6_lstm-Problem2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "8tQJd2YSCfWR"
   },
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "D7tqLMoKF6uq"
   },
   "source": [
    "Deep Learning\n",
    "=============\n",
    "\n",
    "Assignment 6\n",
    "------------\n",
    "\n",
    "After training a skip-gram model in `5_word2vec.ipynb`, the goal of this notebook is to train a LSTM character model over [Text8](http://mattmahoney.net/dc/textdata) data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "collapsed": true,
    "id": "MvEblsgEXxrd"
   },
   "outputs": [],
   "source": [
    "# These are all the modules we'll be using later. Make sure you can import them\n",
    "# before proceeding further.\n",
    "from __future__ import print_function\n",
    "import os\n",
    "import numpy as np\n",
    "import random\n",
    "import string\n",
    "import tensorflow as tf\n",
    "import zipfile\n",
    "from six.moves import range\n",
    "from six.moves.urllib.request import urlretrieve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "output_extras": [
      {
       "item_id": 1
      }
     ]
    },
    "colab_type": "code",
    "collapsed": false,
    "executionInfo": {
     "elapsed": 5993,
     "status": "ok",
     "timestamp": 1445965582896,
     "user": {
      "color": "#1FA15D",
      "displayName": "Vincent Vanhoucke",
      "isAnonymous": false,
      "isMe": true,
      "permissionId": "05076109866853157986",
      "photoUrl": "//lh6.googleusercontent.com/-cCJa7dTDcgQ/AAAAAAAAAAI/AAAAAAAACgw/r2EZ_8oYer4/s50-c-k-no/photo.jpg",
      "sessionId": "6f6f07b359200c46",
      "userId": "102167687554210253930"
     },
     "user_tz": 420
    },
    "id": "RJ-o3UBUFtCw",
    "outputId": "d530534e-0791-4a94-ca6d-1c8f1b908a9e"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found and verified text8.zip\n"
     ]
    }
   ],
   "source": [
    "url = 'http://mattmahoney.net/dc/'\n",
    "\n",
    "def maybe_download(filename, expected_bytes):\n",
    "  \"\"\"Download a file if not present, and make sure it's the right size.\"\"\"\n",
    "  if not os.path.exists(filename):\n",
    "    filename, _ = urlretrieve(url + filename, filename)\n",
    "  statinfo = os.stat(filename)\n",
    "  if statinfo.st_size == expected_bytes:\n",
    "    print('Found and verified %s' % filename)\n",
    "  else:\n",
    "    print(statinfo.st_size)\n",
    "    raise Exception(\n",
    "      'Failed to verify ' + filename + '. Can you get to it with a browser?')\n",
    "  return filename\n",
    "\n",
    "filename = maybe_download('text8.zip', 31344016)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "output_extras": [
      {
       "item_id": 1
      }
     ]
    },
    "colab_type": "code",
    "collapsed": false,
    "executionInfo": {
     "elapsed": 5982,
     "status": "ok",
     "timestamp": 1445965582916,
     "user": {
      "color": "#1FA15D",
      "displayName": "Vincent Vanhoucke",
      "isAnonymous": false,
      "isMe": true,
      "permissionId": "05076109866853157986",
      "photoUrl": "//lh6.googleusercontent.com/-cCJa7dTDcgQ/AAAAAAAAAAI/AAAAAAAACgw/r2EZ_8oYer4/s50-c-k-no/photo.jpg",
      "sessionId": "6f6f07b359200c46",
      "userId": "102167687554210253930"
     },
     "user_tz": 420
    },
    "id": "Mvf09fjugFU_",
    "outputId": "8f75db58-3862-404b-a0c3-799380597390"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data size 100000000\n"
     ]
    }
   ],
   "source": [
    "def read_data(filename):\n",
    "  f = zipfile.ZipFile(filename)\n",
    "  for name in f.namelist():\n",
    "    return tf.compat.as_str(f.read(name))\n",
    "  f.close()\n",
    "  \n",
    "text = read_data(filename)\n",
    "print('Data size %d' % len(text))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "ga2CYACE-ghb"
   },
   "source": [
    "Create a small validation set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "output_extras": [
      {
       "item_id": 1
      }
     ]
    },
    "colab_type": "code",
    "collapsed": false,
    "executionInfo": {
     "elapsed": 6184,
     "status": "ok",
     "timestamp": 1445965583138,
     "user": {
      "color": "#1FA15D",
      "displayName": "Vincent Vanhoucke",
      "isAnonymous": false,
      "isMe": true,
      "permissionId": "05076109866853157986",
      "photoUrl": "//lh6.googleusercontent.com/-cCJa7dTDcgQ/AAAAAAAAAAI/AAAAAAAACgw/r2EZ_8oYer4/s50-c-k-no/photo.jpg",
      "sessionId": "6f6f07b359200c46",
      "userId": "102167687554210253930"
     },
     "user_tz": 420
    },
    "id": "w-oBpfFG-j43",
    "outputId": "bdb96002-d021-4379-f6de-a977924f0d02"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "99999000 ons anarchists advocate social relations based upon voluntary as\n",
      "1000  anarchism originated as a term of abuse first used against earl\n"
     ]
    }
   ],
   "source": [
    "valid_size = 1000\n",
    "valid_text = text[:valid_size]\n",
    "train_text = text[valid_size:]\n",
    "train_size = len(train_text)\n",
    "print(train_size, train_text[:64])\n",
    "print(valid_size, valid_text[:64])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "4eErTCTybtph"
   },
   "source": [
    "---\n",
    "Problem 2\n",
    "---------\n",
    "\n",
    "We want to train a LSTM over bigrams, that is pairs of consecutive characters like 'ab' instead of single characters like 'a'. Since the number of possible bigrams is large, feeding them directly to the LSTM using 1-hot encodings will lead to a very sparse representation that is very wasteful computationally.\n",
    "\n",
    "a- Introduce an embedding lookup on the inputs, and feed the embeddings to the LSTM cell instead of the inputs themselves.\n",
    "\n",
    "b- Write a bigram-based LSTM, modeled on the character LSTM above.\n",
    "\n",
    "c- Introduce Dropout. For best practices on how to use Dropout in LSTMs, refer to this [article](http://arxiv.org/abs/1409.2329).\n",
    "\n",
    "---"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "WARNING: this might not be the best answer since we don't need to train embedding separately. Just put embedding layer before LSTM, then the embedding matrix will be updated through LSTM's training process."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### a - Embedding"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'abcdefghijklmnopqrstuvwxyz'"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "string.ascii_lowercase"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0\n",
      "489\n",
      "728\n",
      "dt\n",
      "729\n"
     ]
    }
   ],
   "source": [
    "import string\n",
    "\n",
    "def build_bigram_dictionary():\n",
    "  # 26 letters + space\n",
    "  all_letters = string.ascii_lowercase + \" \"\n",
    "  bigram_list = []\n",
    "  dictionary = dict()\n",
    "  dic_index = 0\n",
    "  for i in all_letters:\n",
    "    for j in all_letters:\n",
    "      dictionary[i+j] = len(dictionary)\n",
    "  reverse_dictionary = dict(zip(dictionary.values(), dictionary.keys())) \n",
    "  return dictionary, reverse_dictionary \n",
    "\n",
    "dictionary, reverse_dictionary = build_bigram_dictionary()\n",
    "print(dictionary['aa'])\n",
    "print(dictionary['sd'])\n",
    "print(dictionary['  '])\n",
    "print(reverse_dictionary[100])\n",
    "print(len(dictionary)) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[' a', 'na', 'rc', 'hi', 'sm']"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "[a+b for a,b in zip(text[:10],text[1:11])][::2]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[702, 351, 461, 197, 498, 716, 467, 170, 351, 517]\n",
      "[' a', 'na', 'rc', 'hi', 'sm', ' o', 'ri', 'gi', 'na', 'te']\n"
     ]
    }
   ],
   "source": [
    "def build_dataset(text):\n",
    "  bigram_data = [a+b for a,b in zip(text[:len(text)],text[1:len(text)+1])][::2]\n",
    "  data = [dictionary[i] for i in bigram_data]\n",
    "  return data\n",
    "data = build_dataset(text)\n",
    "print(data[:10])\n",
    "print([reverse_dictionary[i] for i in data[:10]])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Build embedding using skip-gram (similar to Assignment 5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "data: [' a', 'na', 'rc', 'hi', 'sm', ' o', 'ri', 'gi']\n",
      "\n",
      "with num_skips = 2 and skip_window = 1:\n",
      "    batch: ['na', 'na', 'rc', 'rc', 'hi', 'hi', 'sm', 'sm']\n",
      "    labels: ['rc', ' a', 'hi', 'na', 'sm', 'rc', ' o', 'hi']\n",
      "\n",
      "with num_skips = 4 and skip_window = 2:\n",
      "    batch: ['rc', 'rc', 'rc', 'rc', 'hi', 'hi', 'hi', 'hi']\n",
      "    labels: ['hi', 'na', 'sm', ' a', 'na', ' o', 'sm', 'rc']\n"
     ]
    }
   ],
   "source": [
    "import collections\n",
    "\n",
    "data_index = 0\n",
    "\n",
    "def generate_batch(batch_size, num_skips, skip_window):\n",
    "  global data_index\n",
    "  assert batch_size % num_skips == 0\n",
    "  assert num_skips <= 2 * skip_window\n",
    "  batch = np.ndarray(shape=(batch_size), dtype=np.int32)\n",
    "  labels = np.ndarray(shape=(batch_size, 1), dtype=np.int32)\n",
    "  span = 2 * skip_window + 1 # [ skip_window target skip_window ]\n",
    "  buffer = collections.deque(maxlen=span)\n",
    "  for _ in range(span):\n",
    "    buffer.append(data[data_index])\n",
    "    data_index = (data_index + 1) % len(data)\n",
    "  for i in range(batch_size // num_skips):\n",
    "    target = skip_window  # target label at the center of the buffer\n",
    "    targets_to_avoid = [ skip_window ]\n",
    "    for j in range(num_skips):\n",
    "      while target in targets_to_avoid:\n",
    "        target = random.randint(0, span - 1)\n",
    "      targets_to_avoid.append(target)\n",
    "      batch[i * num_skips + j] = buffer[skip_window]\n",
    "      labels[i * num_skips + j, 0] = buffer[target]\n",
    "    buffer.append(data[data_index])\n",
    "    data_index = (data_index + 1) % len(data)\n",
    "  return batch, labels\n",
    "\n",
    "print('data:', [reverse_dictionary[di] for di in data[:8]])\n",
    "\n",
    "for num_skips, skip_window in [(2, 1), (4, 2)]:\n",
    "    data_index = 0\n",
    "    batch, labels = generate_batch(batch_size=8, num_skips=num_skips, skip_window=skip_window)\n",
    "    print('\\nwith num_skips = %d and skip_window = %d:' % (num_skips, skip_window))\n",
    "    print('    batch:', [reverse_dictionary[bi] for bi in batch])\n",
    "    print('    labels:', [reverse_dictionary[li] for li in labels.reshape(8)])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Train skip-gram"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import math\n",
    "\n",
    "batch_size = 128\n",
    "embedding_size = 128 # Dimension of the embedding vector.\n",
    "skip_window = 1 # How many words to consider left and right.\n",
    "num_skips = 2 # How many times to reuse an input to generate a label.\n",
    "# We pick a random validation set to sample nearest neighbors. here we limit the\n",
    "# validation samples to the words that have a low numeric ID, which by\n",
    "# construction are also the most frequent. \n",
    "valid_size = 16 # Random set of words to evaluate similarity on.\n",
    "valid_window = 100 # Only pick dev samples in the head of the distribution.\n",
    "valid_examples = np.array(random.sample(range(valid_window), valid_size))\n",
    "num_sampled = 64 # Number of negative examples to sample.\n",
    "\n",
    "vocabulary_size =  len(dictionary)\n",
    "\n",
    "graph = tf.Graph()\n",
    "\n",
    "with graph.as_default(), tf.device('/cpu:0'):\n",
    "\n",
    "  # Input data.\n",
    "  train_dataset = tf.placeholder(tf.int32, shape=[batch_size])\n",
    "  train_labels = tf.placeholder(tf.int32, shape=[batch_size, 1])\n",
    "  valid_dataset = tf.constant(valid_examples, dtype=tf.int32)\n",
    "  \n",
    "  # Variables.\n",
    "  embeddings = tf.Variable(\n",
    "    tf.random_uniform([vocabulary_size, embedding_size], -1.0, 1.0))\n",
    "  softmax_weights = tf.Variable(\n",
    "    tf.truncated_normal([vocabulary_size, embedding_size],\n",
    "                         stddev=1.0 / math.sqrt(embedding_size)))\n",
    "  softmax_biases = tf.Variable(tf.zeros([vocabulary_size]))\n",
    "  \n",
    "  # Model.\n",
    "  # Look up embeddings for inputs.\n",
    "  embed = tf.nn.embedding_lookup(embeddings, train_dataset)\n",
    "  # Compute the softmax loss, using a sample of the negative labels each time.\n",
    "  loss = tf.reduce_mean(\n",
    "    tf.nn.sampled_softmax_loss(softmax_weights, softmax_biases, embed,\n",
    "                               train_labels, num_sampled, vocabulary_size))\n",
    "\n",
    "  # Optimizer.\n",
    "  # Note: The optimizer will optimize the softmax_weights AND the embeddings.\n",
    "  # This is because the embeddings are defined as a variable quantity and the\n",
    "  # optimizer's `minimize` method will by default modify all variable quantities \n",
    "  # that contribute to the tensor it is passed.\n",
    "  # See docs on `tf.train.Optimizer.minimize()` for more details.\n",
    "  optimizer = tf.train.AdagradOptimizer(1.0).minimize(loss)\n",
    "  \n",
    "  # Compute the similarity between minibatch examples and all embeddings.\n",
    "  # We use the cosine distance:\n",
    "  norm = tf.sqrt(tf.reduce_sum(tf.square(embeddings), 1, keep_dims=True))\n",
    "  normalized_embeddings = embeddings / norm\n",
    "  valid_embeddings = tf.nn.embedding_lookup(\n",
    "    normalized_embeddings, valid_dataset)\n",
    "  similarity = tf.matmul(valid_embeddings, tf.transpose(normalized_embeddings))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Average loss at step 0: 3.851841\n",
      "Nearest to ca: bh, c , ch, wd, dc, sv, rp, ap,\n",
      "Nearest to cx: dp, ry, vr, qa, sc, ko, rl, uz,\n",
      "Nearest to de: sy, me, pb, lf, xt, qa, ld, mj,\n",
      "Nearest to dg: uy, ez, cl, av, zs, jf, yz, bk,\n",
      "Nearest to cr: hj, wf, im, fj, cw, ah, aa, by,\n",
      "Nearest to ci: kp, ik, my, fh, ix, lu, ql, zv,\n",
      "Nearest to bt: hp, lb, l , fb, mr, up, uh,  c,\n",
      "Nearest to aj: ck, kd, oe, ly, vv, dw, hl, ba,\n",
      "Nearest to do: nz, ca,  d, ft, xm, ls, mg,  h,\n",
      "Nearest to b : lh, nx, pd, db, ex, is, ff, yy,\n",
      "Nearest to av: wq, th, dg, ck, nr, nm, bf, bp,\n",
      "Nearest to ah: jf, uy, cr, uc, ri, xg,  y, xq,\n",
      "Nearest to bi: hm, ih, gp, gn, my, rg, ha, up,\n",
      "Nearest to bf: jm, dk, nd, ux, sn, dn, zj, xz,\n",
      "Nearest to dl: qa, kl, nv, tk, fr, si, x , ff,\n",
      "Nearest to az: wa, gf, dt, sm, sc, rs,  u, ei,\n",
      "Average loss at step 2000: 1.880756\n",
      "Average loss at step 4000: 1.663021\n",
      "Average loss at step 6000: 1.638268\n",
      "Average loss at step 8000: 1.615224\n",
      "Average loss at step 10000: 1.664438\n",
      "Nearest to ca: c , wd, ba, gj, xa, co, rp, tu,\n",
      "Nearest to cx: dp, vr, qa, ap, gw, uz, dj, rl,\n",
      "Nearest to de: mj, tx, a , hw, xt, sy, xo, re,\n",
      "Nearest to dg: av, uy, ar, ct, zs, yz, ty, ez,\n",
      "Nearest to cr: hj, sk, fj, cw, cl, pl, tr, sq,\n",
      "Nearest to ci: rh, ti, ql, kp, ix, qu, lu, xq,\n",
      "Nearest to bt: hp, dx, lb, fb, sj, ai, mr, ag,\n",
      "Nearest to aj: kd, ck, oe, dz, ou, hl, fw, xy,\n",
      "Nearest to do: to, nz, ft, mg, lx, xm, qp, fo,\n",
      "Nearest to b : lh, db, nx, st, tt, uy, r , pd,\n",
      "Nearest to av: dg, wq, nr, ad, zl, iv, rk, ov,\n",
      "Nearest to ah: jf, ib, uc, xq, uy,  y, xn, yd,\n",
      "Nearest to bi: fn, gp, gn, lo, rg, ih, na, fw,\n",
      "Nearest to bf: ux, nc, jm, dk, it, bl, sh, sn,\n",
      "Nearest to dl: qa, kl, bb, sf, va, gd, ac, tk,\n",
      "Nearest to az: wa, gf, dt, hm, sm, nl, yt, ar,\n",
      "Average loss at step 12000: 1.621317\n",
      "Average loss at step 14000: 1.525933\n",
      "Average loss at step 16000: 1.652667\n",
      "Average loss at step 18000: 1.607346\n",
      "Average loss at step 20000: 1.612534\n",
      "Nearest to ca: co, cu, na, ua, mo, zy, gq, xa,\n",
      "Nearest to cx: dp, vr, gw, qa, kr, dj, ko, uz,\n",
      "Nearest to de: mn, xo, tx, di, ji, ww, kt, da,\n",
      "Nearest to dg: uy, av, gu, yz, ez, ck, zs, jf,\n",
      "Nearest to cr: tr, hj, cl,  l, pk, dr, qu, fj,\n",
      "Nearest to ci: tu, rh, yj, ce, ql, si, ti, sz,\n",
      "Nearest to bt: hp, fb, lb, dx, ny, xf, ag, bh,\n",
      "Nearest to aj: kd, ts, dz, oe, hl,  b, eb, vt,\n",
      "Nearest to do: to, ga, cd, mg, fl, gf, di, lx,\n",
      "Nearest to b : db, lh, nx, vo, a , pd, ff, uy,\n",
      "Nearest to av: iv, wq, ov, ad, am, dg, zl, ev,\n",
      "Nearest to ah: jf, ib, xq, xn, uc, ge, yo, uy,\n",
      "Nearest to bi: ni, fn, ze, gp, po, zv, pc, bj,\n",
      "Nearest to bf: nc, ux, sn, jm, dk, sh, nd, yp,\n",
      "Nearest to dl: qa, kl, sf, sl, bb, fr, fl, lk,\n",
      "Nearest to az: wa, gf, dt, yt, hm, e , sm, nl,\n",
      "Average loss at step 22000: 1.526133\n",
      "Average loss at step 24000: 1.601442\n",
      "Average loss at step 26000: 1.548605\n",
      "Average loss at step 28000: 1.555896\n",
      "Average loss at step 30000: 1.578255\n",
      "Nearest to ca: cu, ba, c , ua, ga, co, ve, ka,\n",
      "Nearest to cx: vr, dp, gw, qa, dj, kr, ou, tj,\n",
      "Nearest to de: di, tx, q , dt, da, sy, kt, mj,\n",
      "Nearest to dg: gu, ck, uy, ar, yz, yl, jy, ib,\n",
      "Nearest to cr: tr, cl, gr, pr, sk, hj, sq, pk,\n",
      "Nearest to ci: tu, yj, vi, xq, wq, vb, ui, qu,\n",
      "Nearest to bt: hp, ny, ce, fb, xf, dx, lb, bh,\n",
      "Nearest to aj: fl, kd, dz, hl, oe, gs, ts, az,\n",
      "Nearest to do: mg, ga, da, ja, gf, ne, to, ih,\n",
      "Nearest to b : db, nx, x , bi, lh, vo, yi, kl,\n",
      "Nearest to av: iv, ad, im, ov, zl, ev, wq, am,\n",
      "Nearest to ah: jf, xq, xn,  y, ge, yo, uy, xg,\n",
      "Nearest to bi: fn, bo, xf, gp, mi, po, bj, b ,\n",
      "Nearest to bf: nc, sh, jm, ux, dk, sn, yp, lt,\n",
      "Nearest to dl: sf, kl, lk, qa, bb, sl, di, cl,\n",
      "Nearest to az: dt, gf, wa, yt, ab, gu, aj, rj,\n",
      "Average loss at step 32000: 1.555490\n",
      "Average loss at step 34000: 1.537218\n",
      "Average loss at step 36000: 1.564124\n",
      "Average loss at step 38000: 1.620148\n",
      "Average loss at step 40000: 1.585204\n",
      "Nearest to ca: na, co, cu, ga, va, ta, c , bc,\n",
      "Nearest to cx: dp, vr, gw, dj, kr, qa, pu, tj,\n",
      "Nearest to de: di, tx, dt, dy, ye, sy, ke, qe,\n",
      "Nearest to dg: gu, ck, uy, jy, ib, dw, yz, av,\n",
      "Nearest to cr: tr, pr, cl, gr, sq, pk, sk, dr,\n",
      "Nearest to ci: tu, qu, ql, xq, yj, wx, tr, ce,\n",
      "Nearest to bt: hp, xf, dx, bm, ny, fb, yf, bh,\n",
      "Nearest to aj: fl, hl, az, kd, dz, gs, ei, oe,\n",
      "Nearest to do: mg, cd, ja, da, ga, gf, qp, vu,\n",
      "Nearest to b : go, cv, db, tw, bi, pc, fr, oh,\n",
      "Nearest to av: iv, ad, am, ov, ev, im, zl, ik,\n",
      "Nearest to ah: jf, ge, xq,  y, qp, ee, nx, xn,\n",
      "Nearest to bi: mi, fn, ku, xf, bo, gp, b , wp,\n",
      "Nearest to bf: jm, sh, ux, yp, dk, nc, bh, sn,\n",
      "Nearest to dl: sf, kl, lk, sl, qa, fl, dw, bb,\n",
      "Nearest to az: dt, gf, sm, aj, ab, wa, gu, yt,\n",
      "Average loss at step 42000: 1.522748\n",
      "Average loss at step 44000: 1.615377\n",
      "Average loss at step 46000: 1.498544\n",
      "Average loss at step 48000: 1.563353\n",
      "Average loss at step 50000: 1.568516\n",
      "Nearest to ca: cu, ga, co, c , wx, gq, va, ve,\n",
      "Nearest to cx: gw, dp, vr, dj, qa, kr, tj, an,\n",
      "Nearest to de: tx, da, dt, di, mm, ty, wp, jb,\n",
      "Nearest to dg: ck, di, gu, av, jy, ib, dw, uy,\n",
      "Nearest to cr: tr, gr, pr, sq, fl,  l, kn, pl,\n",
      "Nearest to ci: tu, yj, ti, di, si, wx, vi, qu,\n",
      "Nearest to bt: hp, xf, bm, dx, yf, bh, fb, ny,\n",
      "Nearest to aj: fl, hl, az, ei, dz, kd, vt, at,\n",
      "Nearest to do: da, to, mg, ja, ki, cd, du, dv,\n",
      "Nearest to b : bi, v , tw, cv, fr, go, fn, db,\n",
      "Nearest to av: iv, ad, ov, am, im, op, zl, ev,\n",
      "Nearest to ah: jf, qp, ge, nx, xq, yo, uy, ps,\n",
      "Nearest to bi: fn, mi, xf, b , ba, ku, mu, po,\n",
      "Nearest to bf: jm, yp, nc, sh, ux, dk, bh, lp,\n",
      "Nearest to dl: sf, kl, lk, vq, sl,  w, qa, sk,\n",
      "Nearest to az: ab, dt, aj, gf, it, uf, yt, xq,\n",
      "Average loss at step 52000: 1.561924\n",
      "Average loss at step 54000: 1.587399\n",
      "Average loss at step 56000: 1.553898\n",
      "Average loss at step 58000: 1.486202\n",
      "Average loss at step 60000: 1.552178\n",
      "Nearest to ca: ga, cu, wx, vp, gq, ba, co, zc,\n",
      "Nearest to cx: gw, dp, dj, vr, kr, ip, tj, pu,\n",
      "Nearest to de: tx, dt, di, he, ke, te, du, fz,\n",
      "Nearest to dg: gu, ck, jy, yl, qu, ng, av, tl,\n",
      "Nearest to cr: tr, gr, pr, pk, cl, fl, qu, kn,\n",
      "Nearest to ci: tu, yj, vi, ti, ce, qu, wx, di,\n",
      "Nearest to bt: hp, bm, xf, bh, ce, dx, yf, mi,\n",
      "Nearest to aj: ei, az, fl, hl, at, dz, ay, kd,\n",
      "Nearest to do: da, to, je, gi, gf, ki, cd, mg,\n",
      "Nearest to b : go, cv, x , bi, fn, pd, ou, fr,\n",
      "Nearest to av: ad, iv, am, ov, ev, ik, zl, wq,\n",
      "Nearest to ah: jf, qp, ge, aw, xq,  y, nx, uy,\n",
      "Nearest to bi: mi, bu, fn, mu, b , ba, ni, xf,\n",
      "Nearest to bf: jm, ux, yp, dk, sn, bh, lp, sh,\n",
      "Nearest to dl: sf, kl, lk, sl, rj, vq, qa, di,\n",
      "Nearest to az: dt, aj, gf, ab, sk, am, ai, xq,\n",
      "Average loss at step 62000: 1.498673\n",
      "Average loss at step 64000: 1.513360\n",
      "Average loss at step 66000: 1.558935\n",
      "Average loss at step 68000: 1.609786\n",
      "Average loss at step 70000: 1.588301\n",
      "Nearest to ca: ga, cu, co, ba, na, ve, mo, fu,\n",
      "Nearest to cx: gw, dp, vr, dj, kr, tj, op, pu,\n",
      "Nearest to de: di, tx, dt, du, sy, tk, fj, mm,\n",
      "Nearest to dg: yl, bb, jy, ck, gu, nr, yz, di,\n",
      "Nearest to cr: tr, gr, kn, pr, cl, sq, fl, dr,\n",
      "Nearest to ci: vi, xq, ti, tu, di, yj,   , ce,\n",
      "Nearest to bt: hp, bm, xf, bh, yf, dx, zr, fb,\n",
      "Nearest to aj: fl, az, ei, hl, dz, kd, hb, ay,\n",
      "Nearest to do: gi, cd, da, mg, to, je, gf, qq,\n",
      "Nearest to b : m , go, cv, x , gd, ou, bi, fd,\n",
      "Nearest to av: ad, iv, am, ev, ov, zl, az, wq,\n",
      "Nearest to ah: jf, qp, aw,  y, nx, xq, yo, ee,\n",
      "Nearest to bi: mu, fn, mi, mq, ka, xf, ji, ni,\n",
      "Nearest to bf: sh, jm, ux, yp, dk, nc, lp, sn,\n",
      "Nearest to dl: sf, kl, sl, lk, rj, vq, gd, tr,\n",
      "Nearest to az: dt, aj, av, gf, ai, am, it, ab,\n",
      "Average loss at step 72000: 1.527642\n",
      "Average loss at step 74000: 1.553728\n",
      "Average loss at step 76000: 1.576837\n",
      "Average loss at step 78000: 1.571109\n",
      "Average loss at step 80000: 1.580166\n",
      "Nearest to ca: co, cu, na, ga, gi, gj, dj, sa,\n",
      "Nearest to cx: gw, dp, vr, dj, kr, tj, rl, op,\n",
      "Nearest to de: tx, mj, dt, i , da, ke, ds, rp,\n",
      "Nearest to dg: ck, ng, gu, yl, di, nc, tt, yz,\n",
      "Nearest to cr: tr, gr, kn, pr, sq, fl, cl, pl,\n",
      "Nearest to ci: ti, tu, qu, vi, xq, zi, ce, di,\n",
      "Nearest to bt: hp, xf, bm, yf, bh, ag, dx, zr,\n",
      "Nearest to aj: fl, ei, ts, dz, cc, az, kd, ay,\n",
      "Nearest to do: da, gi, mg, no, cd, to, ga, du,\n",
      "Nearest to b : v , x , m , cv, a , rh, fd, pd,\n",
      "Nearest to av: iv, am, ad, zl, im, ov, ev, ig,\n",
      "Nearest to ah: jf, yo, nx, qp, ge, aw, gw,  y,\n",
      "Nearest to bi: mu, mi, fn, xf, ba, nb, ka, gp,\n",
      "Nearest to bf: jm, ux, nc, sh, dk, lp, bh, yp,\n",
      "Nearest to dl: sf, kl, lk, sl, rj, gd, tl, sk,\n",
      "Nearest to az: dt, aj, gf, sw, sk, xq, pg, gu,\n",
      "Average loss at step 82000: 1.533572\n",
      "Average loss at step 84000: 1.557151\n",
      "Average loss at step 86000: 1.584758\n",
      "Average loss at step 88000: 1.566685\n",
      "Average loss at step 90000: 1.558346\n",
      "Nearest to ca: cu, ga, na, co, ve, wx, ba, gq,\n",
      "Nearest to cx: gw, dj, kr, vr, dp, an, cr, tj,\n",
      "Nearest to de: di, tx, ds, da, mj, fj, xn, rp,\n",
      "Nearest to dg: gu, nc, bb, jy, yz, ck, tl,  b,\n",
      "Nearest to cr: pr, tr, gr, kn, cl, sq, fr, fl,\n",
      "Nearest to ci: di, vi, qu, yj, xq, ti, vb, tu,\n",
      "Nearest to bt: bm, hp, xf, yf, ag, dx, zr, bh,\n",
      "Nearest to aj: ay, cc, ei, dz, fl, hl, az, ts,\n",
      "Nearest to do: cd, ki, to, mg, da, gi, je, du,\n",
      "Nearest to b : m , c , a , fd, g , v , z , mh,\n",
      "Nearest to av: iv, am, ad, zl, ev, yx, az, wq,\n",
      "Nearest to ah: qp, yo, ji, ow, xv, nx, jf, mt,\n",
      "Nearest to bi: mi, ba, mu, nb, iy, ka, mq, tf,\n",
      "Nearest to bf: jm, ux, yp, sh, bh, sn, dk, lp,\n",
      "Nearest to dl: sf, kl, lk, tl, rr, vq,  b, gd,\n",
      "Nearest to az: dt, sk, av, sw, aj, rj, gf, am,\n",
      "Average loss at step 92000: 1.602411\n",
      "Average loss at step 94000: 1.500287\n",
      "Average loss at step 96000: 1.503936\n",
      "Average loss at step 98000: 1.608041\n",
      "Average loss at step 100000: 1.505967\n",
      "Nearest to ca: ga, co, cu, ta, na, vp, ve, jy,\n",
      "Nearest to cx: gw, vr, dj, kr, dp, oz, tj, dr,\n",
      "Nearest to de: tx, ds, fj, di, ke, do, mj, dt,\n",
      "Nearest to dg: tt, ck, tl, jy, yz, bb, bp, gu,\n",
      "Nearest to cr: gr, cl, sq, pr, tr, kn, fr, fl,\n",
      "Nearest to ci: vi, di, tu, vh, qu,   , yj, gm,\n",
      "Nearest to bt: bm, xf, hp, ag, dx, yf, bh, zr,\n",
      "Nearest to aj: ay, fl, az, dz, hl, cc, ei, ts,\n",
      "Nearest to do: mg, cd, to, da, je, fr, gi, ja,\n",
      "Nearest to b : v , m , c , ku, x , cv, w , z ,\n",
      "Nearest to av: ad, iv, zl, ov, ev, az, am, nr,\n",
      "Nearest to ah: ji, ee, ge, yo, qp, aw, jf, cv,\n",
      "Nearest to bi: mi, ba, mu, wa, bo, ni, iy, fo,\n",
      "Nearest to bf: jm, ux, lp, bh, yp, nc, dk, lt,\n",
      "Nearest to dl: sf, kl, tl, sl, nl, rr, nr, lk,\n",
      "Nearest to az: dt, av, sw, aj, sk, rj, pg, fb,\n"
     ]
    }
   ],
   "source": [
    "num_steps = 100001\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.initialize_all_variables().run()\n",
    "  print('Initialized')\n",
    "  average_loss = 0\n",
    "  for step in range(num_steps):\n",
    "    batch_data, batch_labels = generate_batch(\n",
    "      batch_size, num_skips, skip_window)\n",
    "    feed_dict = {train_dataset : batch_data, train_labels : batch_labels}\n",
    "    _, l = session.run([optimizer, loss], feed_dict=feed_dict)\n",
    "    average_loss += l\n",
    "    if step % 2000 == 0:\n",
    "      if step > 0:\n",
    "        average_loss = average_loss / 2000\n",
    "      # The average loss is an estimate of the loss over the last 2000 batches.\n",
    "      print('Average loss at step %d: %f' % (step, average_loss))\n",
    "      average_loss = 0\n",
    "    # note that this is expensive (~20% slowdown if computed every 500 steps)\n",
    "    if step % 10000 == 0:\n",
    "      sim = similarity.eval()\n",
    "      for i in range(valid_size):\n",
    "        valid_word = reverse_dictionary[valid_examples[i]]\n",
    "        top_k = 8 # number of nearest neighbors\n",
    "        nearest = (-sim[i, :]).argsort()[1:top_k+1]\n",
    "        log = 'Nearest to %s:' % valid_word\n",
    "        for k in range(top_k):\n",
    "          close_word = reverse_dictionary[nearest[k]]\n",
    "          log = '%s %s,' % (log, close_word)\n",
    "        print(log)\n",
    "  final_embeddings = normalized_embeddings.eval()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(729, 128)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "final_embeddings.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA3IAAANmCAYAAAC/kLlsAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzs3XtclHXaP/DPDQqSgagh2apo6z5ycDjpcDBBUCE8Cz5k\nrJbyIOYhk6dWO2yr9lJ3t81YxcPWpvGgIVEeKN02CxESj6ODaY/iWsn4+MvDKEhqoAjf3x84s6Cg\nHGbmnnvm8369eL3GEea+mLlH57q/1/e6JCEEiIiIiIiISDkc5A6AiIiIiIiIWoeJHBERERERkcIw\nkSMiIiIiIlIYJnJEREREREQKw0SOiIiIiIhIYZjIERERERERKYzJEjlJkhwkSdJKkvT53T93lSTp\nK0mSTkuStEuSpC6mOhYREREREZE9M+WK3HwAJxv8+TUA+UKIAQAKALxuwmMRERERERHZLZMkcpIk\n9QIwGsD6BndPAJB193YWgImmOBYREREREZG9M9WK3F8BLAAgGtznKYS4BABCiIsAepjoWERERERE\nRHat3YmcJEljAFwSQhwDID3gW8UD/o6IiIiIiIhaqIMJHuMpAOMlSRoNwAWAqyRJmwBclCTJUwhx\nSZKkxwFcbuqHJUligkdERERERHZNCPGgRbH7tHtFTgjxhhCijxDiSQDPAigQQjwHYAeA6Xe/bRqA\nzx7wGPyy4q/FixfLHgO/+Bop+Yuvj/V/8TWy7i++Ptb/xdfI+r/4Gln3V1uYc47cnwHESJJ0GsCI\nu38mIiIiIiKidjJFaaWREKIIQNHd2+UARpry8YmIiIiIiMi8K3JkI6KiouQOgR6Cr5F14+tj/fga\nWTe+PtaPr5H142tke6S21mSaLABJEnLHQEREREREJBdJkiAs3eyEiIiIiIiILIuJHBERERERkcIw\nkSMiIiIiIlIYJnJEREREREQKw0SOiIiIiIhIYZjIERERERERKQwTOSIiIiIiIoVhIkdERERERKQw\nTOSIiIiIiIgUhokcERERERGRwjCRIyIiIiIiUhgmckRERERERArDRI6IiIiIiEhhmMgREREREREp\nDBM5IiIiIiIihWEiR0REREREpDBM5IiIiIiIiBSGiRwREREREZHCMJEjIiIiIiJSGCZyRERERERE\nCsNEjoiIiIiISGGYyBERERERESkMEzkiIiIiIiKFYSJHRERERESkMEzkiIiIiIiIFIaJHBERERER\nkcIwkSMiIiIiIlIYJnJEREREREQKw0SOiIiIiIhIYZjIERERERERKQwTOSIiIiIiIoVhIkdERERE\nRKQwTOSIiIiIiIgUhokcERERERGRwjCRIyIiIiIiUhgmckRERERERArDRI6IiIiIiEhhmMgRERER\nEREpDBM5IiIiIiIihWEiR0REREREpDBM5IiIiIiIiBSGiRwREREREZHCMJEjIiIiIiJSGCZyRERE\nRERECsNEjoiIiIiISGGYyBERERERESkMEzkiIiIiIiKFYSJHRERERESkMEzkiIiIiIiIFIaJHBER\nERERkcIwkSMiIiIiIlIYJnJEREREREQKw0SOiIiIiIhIYZjIERERERERKQwTOSIiIiIiIoVhIkdE\nRERERKQwTOSIiIiIiIgUhokcERERERGRwjCRIyIiIiIiUhgmckRERERERArDRI6IiIiIiEhhmMgR\nEREREREpDBM5IiIiIiIihWEiR0REREREpDBM5IiIiIiIiBSGiRwREREREZHCMJEjIiIiIiJSGCZy\nRERERERECsNEjoiIiIiISGGYyBERERERESkMEzkiIiIiIiKFYSJHRERERESkMEzkiIiIiIiIFIaJ\nHBERERERkcIwkSMiItTV1ckdAhEREbUCEzkiIjuQnZ2N0NBQBAcHY/bs2airq4Orqyt+97vfISgo\nCAcPHpQ7RCIiImoFJnJERDautLQUubm52L9/P7RaLRwcHJCdnY1ffvkF4eHhKCkpwZAhQ+QOk4iI\niFqhg9wBEBGRee3evRtarRZqtRpCCFRXV8PT0xOOjo5ISEiQOzwiIiJqA0kIIW8AkiTkjoGIyJat\nWbMGFy5cwPLlyxvd7+bmhp9//lmmqIiIiMhAkiQIIaTW/AxLK4mIbNyIESOwZcsW6PV6AEBFRQXO\nnTsHXkQjIiJSLpZWEhHZOB8fHyxbtgyxsbGoq6uDk5MT1qxZA0lq1YU/IiIisiIsrSQiIiIiIpIR\nSyuJiOih9Ho9NBqNsdSSiIiIlIeJHBGRHcnJyYWXlzdiYmbBy8sbOTm5codEREREbcDSSiIiO6HX\n6+Hl5Y2qqj0A/AEch4tLNHS6Unh4eMgdHhERkd1iaSURETWrrKwMTk59UZ/EAYA/Onb0QllZmXxB\nERERUZswkSMishN9+/bF7dtlAI7fvec4amp06Nu3r3xBERERUZswkSMishMeHh7YsGEdXFyi4eYW\nDBeXaGzYsI5llURERArEPXJERHZGr9ejrKwMffv2ZRJHRERkBdqyR46JHBERERERkYzY7ISIiIiI\niMgOMJEjIiIiIiJSGCZyRERERERECsNEjoiIiIiISGHanchJkuQsSdIhSZJKJEn6X0mS/nj3/q6S\nJH0lSdJpSZJ2SZLUpf3hEhERERERkUm6VkqS9IgQ4hdJkhwB7APwCoDxAK4KIf4iSdKrALoKIV5r\n4mfZtZKIiIiIiOyWbF0rhRC/3L3pfPcxKwBMAJB19/4sABNNcSwiIiIiIiJ7Z5JETpIkB0mSSgBc\nBFAohDgJwFMIcQkAhBAXAfQwxbGIiIiIiIjsXQdTPIgQog5AkCRJbgB2SZIUBeDeeslm6yeXLFli\nvB0VFYWoqChThEVERDKLj4/H+fPnUV1djfnz52PGjBlwdXVFamoqvvrqK/Ts2RMff/wxunfvLneo\ndmHo0KEoLi6WOwwiIrtXWFiIwsLCdj2GSfbINXpASfoDgCoAKQCihBCXJEl6HMAeIYRPE9/PPXJE\nRDbq2rVrcHd3R3V1NdRqNYqKivDYY49h8+bNePbZZ7F06VJcvnwZq1evljtUIiIi2ciyR06SpMcM\nHSklSXIBEAOgBMDnAKbf/bZpAD5r77GIiEhZVq5cicDAQISFheH8+fM4c+YMHB0d8cwzzwAApk6d\nin379skcpf1wdXU13n7xxRfh4+OD2NhYjBkzBtu2bZMxMiIiai1TlFb2BJAlSZKE+sRwkxBi9909\nc59IkvRfAHQAnjHBsYiISCGKiopQUFCAQ4cOwdnZGdHR0aiurr7v++r/+yBLMDzX27Ztw5kzZ3Dq\n1ClcuHABvr6+SElJkTk6IiJqjXavyAkhTgghgoUQQUKIACHEirv3lwshRgohBgghYoUQ19ofLhER\nKUVlZSW6du0KZ2dnlJaW4uDBgwCA2tpabNmyBQCQnZ2NoUOHPvSxsrOzERoaiuDgYMyePRt1dXVI\nTk6Gv78/AgICsGrVqmZ/tiWPb2/27t2LpKQkAEDPnj0xfPhwmSMiIqLWMkmzEyIionvFxcXhvffe\ng5+fHwYMGIAhQ4YAADp37ozDhw9j6dKl8PT0RG5u7gMfp7S0FLm5udi/fz8cHR0xd+5cLFu2DD/9\n9BOOHz8OAPj555+b/Xk29yAiIltk8mYnrQ6AzU6IiOyKq6srrl+/3uLvX7t2Lf70pz+hR48eEEKg\nuroacXFx2LlzJ0aNGoUxY8YgNjYWkiQhPT0dmZmZkCQJM2bMwEsvvdTq49kyw3Oxfft2/P3vf8c/\n/vEPXLp0CX5+fli/fj0SEhLkDpGIyC61pdkJV+SIiMjiNBoN+vbtCw8Pj4d+rxAC06ZNw/Llyxvd\nv3z5cuzatQvvv/8+PvnkE8ydOxdZWVnQaDSora1FWFgYIiMjuQevAcNzER8fj4KCAvj5+aFPnz7G\n1VIiIlIOkwwEJyIiaomcnFzU1johJmYWvLy8kZPz4LJKABgxYgS2bNkCvV4PAKioqMC5c+dQW1uL\n+Ph4LF26FCUlJSguLkZ8fDw6deqEzp07IyEhAXv37jX3r6QoDUtQV69ejVOnTmHXrl0tSqiJiMi6\ncEWOiIgsQq/XIyVlDqqq9qCqyh/AcaSkRGPkyOEPTCR8fHywbNkyxMbGoq6uDk5OTkhPT0d8fDzq\n6uogSRL+/Oc/o7S0tNHPsWy/eXq9HmVlZcZVUa5aEhEpD/fIERGRRWg0GsTEzEJl5VHjfW5uwcjP\nfx9qtbrdj19SUoLk5GQcPHgQtbW1CA8Px6ZNmxAREfHAZij2JicnFykpc+Dk1Be3b5dhw4Z1SEqa\nLHdYFqPT6TBq1CgMHToU+/fvR69evfDhhx9izJgxhj0qOHHiBM6ePYvevXvLHS4R2QlZBoITERG1\nRN++9YkDcPzuPcdRU6ND3759W/U4er0eGo3GWGppEBQUhOnTp0OtViM8PBypqakICAgwReg2o+Gq\naGXlUVRV7UFKypz7nktb9/3332PevHn47rvv0KVLFxQWFqKkpARarRapqalITExkEkdEVo+JHBER\nWYSHhwc2bFgHF5douLkFw8UlGhs2rGvV/qycnFx4eXk3u8cuLS0NJ06cwPHjxzFv3jxcvXoV3bt3\nN/WvolhlZWVwcuoLwP/uPf7o2NELZWVl8gUlg379+kGlUgEABg0aZPz99+3bh/Xr12PDhg0yRkdE\n1DLcI0dERBaTlDQZI0cOb7Q/q6Vas8dOr9fjyJEjmDdvHhYsWGDi30K5Gq+K1j+HbVkVVTpnZ2fj\nbUdHR1RXV+PixYtITU3Fjh078Mgjj8gYHRFRyzCRIyIii/Lw8GhTl0TDalJ9Egc0XE1q+HiN94BV\noGtXrsgZeHh4oHdvD/zf/0WjY0cv1NToWr0qagvu3Zt/+/ZtJCYm4u2338avf/1rmaIiImodNjsh\nIiJF0Ov18PLyRlXVHhhWk1xcoqHTlRoTkZZ8D93ftdKe6HQ6jBs3DseP1+/VfPfdd/HFF1/gwIED\n8Pb2hhACkiThiy++wOOPPy5ztERkLzgQnIiIbJZhj11KSvOrSS1dtbNnrq6uuH79Oh577DHMnTsX\nhYWF6N27Nzp06ICUlBQkJCTIHaJZeXl5GZM4AHj++ecRGRlpl0ktESkbm50QEZFiJCVNhk5Xivz8\n96HTld7XNt9UnTFtmWFm3NatW3Hu3DmcPHkSGzduxIEDB2SOzPIe1jyHiMiaMZEjIrJh6enpUKlU\n8Pf3x6pVq6DT6eDj44OpU6fC19cXzzzzDKqrq+UOs1U8PDygVqubXD0xRWdMe7Fv3z4kJiYCADw9\nPREdHS1zRJbFUQxEpHRM5IiIbJRWq0VWVhY0Gg0OHDiA9evXo6KiAqdPn8aLL76IkydPwtXVFevW\nrZM7VJN62KodEcBRDESkfEzkiIhsVHFxMeLj49GpUyd07twZCQkJ2Lt3L/r06YOwsDAAwNSpU1Fc\nXCxzpKb3oFW71hg7dix+/vlnE0VlHQwNxp566ils3boVQghcunQJhYWF8gZmYQ8rw/3Tn/4kU2RE\nRC3DRI6IyE4YPsAb9kgZ3Ptn+redO3fCzc1N7jBMyvB6T5o0Cb169YKfnx+ef/55DBo0CF26dJE5\nOst5WBnuH//4R5kjJCJ6MCZyREQ2KiIiAnl5eaiursbNmzeRl5eHyMhI6HQ6HDp0CACwefNmDB06\nVOZIrUN2djZCQ0MRHByM2bNno66uDv369UN5ebncoZmUYYXxypUrePbZZ1FUVIScnBycPXsWKpVK\n5ugsy1CG+7vfxcPHpy/eeedtzJo1CwsXLkRVVRWCg4Px3HPPyR0mEVGTmMgREdmooKAgTJ8+HWq1\nGuHh4UhNTYW7uzsGDBiAtWvXwtfXF9euXcPs2bPlDlV2paWlyM3Nxf79+6HVauHg4IDs7GybXa00\ndGscPnwMPD17IiAgAIsWLUKPHj3kDs3irl69Co1Gg8OHD0Or1cLR0REqlQqPPPIItFotNm3aJHeI\nRERN4hw5IiIblpaWhrS0NOOfdTodOnTogI0bN8oYlfXZvXs3tFot1Go1hBCorq6Gp6en3GGZRcNu\njYah6VevRiMuLq5Vj5Oeno7MzExIkoQZM2bgpZdewtKlS5GdnY0ePXqgV69eGDx4MF5++WWz/B6m\n0tRrb48JLREpDxM5IiI7U1tbC41GwwHIDQghMG3aNCxfvrzR/ZmZmTJFZD6mGJresCNqbW0twsLC\nMGTIEGzfvh0nTpzArVu3EBwcjMGDB5vvFzGR5l77FStWyBQREVHLsLSSiMiO7N9/EGVllzgA+R4j\nRozAli1bjDPEKioqcO7cOZssrTTF0PSmOqL+85//xIQJE9CxY0c8+uijGDdunBmiN73mXnsnJyfc\nuXNH5uiIiJrHRI6IyE5wAHLzfHx8sGzZMsTGxiIgIACxsbG4ePEihBA2l8yZY2i6kp+n5l77mTNn\nwt/fn81OiMhqSYZ21LIFIElC7hiIiOyBRqNBTMwsVFYeNd7n5haM/Pz3oVarZYzMOtXW1uLxxx/H\nxYsX4ejoKHc4JqfX61FWVtamEtuSkhIkJyfj4MGDqK2tRXh4ODIzMzFz5kzs378fNTU1GDRoEF54\n4QXZ98hVVlZi8+bNZm/qM3ToUBQXF6OoqAgrVqzAjh07zHo8IrItkiRBCNGqK2JckSMishOmKKmT\nw5YtW+Dr64sRI0ZY5Hh6vR4ajQY+Pj5ITU21ySQOaN/Q9KY6og4aNAgTJkxAQEAAxowZA39/f6uY\nS1dRUYF169a1+PsNr39rV6qLi4uNt5W6OklEysJEjojITpiqpM7Sc+c2bNiA9evXY/fu3WY/lqEt\nf0zMLJw/fxUqVYDZj6lUaWlpOHHiBI4fP4558+YBAJ5//nls2rQJmzZtQllZGQYNGmTxuNLT06FS\nqeDv749Vq1bh9ddfxw8//IDg4GC8+uqrD/zZhq9/a/eQurq6Gm9fv34diYmJ8PHxYWkmEZkNSyuJ\niOxMe0rqzC07OxsZGRmoqalBSEgInnjiCbzzzjvo1asXxo8fj7fffttsx9br9fDy8m7Ult/FJRo6\nXanVPU/WKCcnF8899zyEcIQQt/Dss89i8+Zsi8ag1WqRnJyMQ4cOGbtpfvTRR3juuedw/PjxB/5s\ne19/Nzc3/PzzzygqKsLEiRNx8uRJPP7443jqqaewYsUKDBkyxDS/JBHZJJZWEhHRQ7WnpA7498pD\nUVERhg0bhrFjx8Lb2xtz5sxpV1z3DuV2dHREv379oFarkZ2dbdYkDvh3W/76D/FAw7b89GCGRjq1\ntRrU1f0CIUqQl/elxRvpNNVN85tvvmnRz5ry9Q8JCUHPnj0hSRICAwN5DhGRWXCOHBERtUrD/T8a\njQanTp1Cnz598PTTT2Pbtm1ISEho0+PKPZS78R7C+hUZJewhtAammE1nDq2p+DHl6+/s7Gy87ejo\nyDEGRGQWXJEjIqI2CwkJgZeXFyRJQlJSUqOGD61lGMys1WpRUlKCU6dOYdGiRa36MN4e5mjL31bx\n8fFQq9VQqVRYv349AODLL7/EoEGDEBQUhJiYGIvH9CDW0kgnIiICeXl5qK6uxs2bN5GXl4ehQ4fi\n+vXrD/3Z9r7+3CZCRJbGFTkiImqze7vztadb34gRIzBx4kSkpaXBw8MDFRUVLfoAbkpJSZMxcuRw\nlJWV4ciRI1i5Mh3vvPM2QkNDsW7dOuzatQu///3vUVdXh8ceewxff/21WeLIzMyEu7s7qquroVar\nMX78eMycORPFxcXo06cPrl27ZpbjtpUhCUpJiUbHjl6oqdHJkgQ37KYpSRJSU1MRFBSEIUOGwN/f\nH6NGjXpgiW7D17+1e0ibO/fZwZKIzIXNToiIqFVcXV1x/fp1FBUVYfTo0Th58iR69+6N0aNH44UX\nXkB8fHybH/vTTz/FH//4R9TV1cHJyQlr167Fa6+9hhUrViA4ONiEv8WDlZaWYuHChdi+fTscHR0x\nd+5chIaG4s0332yUTLm7u5vl+EuWLEFeXh4AQKfT4ZVXXsHp06exadMmsxzPVKy5kQ4RkTVrS7MT\nrsgREVGrNFxhGDx4MF588UV8//33GD58eLuSOABITExEYmJio/sKCgra9Zht0dR+vcOHD2PYsGHo\n06cPAJgtiSsqKkJBQQEOHToEZ2dnREdHIygoCKWlpWY5nil5eHhYTQInV1LJZJaILIV75IiIqFV+\n/vln4+0uXbpgx44dOHXqFNauXWuyY7R1KLOpNLVfb/HixRbZB1VZWYmuXbvC2dkZpaWlOHjwIKqq\nqrB3715j98OKigqzx6Fk7ZkHp8TjEpF9YiJHRERtcu3aNVy7ds3kyZY1fBgeMWIEtmzZYvzdKioq\n4O/vj71790Kn0xnvM4e4uDjU1NTAz88Pb7zxBsLDw9GjRw/8/e9/R0JCAoKCgvDss8+a5di2wDAK\noapqDyorj6Kqag9SUuaY/aKAXMclIvvFPXJERNRqOTm5SEmZAyen+m6FGzasQ1LS5HY/rjUN5W5q\nv15FRQVef/11CCHQo0cP7Nq1y6Ix0cNpNBrExMxCZeVR431ubsHIz38farXa5o5LRLahLXvkmMgR\nEVGrmDPZ4ofhplli31V6ejoyMzMhSRJSUlIwf/58bNy4Ee+++y4cHBzg7++PrKwssxzblOS6GGBN\nFyGISHnY7ISIiMzOnMOfrXkot1xNLMy1+tmQVqtFVlYWNBoNamtrERYWBrVajeXLl+PgwYPo2rWr\n1Y08aI5coxCsZQQDEdkPrsgREVGrmHvlwZC4NPwwbOrEpa0xmTOZaoqlVnkyMjJQXl6OJUuWAAAW\nL16M7t27Q6/XY+nSpSY7jiWxayURKUlbVuTY7ISISOGys7MRGhqK4OBgzJ49G+fOncN//Md/oLy8\nHEIIREZGIj8/32THM6w8uLhEw80tGC4u0SZdeUhKmgydrhT5+e9DpyuVPYmTs4mFYfWzPokDGq5+\nmpMQwvChwqzHMScPDw+o1WqLJ1NyHZeI7A8TOSIiBSstLUVubi72798PrVYLBwcHFBUV4bXXXsOs\nWbPw7rvvws/PDyNHjjTpcc2dbFnTh2G5king3lJTwFylphEREcjLy0N1dTVu3ryJvLw8DB48GFu2\nbEF5eTkAjjwgIrI23CNHRKRgTQ2u9vT0xKJFi/DJJ5/g/fffx7Fjx8xybGsa/mxOcu7bs9S+q6Cg\nIEyfPh1qtRqSJCE1NRXh4eH4/e9/j2HDhqFDhw4ICgrChx9+aLJj3ttc5eLFi+jduzfmzJkDAHjr\nrbfg6uqKl19+2WTHJCKyJdwjR0SkYGvWrMGFCxewfPnyRvdXVVVBrVbj1q1bKC4uhqenp0wR2ga5\n9+3Z2r4rrVaL5ORkHDp0yNhc5aOPPsL8+fNRWFgIAPDz88NXX32FX/3qV616bCaARKRE7FpJRGRn\nRowYgYkTJyItLQ0eHh6oqKjA9evXsWLFCkydOhVeXl6YMWMGduzYIXeoipaUNBkjRw6XLZmSY/XT\nnMljcXEx4uPj0alTJwBAQkICvvnmG+j1ely8eBGXL19Gt27dWp3EERHZE+6RIyJSMB8fHyxbtgyx\nsbEICAhAbGwsysrKcOTIEbz66qtISkqCs7OzIuZ/WTtz79vLyMiAr68vnnvuObM8fmvk5OTCy8sb\nMTGz4OXljZycXLMez9BcJTExEZ9++ilyc3MxeXLTK57p6elQqVTw9/dHRkYGAGD58uUYMGAAIiMj\ncfr0abPGSkRkLVhaSUREZAV8fHywe/duPPHEE7LGYYmRByUlJUhOTsbBgwcblVZ27NgRqampuHr1\nKoqKiu4rCW6qJDMrKwvJyck4fPgwbt++bezeytJKIlISllYSEZHN7aeyB7Nnz8aPP/6IUaNGITEx\nET/++COOHDkCBwcHLF68GPHx8RaLxZwD3w3uba4yc+ZMBAQEAACuX7+OXr16Nbmvs6mSzH/84x+I\nj4+Hs7MznJ2dMX78eJPESERk7ZjIERHZELkGV1P7/O1vf8OuXbuwZ88e/OUvf4G7uzuOH68fOVBZ\nWWnRWCzVpTMtLQ1paWn33W/4vVvCUJJJRGSPuEeOiMhGyDm4mkxDCIH8/HzMnTvXeF+XLl0sGoO5\nB743R6/XQ6PRPPB8bWre3ZgxY7B9+3bcunUL169fZ2MfIrIbXJEjIrIRliiJI/OyltUlS3fpbOlK\nclPz7oKCgjB58mT4+/vD09MTISEhZo2ViMhasNkJEZGNsESTCjKffv364ejRo3jnnXdQXV2Nv/71\nrwCAa9euwd3dXebozMcU5y33hRKR0rWl2QlLK4mIzKBhi/RVq1ZBp9PB19cXM2fOxMCBAxEXF4db\nt26Z9JhylcSRaRhW4958801UVFRApVIhKCjIOCDbVhlWkuuTOKDhSnJLWHpUAhGRteCKHBGRiTXV\nIn3Tpk1Qq9XQarVQqVSYPHkyJkyYgN/+9rcmPz5XJ0hJ2rMix1VoIrIVXJEjIrICDVukd+7cGQkJ\nCdi7dy+efPJJqFQqAMCgQYNavOLQWuYeXE3m05KGH7amPSvJ7V3NIyJSMiZyRGRS2dnZCA0NNQ7l\nraurg6urK958800EBgZiyJAhdvUhFajvRAgAzs7OxvscHR1x584duUIiK2TPJYJJSZOh05UiP/99\n6HSlLR6Z0XhUAmCuUQlERNaIiRwRmUxpaSlyc3Oxf/9+aLVaODg4IDs7G7/88guGDBmCY8eOISIi\nAh988IHcoZpVUy3SIyMjwTJy69bcRYiXX34ZAwcORExMDK5evWqWY3N0RNtWkpW+LzQ+Ph5qtRoq\nlQrr168HAGzYsAEDBgxAWFgYZs6ciZdeeknmKInIWjGRIyKT2b17N7RaLdRqNYKCglBQUICzZ8/C\nyckJo0ePBmDekkJr0bBFenh4OFJTU+Hu7m41reXpfg+6CBESEoLvvvsOkZGRWLJkiVmOzxLBtmvr\nap41yMzMhEajgUajwapVq/DTTz9h2bJlOHz4MPbt24fS0lK5QyQiK8Y5ckRkMkIITJs2DcuXL290\n/4oVK4y37aWkMC0tDWlpaY3uO378uPH2K6+8YumQ6AEaXoQQQqC6uhqenp5wcHDAM888AwCYOnUq\nJk2aZJYL1J2GAAAgAElEQVTjNy4RrG/awRLBlvPw8FDMKlxDK1euRF5eHgDg/Pnz2LRpE6KiooxD\n4BMTE3HmzBk5QyQiK8YVOSIymREjRmDLli3GcrCKigqcO3eOJYV32WMjC6UwXITQarUoKSnBqVOn\nsGjRovu+z1yrqkovEaTWKyoqQkFBAQ4dOoRjx44hMDAQPj4+/PeSiFqMiRwRmYyPjw+WLVuG2NhY\nBAQEIDY2FhcuXGBJIey7kYUSNHcRora2Flu2bAFQv4du6NChZotBySWC1HqVlZXo2rUrnJ2dUVpa\nioMHD+LGjRv45ptvUFlZiTt37mDr1q1yh0lEVoxz5IjI7Ox9rhlnXSnDp59+ij/+8Y+oq6uDk5MT\n1qxZg5EjR+KFF17Arl274OnpidzcXHTv3l3uUMkG3L59GxMnToROp8OAAQNw7do1LFmyBP/617/w\nzjvvoFu3bvD29kavXr2wdOlSucMlIjNryxw5JnJEZFY5OblISZkDJ6f6PUAbNqyTZaVBCCHbyqBG\no0FMzCxUVh413ufmFoz8/PehVqtliYlaxtXVFdevX5c7DLIjN2/eROfOnVFbW4v4+HikpKRgwoQJ\ncodFRGbGgeBEZFUs2VI9PT0dKpUK/v7+WLVqFXQ6Hby9vTFt2jSoVCqcP3/e5MdsKc66UjbuayRL\nWrhwIQYMGABfX188+eSTTOKIqFlM5IjIbCzVUl2r1SIrKwsajQYHDhzA+vXrUVFRgTNnzuDFF1/E\niRMn0Lt3b5MeszXYyEKZcnJyUVvrxH2NZDE5ObnIzPwYly49iv/7vysIDQ2XOyQismIsrSQis7HU\n3rCMjAyUl5cbZ3wtXrwYjz32GFauXIkffvjBZMdpL3vfK6gk3NdIlsZzjsi+sbSSiKyKXCtRhotD\nnTt3NutxWsvDwwNqtZofyhSAA7rJ0njOEVFrMZEjIrOyREv1iIgI5OXlobq6Gjdv3kReXh4iIyPt\nYh6TTqeDSqWSOwybw32NZGk854iotZjIEZHZmXslKigoCNOnT4darUZ4eDhSU1Ph7u5uN/Pr7OX3\ntCTua7QeWVlZmDdvntxhmB3POSJqLe6RIyKbY8t70dLT05GZmQlJkpCSkoKJEydi1KhRGDp0KPbv\n349evXrhs88+g7Ozs9yh2gRbPpeUIisrC0ePHkVGRobcoVgEzzki+8Q9ckRk93JycuHl5W2TnQYf\n1J1z3rx5+O6779ClSxds3bpV7lBtBvc1ml98fDzUajVUKhXWr18PAMjMzMSAAQMQFhaGffv2yRyh\nZfGcI6KW6iB3AEREptJwbl1VVX3Xt5SUaIwcOdwmPhQVFxcjPj4enTp1AgAkJCRg7969ePLJJ437\n5AYNGiRrcwQO0KbWyszMhLu7O6qrq6FWqzF69GgsWbIEJSUlcHNzQ1RUFIKDg+UOk4jI6nBFjqiN\nMjIy4Ovri+eee07uUOiuB3V9a+qqv6urK958800EBgZiyJAhihv6bChLb1hG6ejoiDt37sgVEvfr\nUautXLkSgYGBCAsLw/nz57Fp0yZER0ejW7du6NChAyZPNn2DJHo4V1dXuUMgoodgIkfURn/729+Q\nn5+PTZs2yR0K3fWgrm+ZmZnQaDTQaDRYtWoVysvLcfPmTQwZMgTHjh1DREQEPvjgAxmjfzhr686Z\nnZ2N0NBQBAcHY/bs2airq4MQQtHJMVlWUVERCgoKcOjQIRw7dgyBgYHw8fGxi46z1o4XZYisHxM5\nojaYPXs2zp49i7i4OLi7uyMlJQXR0dHo378/Vq9eLXd4dutBXd/uvep/5swZODs7Y/To0QDkL0ls\nCWvqzllaWorc3Fzs378fWq0WDg4OyM7OVlxyTPKqrKxE165d4ezsjNLSUhw8eBC//PILvvnmG1RU\nVKCmpgaffvqp3GHavKYqFoQQGDp0KK5cuQI/Pz+EhobKHCUR3Yt75Ija4G9/+xt27dqFwsJCrF69\nGl9//TUKCwtRWVmJAQMGYM6cOXB0dJQ7TLuUlDQZI0cOb9T1reFVf2dnZ0RHR6O6uhodO3Y0/pzc\nJYktlZaWhrS0tEb3HT9+3Hj7lVdesUgcu3fvhlarhVqthhAC1dXV8PT0vC85zs/Pt0g85padnY2M\njAzU1NQgNDQUKpUKZWVl+Mtf/gLA/jormkpcXBzee+89+Pn5YcCAAQgPD8cTTzyBJUuWICwsDF27\ndkVgYKDcYdq8e/cpJiQkQJIkbNu2DWPHjsWMGTOwZ88eucMkonswkSMygTFjxqBDhw7o3r07PD09\ncenSJTzxxBNyh2W3PDw8GjU3aeqqPwCbKd+So125EALTpk3D8uXLG92/YsUK422lJMcP03D10dHR\nEXPnzsWjjz6KvLw8YyKXm5uLN998U+ZIlcfJyQlffPHFfff7+PjA19eXLfgtZOXKlcjLywMAY8XC\n7du30bNnTxQWFqKurs6YyGk0GrzwwgvYunUr+vXrJ2fYRHaPpZVEJtCw2YSDg4NNfHi1JXFxcaip\nqYGfnx/eeOMNDBkyBIBt7AGRa9zCiBEjsGXLFuMeuIqKCpw7d85mkuOGGq4+BgUFoaCgAGVlZXjy\nySdx+PBhlJeX4/Tp08bzyhKGDh1qsWNZmi2PELFGTe1TrK6uRocOHeDo6Igvv/zS+L0HDhzAnDlz\nsGPHDiZxRFaAK3JEbWSLH1htVXNX/X/++Wfj7UmTJmHSpEmWDKvd5By34OPjg2XLliE2NhZ1dXVw\ncnLCmjVrbCI5vldzq4//8z//g9zcXHh7eyM+Pt6iMRUXF1v0eJZi6yNErFFzFQsODg7o1KkTSktL\ncfXqVZw8eRIvvPACvvrqKzz++OMyR01EAFfkiNqsuQ+stvhB1lbp9XpoNBrFdlZ80LgFS0hMTERJ\nSQm+/fZbaDQahIaG4ocffjA+p5MmTcKHH35okVjMqbnVx4kTJ+Kzzz7DypUr8c9//tPYvVMIgQ0b\nNhgHWs+cORMvvfSSSWNq2Br+7bffhr+/P4KCgvDGG2+Y9DiWJvc5fa/mxpYsXLgQAwcORGxsLA4d\nOoSoqCj0798fO3fulCXO9miuYsEgJycHJSUlAIBOnTpBq9XKESYRNUUIIetXfQhERJa1efPHwsWl\nm+jSJVi4uHQTmzd/LHdIrXb58mXh4tJNAN8KQAjgW+Hi0k1cvnxZlnhs4TltzieffCICAwOFv7+/\nGDx4sDh06JAQQoioqCjRuXNncefOHSGEEHPmzBEbN24U/fr1E9euXRN37twRERERYt68eSaNx9XV\nVQghxBdffCGeeuopUV1dLYQQoqKiwqTHsTRrO6cNz2dVVZUYOHCguHr1qpAkSezatUsIIUR8fLyI\njY0VtbW14ttvvxWBgYGyxGlqly9fFocPHxadO3cWQghRWFgoxo0bJy5fviz8/f1FYWGhzBES2Z67\nOVGr8iiWVhKZgBzNJqjtbKV8yzBuISUlGh07eqGmRmcct2BptvKcNicxMRGJiYmN7tPr9VCr1Th9\n+nSj7p0HDhzAsGHD0KVLF+PPnjlzxixx7d69G8nJycZ9uu7u7mY5jqVY0zkNNN0ExNnZGbGxsQAA\nlUqFTp06wcHBASqVCjqdTpY4TSknJxcpKXPg5NQXN2/eRE5OLp54or6U0sPDAzt37sTo0aPx4Ycf\nQq1WyxwtkX1jaSVRO3FjvvJYW/lWeyQlTYZOV4r8/Peh05UiKWmyLHHY0nPaEob3/erVOdDrK7Bg\nwasoKSnBqVOnsGTJEu6hbYekpMmIigrFr35Vjb59PXHjxs94//33sXDhQuP3ZGVlmbxc9V7NNQFp\nOLbEwcHBmERLktSuRlfZ2dkIDQ01lujW1dVZvIyz4QWZysqjAL5FSsoc+Pr64vPPPwcA9O7dGydO\nnGASR2QFmMgRtcO9/+lVVe1BSsocxe65shd9+/bF7dtlAAzz146jpkaHvn37yhdUO3h4eECtVsu6\n8mVrz+mDNHzfV1d/hTt3PPFf/zULer0eFRUVCAwMxDfffIPKykrcuXMHW7duNXkMhkQxJiYGmZmZ\nqKqqAlC/f8/U5OiQuXnzZpw8eRJarRYZGRlISEjA9u3bjX+fm5uLZ5991qwxtGVsSVsT+IYjLrRa\nLRwcHJCdnY1ffvkFI0eOxHfffYdHH30UixYtQkFBAbZt24Y//OEPbTrWgzR3QaakpETR+4mJbBVL\nK4nawfCfXn0pGdBwFcIWyslslbWVb9kCe3pO73/fv4Pbt6dhyJAhcHd3x9q1a/HGG28gJCQE3bp1\ng7e3t7HM0lQMTZWefvppfPvttxg8eLBxGPuyZctMeiw5OmTeW9L4448/4te//jUOHz6M/v37W2Tc\nw73DylsytqStza4ajrgwlOh6enrCycnJomWcjS/I1JdIV1V9j4kTk+DkVP93Gzask23ln4gak+Qu\n/5AkScgdA1Fb6fV6eHl5o6pqDwz/6bm4REOnK7XJD7C2hnsbTc8entOWvO/Lysqg1+vRu3dvzJw5\nEykpKZgwYYJZYjH38+3q6orr168jKSkJzz//PEaNGgUASE5Oxrhx45CQkGDS4xUVFeEPf/gDvv76\nazg7OyM6OhpvvfUWfvzxR5w4cQLe3t44ffp0o+HzSrdmzRpcuHDhvhEXbm5uxjEpb731FlxdXfHy\nyy/f93emZNgj17GjF27fPou6OoHbt78B/48jMi9JkiCEaNXVIJZWErWDYRXCxSUabm7BcHGJttlV\nCFt0b0lieno6VCoV/P39kZGRAaDp9uPUPGso8zS3h73vc3Jy8Zvf+CAsbBieeOJXqKm5Y5YkzlL7\ncw2rTJMnT0Zubv0xampqUFBQgDFjxpj8eM2VNBrGPXz88cdmL6tsKVONMGluxIU5yjgfpuG+288+\ny4WLy69hL3tfiRSntW0uTf0Fjh8gG2Bo1SxXi2x7Vltba5LHOXr0qPD39xdVVVXixo0bws/PTxw7\nduy+9uPl5eUmOR4pX1Pve0u1z7dkm37DqIPq6mrh5eUlbt++LT777DMxdepUkx9LCCFu3bolRo0a\nJXx9fUV8fLyIjo4WRUVFQgghxo4dK/r372+W47aWqcdt3Dvi4uDBg8bnXgghlixZIt59913jnxv+\nnamUlZWJgQMHGv9sbeMgiGwZOH6ASB4eHh42vQIhp/j4eJw/fx7V1dWYP38+ZsyYAVdXV7zwwgvY\nvXs31q5da5K9MsXFxYiPj0enTp0AAJMmTcLevXuRl5dnbLJgaD8eEhLS7uOR8jX1vrfUvlk59uc6\nOzsjKioKX375JXJzc5GUlGSW4zg5OeGLL75o8u927NhhlmO2ljnGbTQ14qJh6eScOXOMJbseHh5m\nKasEGu/zs6e9r0RKxNJKIrJqmZmZ0Gg00Gg0WLVqFcrLy3Hz5k2Eh4ejpKTEbA0PDFe7du/efV/7\ncaLmWKp7pyW7hIoGJXzPPPMMMjMzUVxcjLi4OJMfqzmmKmE0FUuP2zBXGW1T5eQGP/74I4KDg/Ef\n/9HfKkacENH9mMgRkVVbuXIlAgMDERYWZlwR69Chg8kbLERERCAvLw/V1dW4efMmtm/fjmvXrjW5\nV4esS8P2+AsWLIBKpcKrr74qSyyW2jdryf25DVdoYmNj8c033yAmJgYdOlimqMcaZ3VaMpE215gb\nrVaLrKwsaDQaHDhwAB988IFxfMW//vUv/Od//ic2btyIQYMG2cXeVyIlYmklEVmthgN5Dd3rqqur\n0alTpza3+W5OUFAQpk+fDrVaDUmSMHPmTMyaNQsTJkwwth8PDw836THJNBq2xzd8GDX1+dEaSUmT\nMXLkcLN3k7TUcRqW8HXo0AFXrlwxy3GaYo4SRlOwZMmhucpo7y0nT0hIwN69e3H58mVMnDgR27Zt\ng7e3twl+AyIyFyZyRGS12jKQtz3S0tKQlpbW6L7m9urILSsrC0eOHMHq1avlDkV2hvb4EyZMwI0b\nNzBo0CC8/vrr9+03siRL7Zu15P5cOUZLWPOsTksl0k3NdjPH6p8QApIkoUuXLujTpw/27t3LRI7I\nyrG0koisVlxcHGpqauDn54c33nijRQN5Tc3a9uY0JOeqkzUxPA+fffYZHnnkEWi12gcmcYZSTJ1O\nh5ycHIvEqHRylTdasoSxLSxRcmiuMtp7y8nz8vIQEREBZ2dnbN++HRs3buT7g8jKcUWOiKxWc93r\nzNWt7V6GwbhOTvUfJjdsWGeRjf46nQ5xcXEICwvD/v37MXjwYEyfPh1LlizBlStX8NFHH5k9BiVp\n7QqtoRTz7Nmz2Lx5s9m6L9oKOcsb2TWxnjlW/+4tJ09NTYW7uzsAwMXFBTt37kRsbCxcXV0xduzY\ndh+PiExPMleJUosDkCQhdwxEpAyWLO3S6/Xw8vJGVdUeGMqZXFyiodOVmv3YOp0Ov/nNb3Ds2DH4\n+vpi8ODBCAgIwIYNG/D5558jMzMT8fHxOHLkyH2d5uyRm5ubMblveLs5hlLM8PBwlJaWol+/fpg2\nbRrmz59viXAVR6PRICZmFiorjxrvc3MLRn7++1Cr1RaJQY6yTiIiS5IkCUKIVpXatLu0UpKkXpIk\nFUiS9L+SJJ2QJOmlu/d3lSTpK0mSTkuStEuSpC7tPRYRWV7DjoBysnRpl6Xbi9+rX79+8PX1BQD4\n+flh5MiRAACVSmWxGJSi4cXAllwYNJRi/vnPf0ZERAS0Wi2TuAewhvJGdk20DGsuJSei+5lij9wd\nAC8LIfwAhAOYK0mSN4DXAOQLIQYAKADwugmORUQW1rAjoFzM1X77QeT+8Ors7Gy87eDgYPyzg4MD\n7ty5Y5EYlKLhXkHuGzQ9S446IPlY45gHInqwdidyQoiLQohjd2/fAHAKQC8AEwBk3f22LAAT23ss\nIntjDathrq6uxtsrVqxASEgIAgMD8dZbb1ksBjlWx+T+8MqS85b74YcfjKsIlto/aW+SkiZzKLQN\nk+NiGRG1n0m7VkqS1BdAIICDADyFEJeA+mQPQA9THovIHljDaphhhePrr7/GmTNncPjwYZSUlODI\nkSMWi0+u1TE5P7w+aJWJq07/1pZVBEOSbNgrRy3D8kbbJXcpORG1jcmanUiS9CiAQgBLhRCfSZJU\nLoTo1uDvrwohujfxc2x2QtQMwwfNF198Ebt370bv3r3RsWNHpKSkICEhwSIxGJpHLFiwAFu3boW7\nuzuEELh58yZef/11JCcnWyQOQwfJhp3ruCpg39rakMZwTt+5cwdPP/00ysvLMX36dO6TI7slZ3Mn\nIqrXlmYnJhk/IElSBwBbAGwSQnx29+5LkiR5CiEuSZL0OIDLzf38kiVLjLejoqIQFRVlirCI2mzo\n0KEoLi7GhQsXMH/+fHzyySeyxCFJErZv344zZ87g1KlTuHDhAnx9fZGSkmLxWIQQeP3115Gammrx\nYwOWG75rzSzZuc/wHrBmbR0WbSjF7Nu3L3bv3m2ZYMmqKeF8NyeOeSCyvMLCQhQWFrbrMUyyIidJ\n0kYAV4QQLze4720A5UKItyVJehVAVyHEa038LFfkiJrh6uqK1NRU+Pv7Y/r06QCASZMmYcqUKRZb\nkTOsCn799ddYtGgR8vPz0blzZ/z000/o2LEj/6O3ELlm2lmztqwi8Hkkah7HPBDJR67xA08BmAJg\nuCRJJZIkaSVJigPwNoAYSZJOAxgB4M/tPRaRpRgafOh0OqhUKpmjkZdhP1ZMTAx++9vfIjw8HP7+\n/khMTMSNGzdkjs4+yNGIwPAeuHnzJkaOHGmcZff555+b7Zit1dqGNGzoQM0xnO9FRUWIiorCxIkT\n0b9/f7z22mv46KOPEBISgoCAAJw9e1bmSM2L+yCJlKXdpZVCiH0AHJv565HtfXwiOVhTO/PIyEi8\n9957eP7553Hp0iXs2bMHU6ZMsdjxDV0AtVotVq1ahQMHDjT6T37x4sUYNmwYhg8fbrGY7E1bSwjb\nw3Ded+rUCXl5eXj00Udx9epVhIWFYfz48WY5Zlu0puRWjueRlKHhv/PHjx9HaWkp3N3d0a9fP6Sm\npuLw4cPIyMjA6tWrkZ6eLmOkRET/ZtKulURkWg4ODpg4cSJ+85vfwM/PD9OnT8eQIUMsHkdOTi6G\nDBmOs2f/332dAd966y0mcWYm50w7w97IgIAAjBw5Ej/99BMuX252y7MsWrqKIPdsQHMxjCkpKirC\nuHHjZI5G+dRqNXr06AEnJyf0798fTz/9NABApVKxi2M7NTdSJzk5Gdu2bbNwNETKx0SOyEpdvXoV\n3brVN35dtGgRNm7ciI8++sjiKweGcrRbtz5GXV0/VFVFY8qUKRg+fDiqq6v5H7AFyDnTLjs7G1eu\nXEFJSQlKSkrQo0cPVFdXm/245iD3bEBzadikQ+4KAlvg7OxsvO3g4GD8s4ODA+7cuSNXWDbBnhvK\nEJmDSbpWEtkauRvwXLhwAVFRUViwYMF9zRlCQvwf/gAm9O9yNB8AZwDkwtX1R3To0AFbt27lB0cL\nsXTXTsN7oLKyEj169ICDgwP27NkDnU5n1uOamy12P204D6+yshJjx47F999/j+HDh2PdunUyR6cM\ncv+bby8anqv3jtUhotZjIkfUBLn3yPXs2ROnT59u1JWvfl/PcRw+HI2IiAiLxfLvcrRTAJ4EIFBT\no0NY2FiWGVmYh4eHxRIPw3k/ZcoUjBs3DgEBARg8eDB8fHwscnxzsuTzaAoPG1ze8N8ojUaDU6dO\noU+fPnj66aexbds2i3W4VbLm/p2X60KVrY5DMDyf27Zts4qxOkRKx0SOqAmGBh9XrlwxljfKwRqa\nMxjK0ZKTn0VNzS04O9eXo/3003l2rbRhhvdA9+7dsX//fpmjsW+tSSZCQkLg5eUFAEhKSkJxcTET\nuRYwnO/Dhg3DsGHDjPcXFBQYb9/7d+Zki0lcQ3v37kVSUhKA+guX3GdN1DbcI0d0D71eD41Gg/z8\nfEyZMgVpaWmyxWItzRmSkiZj//4C9Ov3K+h0pZy7ZUcM7we26LeM+Ph4qNVqqFQqrF+/HkB92d/L\nL7+MgQMHIiYmBlevXm325+9N+lj63HZynvsNxyFER0cjMTERPj4+eO655yweCxFZLyZyRA3k5OTC\ny8sbMTGzMH78ZCxe/BYmTJggWzzW1Jyhe/fueOSRR4zHliTJ+CGRHxZtU8P3w73dSsk8MjMzodFo\noNFosGrVKpSXl+PmzZsICQnBd999h8jISCxZsqTRzzTc33Xo0CHodDrU1dUhNze32S6B9GByn/sN\n/009duwYMjIycPLkSfzwww+KXiE3nKuRkZHIzc1FXV0dLly4gD179sgcGZEyMZEjustahwUnJU2G\nTleK/Pz3ZV0N8/LywvHj9SuDer0eERERmD17tuzlp2Qe1vp+sHUrV65EYGAgwsLCcP78eZw5cwaO\njo545plnAABTp07Fvn37Gv1Mww/9ISEhePHFF+Hn54df//rXiI+Pt2j8tsDazv2QkBD07NkTkiQh\nMDBQ0XuTDedqfHw8+vfvL+tYHSJbwD1yRHdZw3605lhTc4aGXTRv3PgOPj4DeNXfBlnz+8FWFRUV\noaCgAIcOHYKzszOio6ObHPVw7wq4YX+Xr68v3nnnHdm6cep0OowdOxYnTpyw+LFNydrO/YbjEBwd\nHRU9AsFwrur1ejz//PNYtGgR/z0hageuyBHdZS370azZvVeqa2s1+OGH/4fy8nK5QyMT4/vB8ior\nK9G1a1c4OzujtLQUBw8eBADU1tZiy5YtAOrn+jV14UTuUkADWyizbum5b84LWLY8DsFazlUiW8BE\njugua9qPZq0MV6qB+69Uk23h+8Hy4uLiUFNTAz8/P7zxxhvGcrNHH30Uhw8fhkqlQmFhIRYtWtTo\n5+QqBUxPT4dKpYK/vz9WrVoFAKipqcHUqVPh6+uLZ555RpHD41t67puzs6S1jUMwFWsrWyVSOknu\nqz6SJAm5YyBqSK/X29SwYFNqONeuPpk7DheXaOh0pXyubJQ9vB8yMjLw3nvvYdCgQdi0aZPc4bSa\nRqNBTMwsVFYeNd7n5haM/Pz3oVarzXJMrVaL5ORkHDp0CLW1tQgLC8OmTZsQHByM/fv3IywsDCkp\nKfDz88PLL79slhjM7WHnvmG+3+LFi/H5559DkiRcvnwZTz/9NDZs2CBDxNZPjnOVSCkkSYIQolVX\na5jIEVGrGPbIdezohZoaHTZsWMdxBKRoPj4+2L17N5544gm5Q7lPSxJpOS6wZGRkoLy83NhBc/Hi\nxXjsscfw7rvvGlfo9+zZg9WrV2Pbtm1miUFubm5uxj1fQH1pbGRkJLKyshAYGGiSY9jahRReDCRq\nXlsSOZZWElGrWEsXTSJTmD17Ns6ePYu4uDi4u7tj+vTpiIyMRL9+/bBt2zYsWLAA/v7+GD16NGpr\nay0aW0v3EllDGazhgqw9z7GbOnUqXnnlFZMlcba4l+zJJ5+U/VwlsiVckSMiq5Seno7MzExIkoQZ\nM2bgpZdeanRfSkoK5s+fL3eYZAOefPJJHDlyBKtXr8bu3btRWFiI7777DmFhYcjLy0NsbCwSEhIw\nffp0jB8/3iIxtWXlwpKrNyUlJUhOTsbBgwdRW1uL8PBwbNq0CUFBQThw4ABCQ0ORmpoKX19f/Pd/\n/7dZY5FLwxW5JUuWQK/XY+3atSZ5bFtduTI8Z7a20khkClyRIyKboNVqkZWVBY1GgwMHDuCDDz5o\n8r5vv/1W7lDJxowaNQoODg5QqVQQQiA2NhYAoFKpLNrUpy2NhTw8PKBWqy3ywTgoKAjTp0+HWq1G\neHg4UlNT4e7uDm9vb6xduxa+vr64du0aZs+ebfZY5GK4CL1jxw7k5+cbG76Ygq03lnrppZdw5coV\n47manJxssyW4RObEOXJEZHWKi4sRHx+PTp06AQASEhKavG/v3r0ICAiQM1SyMYaZXZIkoWPHjsb7\nHRwcLDq/q3EL/PoVGWsb/5CWloa0tLRG9508eVKmaCzPUDb617/+FT/99BPUajUkScL48eONewfb\nSrbHKGsAACAASURBVAmvf3tMnjwZubm5GDVqFGpqalBQUID33ntP7rCIFIeJHBFZPSGEoeRA7lDI\nBjV3Xsl5vhn2vaWkRDdqLGTtZWj2VDJnKKssKCgw+WMr9fVvqVGjRiEtLQ01NTX45z//icjIyEaD\nz4moZVhaSURWJyIiAnl5eaiursbNmzeRl5eHoUOHNrpv+/btiIiIkDtUMqGlS5fC29sbkZGR+O1v\nf4v09HRotVoEBgYiKCgICxcuhEqlMvlxrXVml9IaC9lic46H0ev10Gg0ZpmDprTXvzWcnZ0RFRWF\nL7/8Erm5uZg82XZ+NyJLYrMTIrJKK1euxIYNGyBJElJTUzFv3jz89a9/xYcfftjoPrINR44cwcyZ\nM3Ho0CHcunULwcHBmDVrFrKysrBu3To89dRTWLhwIb788kscP35c7nDpHrbanONBDKNYnJzqyyA5\niuXhDLP3AOCLL77A+vXrcfToUfzwww/o0IFFYmTfOEeOiGyaPZVt2ZtVq1bh2rVrWLx4MQDgd7/7\nHXr27InVq1cbGzycOHECU6ZMsVgix/Ot5ext0LM9Jq6m0LDT5507d/D4449j4sSJWL9+vcyREcmP\nXSuJyGbZY9mWPZP7Ah/Pt9Zp3JwDsLXmHPey9a6S5tJwgHqHDh1w5coVJnFE7cBEjoisnl6vR0rK\nHFRV7UFl5VFUVe1BSsocs+xLIXk89dRT2LFjB27duoUbN25g586dkCQJ7u7u2LdvHwAgOzvbIrHw\nfGs9axhKbkkPS1zT09OhUqng7++PVatWQafTwcfHB8nJyRgwYACmTJmCr7/+Gk899RQGDBiAI0eO\nyPSbWJ459xUS2RsmckRWZOjQoXKHYJVs/eo3X3dg8ODBGD9+PAICAjBmzBj4+/ujS5cuyMzMxNy5\ncxEcHGyxWGz9fDMXW27Oca8HJa73zrxcv349Kioq8P3332PBggU4ffo0Tp8+jY8//hj79u3DO++8\ng+XLl8v9K1kEV7qJTIt75Iis2NixY7F582a4ubnJHYqsuB/FPty8eROdO3dGVVUVIiMj8cEHH+BX\nv/qVcZ/aL7/8grFjx+LEiRNmjYPnG7VUU/soMzIyUF5ebpwlt3jxYjz22GNYs2YNTp8+DQCYNm0a\n4uLikJSUhLNnz2LSpEnQarVy/RoWwfcV0YNxjxyRgjQsvcnIyABQ39GroZ07d9p9Egc8+Op3dnY2\nQkNDERwcjNmzZ6Ourg5z5sxBSEgIVCoV3nrrLbnDf6iGr/uCBQugUqkQEBCATz75RMaoLG/mzJkI\nCgrCoEGDkJiYiFOnTje6ev/ZZzssMhLA3soEqe08PDygVqsfeG4YLlY3nJPm4OBg/LOlh83LhSvd\nRGYghJD1qz4EIvty9OhR4e/vL6qqqsSNGzfEwIEDRVRUlHBwcBADBw4UH3zwgRBCiL59+4qrV6/K\nHK31uHz5sjh8+LC4fPmyEEKIU6dOiXHjxok7d+4IIYSYM2eO2LRpk6ioqBBCCFFbWyuioqLEiRMn\nZIu5JVxdXYUQQmzZskXExsYKIYS4dOmS6NOnj7h48aKcocnm8uXLwsWlmwC+FYAQwLfCxaWb8bW3\nVAwNzzeiltBqtSIgIMD477tKpRLHjh0TAwcONH7P9OnTxdatW4X4/+zdeVRT5/Y38G9ARBQQtKJt\nVUStMgYSQFEEBBF564gtUhwqSPWqlet021vtr4pVe2vVVq1DbUWcAFEcqtYOzggORIJiVagTsbYq\noMhgARme9w9MDBiUIck5CfuzFmtBCDk74eTkGfdmjGVnZ9f4nb7iw3uaED571idqUD+KinYQwoHk\n5GQEBQWhVatWAIDRo0ejTZs2SEtLg0Qigbu7O0aPHs15QWK+6dChQ42R72PHjkEqlcLd3R2MMZSW\nlqJjx45ISEjA999/j4qKCty/fx9Xr16Fo6Mjh5HXT0pKCkJDQwEAVlZWGDhwICQSCYYNG8ZxZNon\nH70vKXlx9F5bM2O1zzdC6kMkEiEsLAzu7u6KmpcWFhY1rue1r+3N4Vovn+mOiPCFkZE1ystlNNNN\nSBNRR44QHmCM4cSJE3jy5Ak8PDxw9+5dXL9+neuweI8xhokTJ9ZIFJCdnY3BgwcjLS0N5ubmCA8P\nR2lpKYdRNh5rxvuHa2YFrN5Po8/p7Il+mTVrFmbNmlXjNuX6h5s3b1Z8b21t3WyK3IeGhsDf34/q\nMxKiJrRHjhAOeHl5Yf/+/SgtLcWTJ08QGxuLBw8eoHXr1rh48SJcXFx0tvOhTYMGDUJiYqIijXV+\nfj7u3LkDU1NTmJmZ4cGDB/j55585jvLV5B02Ly8vJCQkoKqqCrm5uTh9+jT69OnDcXSaFx4ejr17\n99a4jfapEX3WnFPw12dfISGkfmhGjhAO1F56M2jQIDx48AA3btxAZmYmzp07B6B5z8jUh52dHZYs\nWYKAgABUVVWhZcuWWLduHUQiEezs7NClSxedSO0vX1YVFBSEc+fOwdnZGQYGBli+fDmsrKw4jo47\nNHpP9FF8fAIiIqajZcvqWefo6PV6XaqB1GRjY4O0tDS0a9eO61CIHqDyA4TwwNOnTzFq1CjIZDL0\n7t0bBQUFWLBgAcLDw3HhwgW64DcTqlKZ66Nt27Zh5cqVMDAwgFAohKGhIYyNjXHhwgUUFRVh5cqV\nGDp0KNdhEqJ2lIKfdO/enT7XiUqNKT9AM3KEcEi54X748OEXGvK3bt3iOkSdpIsdouYySn/16lV8\n8cUXOHv2LCwtLfH48WPMnj0bMpkMEokEN27cgK+vL27evImWLVtyHS4hasWHJD5Ee2JjY7FmzRqU\nl5ejb9++WLduHa20IWpFe+QI4Uh8fEKNGlmRkTNr/Bwfn8B1iDqp9uuqC69jbm4uIiKmo6TkBAoK\n0lBScgIREdP1cv/M8ePHERwcDEtLSwCAhYUFAGDMmDEAgJ49e6JHjx7IzMzkLEZCNKVmEh+Akvjo\nr8zMTCQkJODMmTOQSqUwMDBAbGxss8hQSrSHOnKEcEBVw33t2h9QUrJH7xvymqSrHaLGFspVLiq/\nevVqrFixAmvXrgUAzJ49G4MGDQIAnDhxAuPHj9dY/Oqg3LhhjFFjh+glSuLTfCiXxxGJRDh+/Dhu\n377NdVhEz1BHjhAOqGq4A50BtFH8XJ+GPKmpsR0irjVmlF4qlWLr1q2QSCQ4e/YsNm3aBC8vLyQl\nJQEA0tLS8OTJE1RWVuL06dPw8fHR9NOoFz8/P+zevRuPHj0CUJ1pFAB2794Nxhhu3ryJ27dvo3fv\n3lyGSYjGhIaGQCbLxNGjGyGTZerlEmryvDyOVCpFeno6rl27hgULFtDSSqJWtEeOEA6oqpEF3AXw\n5Nk9aLlNY+hq7bHGFMpVVVT+/PnzkEqlKCoqgrGxMVxdXfHRRx/h+++/h5+fHyZPnqytp1Qne3t7\nfPrpp/Dx8UGLFi0gEokgEAjQtWtX9OnTB0VFRdi4cSPtjyN6jYrN679BgwZh1KhRmDVrFjp06ID8\n/HwUFRVxHRbRM9SRI4QDqhruERGTER39Tr0b8o0lk8kQGBgIDw8PnDlzBm5ubggLC0NUVBTy8vKw\nY8cOjBkzBpcuXYK5uTkAoFevXkhJSeF9w6MxHSK+aGqqfcYYDAwMYGNjgy1btsDT0xNCoRBTp05F\n+/btceDAAQ1F3nATJkzAhAkTuA6DENIIZmZm1CGpB1XlcdauXUvLxolaUfkBQjhUO7uiNrItymQy\nvPXWW7h48SLs7e3h5uYGZ2dnREdH4+DBg9i8eTO6desGZ2dnhIWFITU1Ff/3f/+H3377TSPxaIK6\nXsfaGcfWr1+Po0ePYuHChXj69Cl69OiBmJgYtG7dWo3R1096ejrCw8Nx7tw5VFZWwsPDAzt27MD+\n/fuxefNmxMTEYNu2bdi6dSvMzc0hEAhw584d3nbOdTHTKCHNkbm5OQoLC7kOgxC905jyA7RHjhAO\ndejQAe7u7oqGa+2fNcXGxgb29vYAAAcHB/j7+wMAHB0dIZPJEBISgoSE6myPO3fuREiIbu3hUMfr\nqCrj2I4dO7BkyRIcO3YMFy5cgKurK1auXKnGyOtPuah8v379MGXKFDg7O8PLywv3799Hv379sGXL\nFhgZGeGTTz5BeHg49u3bBwBITU3lVYdJFzONEtIcxMbGom/fvhCLxZg2bRqqqqoUe7zy8vLQv39/\n/PzzzxxHyW+5ubmQSCS8T7pFdBMtrSSkGTI2NlZ8b2BgoPjZwMAAFRUV8PDwwM2bN5GXl4f9+/fj\ns88+4ypUzihnHGOMobS0FBKJBNnZ2fD09ARjDOXl5ejXrx9nMc6aNQuzZs2qcZufnx/KysoUP7/5\n5pv417/+haysLHz++eeYOHEirzrnyplGq2trZSAiwhf+/n686WgS0hwpD2YZGhriww8/VKTPz8nJ\nwYgRI/DFF1/Az8+P61B5q7nUByXcoY4cIc1QfZYzjxo1CnPmzIG9vb2i5ldzIs84tnTpUsVthw4d\nQnx8PGJjYzmMrH7kSxWrqqoAAP369eNl55wKJBPCT6oGszp27IinT5/C398f69atg5eXF9dh8hYN\nUhFtoKWVhDRDyput69p4HRISgtjYWLz33nvaCotXBg0ahMTERMVymPz8fAiFQqSkpODmzZsAgH/+\n+QfXr1/nMkyVlJcq/vnnXezZU72kMigoiHed81eVXjAzM+MoMkKat7rS57do0QKurq745ZdfuA6R\n13S1HA7RLZTshBBC6rB792588cUXioxj69atwz///IOPP/4YZWVlEAgEWLJkCYYNG8Z1qAq5ubmw\ntrZFSckJyOsTtmr1BHfu/IE7d+6gT58+2Lp1K68KhMuXHylnGpUvP6LECuo3YMAAJCcncxqDTCbD\nsGHDcPny5Rq3JycnY+rUqWjZsiXOnj1bYxk40a5r165h1KhRSE5OrpE+39HREQUFBXj33XfRt29f\nfPzxx1yHyksvXoszYGLiC5ksk2bkiEqNSXZCHTlCSA1NzR749ddfIyYmBgKBAB988AFGjhxZo8G2\ncuVKPHnyBAsWLFB36BqnC5kVJRIJBg+eioKCNMVt5uZiHD26Ee7u7hxG9nJ1vbbyjtzEiRPxzjvv\nYMSIEQCA8ePHIyQkBMOHD+cqZNIEMpkMw4cPR0ZGRo3bp02bBi8vL4wdO5ajyIiy2oNZa9euxeDB\ng1FYWIinT59i5MiRGDlyJKZOncp1qLz0skEqQmqjjhwhpEmaujFbKpUiPDwc58+fV6TE3759O95/\n/31Fg01XO3K6smm9rlHgtLRkFBcX87oTqoq8I5eUlIRvvvkG+/btQ2FhIUQiEa5fvw4DA9oh0FDy\nOmD3799HSEgIioqKUFFRgQ0bNsDT01MrMcjrWbq6ukIqlcLBwQHe3t6IioqChYUF+vfvj+3bt9c5\nc0eIrtCFAUDCD1R+gBDSaMobswsK0lBScgIREdMblDI5OTkZQUFBaNWqFdq0aYPRo0fj9OnTGoxa\nO9Tx2miLvCi6iYkvzM3FMDHxRUTEeLi6DtDp9P7e3t64ceMGHj58iPj4eLzzzju86cTJZDI4OTm9\ncHtWVhZEIhFcXV1x+/ZtDiJTTb4vNi4uDoGBgZBKpbh06RJcXFy0GkdWVhZmzJiBq1evom3btooZ\nnuXLl2P79u0vxEv4QVfT6XO131ZbZYVI88SPT0FCCOc0sTGbMYbHjx8rMicCQGlpaVPC5ISubVoP\nDQ2BTJaJo0c3Ii0tGdHRO3SiE/oq77//PrZv346YmBhMmjSJ63BqUNXZ2L9/P4KDg5GWlgYbGxsO\nono5d3d3xMTE4PPPP0dGRgbatGmj1eN37doVHh4eAIBx48bVuW+voqICU6ZMgaOjIwIDA1FWVoY1\na9bAwcEBLi4utAxTi3S55iMNCBB9RB05QgiAV2cPrA8vLy/s378fpaWlePLkCfbv34+3334bOTk5\nyM/PR1lZGQ4dOqSB6DUnKCgIU6ZMQWHhJQBRAKoAjERRUQbCwsKwevVqbgOsg3wUuLi4WKc6obUp\nL72fOHEiVq1aBYFAAFtbWw6jelHtzsbhw4exatUqbNiwAYMGDeI6PJW8vLyQlJSEN998E2FhYdix\nY4dWj1+7YV1XQ/v69euIjIzE77//DgsLCyQmJmLZsmW4ePEiLl68iO+++04b4TZ7urQy4euvv4aT\nkxOEQiHWrFkDoH5ldwjRNdSRI4QAUL0kLzp6fYOWg4hEIoSFhcHd3R39+vXD5MmT4erqis8++wzu\n7u4YMmQI7OzsNPgs1C8mJgbp6enPErgsQZs29jAwOIzY2FhcuXIF4eHhXIf4UurooHNJuXFvZWUF\nOzs7Xr7mtTsbjx8/xrRp0zB79mwcO3aM6/BqkDdo79y5AysrK0REROCDDz6AVCrVahwymQznz58H\nUL3M08vLS2Vju3v37oqlq2KxGNnZ2XB2dsbYsWMRGxsLQ0NDrcbdXOnKygSpVIqtW7dCIpHg7Nmz\n+OGHH3Dx4kWakSN6iTpyhBAF5SV5Mllmo5J5zJo1C5cvX0ZGRgYiIyMBAJGRkbhx4wZOnjyJzZs3\n61Sik1WrVsHFxQXffLMS5uam+Pbb/8LaugvOnTuDX3/9lfd1ztTRQW+KrVu3Ks6DxlAuPfDPP//g\nxo0bCA0NVUdoaqXc2bCxscGcOXNe6JT4+vpqvbOkirxBe/LkSTg7O0MsFmPXrl2YOXOmVuOwtbXF\nunXrYG9vj4KCAkybNk1lY1u5BIGhoSEqKyvx008/YcaMGYqC1crLt4lm6MqgkL7u1SZElRZcB0AI\n4ZcOHTqovZGvq1m7Tp06hePHj+P8+fMwNjaGr68vevTogd9//x2//vorNm7ciF27diE6OprrUF8q\nNDQE/v5+nP0P1DESvmfPHkRGRiIyMpKXnWflzgZfkrDURd45fv/99/H+++9zEoO1tTWuXr36wu2b\nN29+4bbaHeKqqircuXMHPj4+6N+/PxISElBcXAxzc3ONxUueDwpFRPjWSKfP92s6Lakk+ozfnzaE\nEJ2ny5vjCwoKYGlpCWNjY2RmZuLcuXPIy8tDZWUlgoKCsHjxYqSnp3MdZr1oKnNabGws+vbtC7FY\njGnTpoExhpiYGPTu3RseHh5ISUlp8jHi4xMwYcIU/PPP61i8eAUvz6HajcXKykocOHAAy5cvR2Bg\noCLJz65du9C3b1/Y2tqq5bVpLD5lHnxVLMoDAQKBAJWVlRg/fjyEQiFcXV0xc+ZM6sRpiTpWbWia\nqr3a3t7e1KEjeolm5AghGqO8Ob6kpLqmWUSEL/z9/Xg/igsAgYGB+O677+Dg4IDevXujX79++Ouv\nvzBw4EBUVVVBIBDgyy+/5DpMzmRmZiIhIQFnzpyBoaEhPvzwQ2zfvh2LFi2CVCqFubk5Bg4cCLFY\n3Ohj6Mo5VHvW8dGjRxgzZgx69OiB1NRU7NmzB0B1B+/8+fP4+eefERUVhSNHjmg9Vj7VRJTHUlVl\njMrKx+jVywbbt29XnDPW1tY1iobPmTMHubm5GDFihM7N8OsLTazaUCflvdoCgQCTJ0+Gs7Mz7ZEj\neokKghNCNEYikWDw4KkoKEhT3GZuLsbRoxvh7u7OYWTaFxQUhLt376K0tBT//ve/YWlpibNnz2Ll\nypVYvXo11qxZg5s3b+L27duYMGFCnanY+WTdunX43//+BysrKzDGUFpaChMTEwiFQmzZsgUA8O23\n3+L69euKzHENpYvnkEwmQ0BAALKysgAAX331FcrLy3Hs2DEsXboU/fr1Q05ODgYMGIA//vhDq7HV\nVTBeJsvUeOP866+/fpY0SIAPPvgAI0eORPfub6GqqjuAfAB9YGDwC3777Zc6M33yqRNKCCHqRAXB\nCSG8oiub4xuisUvSYmJiIJFIIJFIsGbNGnh6eio24CcnJ+O1117DvXv3cPr0afj4+GgidLVjjGHi\nxImQSqVIT0/HtWvXEBUVpdYlTLp2DuXm5uLSpUs1MikaGhqioqICwPO9dMq3aRNXmQdrZxL8/vvv\nERoaiqqqSgBWAAYB8IBA0Ap///03GGMIDw+vkRhJl9LfE+2xsbHBo0ePXridT8uHCdEU6sgRQjSG\n64yJ6taU/X7y7JceHh64e/cu7ty5gydPnqC4uBh//vknxo4di1OnTuH06dPw8vLS4LNQn0GDBiEx\nMVHRUMrPz4eLiwuSkpKQn5+P8vJy7N69u0nH0KVzSH5+jB8/H5mZf7zy/OBiNQpXHePamQT79++P\n27dvo3q1WzcASQDywVgpOnTogHHjxqFXr174/PPPFY+hK+nviXapWjKp7r3ZAwYMAFA92y7PTksI\nH1BHjhCiUbqwOb4+mjIboJz98uLFi3BxcUFZWRn69++PmJgY2NrawsvLC6dPn8a5c+fg6emphWfU\ndHZ2dliyZAkCAgLg7OyMgIAA3L9/H1FRUfDw8ICXlxfs7e2bfBxdOIeUz4+iop/AWHfF+SEQCFQ2\nNrnYs8OXjrFMJoNIJML33/8AI6O9EAhyYGCwGra2vfDJJ5/AyckJ8+bNq/E3ujY7S9QvKCgI7u7u\ncHJywqZNmwC8OCCiiZlb5aXutNeO8ApjjNOv6hAIIYTfUlNTWdu2YgYwxZe5uYilpqa+8m9//PFH\nNmLECMYYY9euXWOtWrVip06dYlu2bGFdu3ZlmzdvZpWVlczOzo65urpq+qkQDWjK+cGFnJwclpqa\nynJycrRyPKlUypydnVlJSQkrLi5mr7/+Ops0aRIrLS1lOTk5zMfHh9na2rKBAwey6dOns8GDB7PS\n0tIXHicubiczMWnHzM1FzMSkHYuL26mV+NUhOzubOTo6ch2GTsvPz2eMMVZSUsIcHR3Zw4cPWbdu\n3djDhw8V99HEe9HU1JQxVv0/dHJyatqTIKQOz/pEDepHUdZKQgiph5qzAdVJIuo7G6Aq+yVQnSb7\n7t278Pb2hoGBAbp27Qo7OzsNPgvt0NW6gU1Rn/ODT6+LtjMP1s4kOH78eOzZs0fxc1ZWFqZNm4ZL\nly4hIiICSUlJGDNmDPbu3VtjvyHXNRGbimZzmmbVqlXYv38/AODu3bu4fv36C69pU67VdaH/G+Er\nWlpJCCH10JQlaS1btsThw4dx5coV7N27F8ePH4e3tze6d++OyspKmJubQyKRYPv27fjmm2+08Gw0\nR5frBjbFq86P5vq6KJs1axYuX76MjIwMfPXVV/jggw9QVlaGtm3b4t1330WXLl1q3FckEqksWK6p\nmojaUFFRgSlTpsDR0RGBgYG4evUqXF1dFb+/ceNGjZ/Jc6qWqMvrMyrjy/JhQrSByg8QQkgDqHtW\nRZ/SqXOZ2p4vVJ0f6npdZDIZhg0bhsuXL2smeC2r67XS1dm2V5HJZOjZsyekUimcnJzw3nvvYfjw\n4di8eTO++eYbCIVCfPrpp3jjjTfw4Ycfch0u7xw4cADR0dH48ccfkZmZCZFIhF9++QVhYWFIS0tD\nu3btatxfneeSubk5CgsLIZPJMHz48Bq1DQlRFyo/QAhPmJmZAQDu3buHMWPGcBwNUSd1zgboWzp1\nyiqo+vxQ5+uiL0u8VM1QNodZy+7duyuyHorFYshkMnzwwQfYvHkzqqqqkJCQgLFjx3IcJT8FBgai\nvLwcDg4OmD9/Pvr37w+g7veEOq/VyhMONPlA+IT2yBGiAfIPltdffx27du3iOBrCV/IGfknJiw18\nXZyN0MTeFH2gztdFvjTvzJkz6Ny5MxITE+Hn54cVK1bA29sb8+bNQ4sWLbB48WL1Pgk1Uh7AqD73\nMxAR4QvGqlBaeqrGbf7+fjr5XqiLvI4gUF1LsLS0FKNHj0ZUVBT8/Pzg5uYGS0tLDiPkL/kS9dpu\n3bql8WMrdxb1ZTCF6AeakSNEg6jmDHkZfUunTntTVFPn63L9+nVERkbi999/R9u2bXHgwAFs3boV\nU6dOxbFjx/Dbb79h4cKFGngW6qNqhtLAoDMMDTtC32dzVc3mGBsbY8iQIZg2bRrCw8M5iEr3aLPY\nd25uLo4dO/ZsibQ1LaskvEIdOUI0jEbvSF30seOjCzXfuKCu10V5aZ6rqyuys7NhZ2eHCRMmYNiw\nYYiJiUGLFvxebKNqAKOq6i4qKx9AXwY16lLX58G4ceNgaGiIgIAAtR9T3wYUtbkEtzks9yW6jd9X\ne0LUzMzMDEVFRVyHQYiCrqdTV0Xbqe11hTpeF1VL8wAgIyMDlpaWePDgQZMeXxvkAxgREb4wMrJG\nebkM0dHrAeCF2/TpPKo9mzN37lzF98nJyQgPD9fYwJ++DCjWtSxXE0twtXksQhqLOnKkWdGXDzOi\nX6jjQ+pL1dK8ffv24fHjx0hKSsLQoUMhkUhgbm7OQXT1V9cAhr4NatTH0KFDcf36dRw6dEhjx6i9\nt3LVqlWYOHEizp8/DwA6k41Rm/uK9W0PM9FPtLSSNFsfffQRnJyc4OzsrPaEJJThimhCUFAQ3N3d\n4eTkhE2bNgEA4uPjIRQKIRQKMW/ePI4jJJpWO+lCXl4e5s2bh+joaPTs2RORkZGYOXMmhxHWn6qs\ngtquEbd48WLY2trC29sbY8eOxVdffaXVum7x8Qk4ceIccnLM4OLST2NL95T3VlpYWCAtLQ3l5eWQ\nyWQAgISEBLz33nsaObY6aXNfsb7tYSZ6ijHG6Vd1CIRoh5mZGWOMscTERBYQEMAYY+zBgwesa9eu\n7P79+2o/TnZ2NnNyclLb45LmLT8/nzHGWElJCXN0dGR//fUX69q1K3v48CGrrKxkfn5+7Mcff+Q4\nSkJ0g0QiYSKRiD19+pQVFRWxt956i61cuZL5+fmxS5cuMcYYmz9/Plu7dq1Gjp+Tk8NMTNox4BID\nGAMuMROTdiwnJ0etx8nOzma9evVS/Lxs2TK2dOlS9r///Y8tW7aMMcaYWCxmN27cUOtxNSUubicz\nMWnHzM1FzMSkHYuL26kXxyLkWZ+oQf0ompEjzVJKSgpCQ0MBAFZWVhg4cCAkEkmTH1eeSevm5Nn/\njQAAIABJREFUzZsAXtwTQUhTrFq1Ci4uLvDw8MDdu3exadMm+Pr6ol27djAwMMC4ceOQlJTEdZh6\nZ82aNbC3t8eECRO4DkUlbWbw0ycpKSkYOXIkjIyMYGpqiuHDhwOA1uq6abPuYu29lRUVFRgzZgwS\nEhJw/fp1GBgYoEePHmo/riZoM6ESJW8ifEcdOaIR8oLYuoKpYfkjZbcimnTq1CkcP34c58+fx8WL\nF+Hi4gKRSERLd7Vgw4YNOHr0KLZv3851KC+g6476jR49Gj///DMOHTqk0bpu2ly6p+o60b17dxga\nGmLx4sUICdGtDoo2l+Bqe7kvIQ1BHTmiEXxNKiL/MPPy8kJCQgKqqqqQm5uL06dPo0+fPo1+XOXs\nVgUFaSgpOYGIiOk0Qt7MLVq0CF9//bVaHqugoACWlpYwNjZGZmYmzp07hydPniApKQmPHj1CZWUl\n4uPj4ePjo5bjPX36FIMHD4ZYLMbu3bvV8pjqUJ9U6upMtz5t2jTcvn0bgYGBaN26NcRiMUQiESws\nLDjv2D2/7oyn604jeHp64uDBgygrK0NxcbEi2Yi26rpps/xIXZ/JISEhiI2NxZgxY9R+TEKI5lFH\njmjUqVOnFMtVACAyMhLbtm3jLB75h1lQUBCEQiGcnZ3h7++P5cuXw8rKqtGPq80lMqR5CgwMRHl5\nORwcHDB//nz069cPb7zxBr788ksMHDgQIpEI7u7uNd5vTSGVSiEQCCCVShEcHKyWx1SX+gwUqWsw\nacOGDXjjjTdw6tQp/PPPP5BKpYiOjka3bt0watQotRyjsZ5fdzY/u4WuOw3h5uaGESNGwNnZGUOH\nDoVQKETbtm0BaLaumzJtLN1TVfZgwYIFiu8rKyvRtWtXtR+XEKJ5VH6AaJxAIOBN/bbCwkIA1SPZ\n7777Lv7zn/+oZfSz5hKZ6nozlN2qeVq6dCm2bduGjh07onPnznB1dUWfPn2wYsUKeHt7Y968eWjR\nogUWL17coMdt2bIlDh8+rPJ3DV0WJZPJEBgYCA8PD5w5cwZubm4ICwtDVFQU8vLysGPHDkyYMAF5\neXkQi8XYs2cPbGxsGnQMTaqdSv3HH3/E77//joiICAgEAgwePFitx2PPk3MhLy8PEyZMQGJiIidL\nyIOCgnD37l2UlpYiPDwcT55cBVAOQAzgDbruNJC8U1NSUgJvb29FhkpN13VTxkX5kdzc3GZX5kET\n1qxZg40bNypm6ufMmcN1SKSZoRk5ohV8WmqpiT0l2lwiw3fqXNama6RSKXbt2oWMjAz89NNPkEgk\nEAgE2Lp1K6ZOnYpjx47ht99+w8KFC9VyvKYkubh58yY++ugjZGVlISsrCzt37kRKSgqWL1+OL774\nAtHR0fDy8oJUKuVVJw54MZV6YmIiJk2ahHXr1iE9PV1jx62qqkJoaCiioqJgZ2enseO8TExMDCQS\nCSQSCWJiYrB+/VoAlTA3B0xMzjbb605jTZkyBSKRCK6urggODoaLiwuGDh2KjRs3Yty4cVyHpxFN\n+QwcMGAA7t+/T0sxn9mwYQOOHDmCnj17ch0KaaaoI0c0qkWLFqiqqlL8XFxcjGXLlsHNzQ3Ozs44\ncOCAVuPR5F42ym71HJ867tp0+vRpBAUFwdjYGGZmZhgxYgQAwM7ODhMmTMCwYcMQExODFi2avhii\nqQMSNjY2sLe3BwA4ODjA398fAODk5ASZTMbrJCrdu3dXDBaIxWJkZ2ejoKAAnp6eAKCx7JL//e9/\n4ezszOlS09qZS4VCR5iamtJ1p5FiY2ORnp6Oq1ev4uOPP9ZaXTeuNPUzMDk5GZ06dVJ77VVdpLx/\ndtWqVbhy5Qp8fX3Rs2dPfPvtt1yHR5oJ6sgRjZA3Aq2trXH16lUwxvD48WOcOnUKM2fOxIULF3D8\n+HHMnTtXq3Fpei8bZbeq6datWxCLxTAyMsK1a9cUt/v6+kIqlXIYmfZlZGTA0tISDx48aPJjqWNA\nQjkduYGBgeJnAwMDVFRU8LozXjuVel5enkaPJ38tVq5cid9++w0ikQhisViRHENbVGUuLS0thUAg\noOuOGjSHpFVN/Qw0MzNr1qsulMn3z548eRKzZs1CVlYWjhw5gvPnz2PRokWorKzkOkTSDFBHjmiE\nvOHTuXNnjBkzBk+ePMF7770HkUiE3bt3K5KM/P3338jJydFaXNpM99zc/fHHHwgODsbWrVuxYsUK\nJCRUj2zfv38f9+/fh1gsfuFvgoKC4O7uDicnJ2zatAnDhg2Dqamp4vd79uzRaBa5pvL29sb+/ftR\nVlaGoqIiHDx4EACwd+9ePH78GElJSZgxY4Zir2ZjqWNA4mUzbsrFRvmodlwWFhawsLBASkoKgOpZ\nFnW6desW2rVrh6qqKmRkZCA9PR1SqRTDhg1T63FeRVXmUgAwMjJCRUWFVmPRR80haVVTPwMFAoHi\ni9Q0dOhQtGjRAu3bt0fHjh3VMmhHyKtQR45ohHJSkXfeeQdt2rTBL7/8ghEjRuC1115Deno60tPT\nYWVlhdLSUq3FRXvZtCMnJwejRo1CbGwsnJycEBwcjD179gAAdu3ahXfffVfl3ynv/1m9ejW2bdsG\nA4Oalyk+NyBEIhFCQkIgFAoxdOhQRUmL+fPnIzo6Gj179kRkZCRmzpzZpOOoY0BC+XWs/ZryvbGm\nKt6YmBh8+OGHKgcI1IXrwtuqMpcKBAJMmTIFQqGQtwXLdUVzGOhT12cgXwd5uFR7lQMNrhCtUB55\n5eKrOgSij+LidjITk3asbVsxA8Di4nay1atXs3//+9+MMcaOHz/OBAIBk8lkWo8tJyeHpaamspyc\nHK0fW99lZ2ezXr16sSFDhrDvv/9ecbuPjw/LyMhg/fv3Z5cvX37h73bs2MHefPNNZmJiwtq3b8/a\ntm3LXn/9dWZqaqq4T2JiIgsPD9fK8+A7+fvL3FzETEzasbi4nVyHpNeUr2cmJu3YzJmzmJ2dHfPz\n8+M6NKJGzeV91djPQDMzMyaTyZiTk5OGItMt3bp1Yw8fPmRRUVFs5cqVitsdHR05adsQ3fasT9Sg\nfhSVHyAaobzXoKRECMAUERHTkZ5+BuHh4XB2doabmxtnmd+4SPfcnBgbG2Pfvn0ICAiAqakpQkND\nERISgq+++gqFhYVwdHSscf/MzEysX78eNjY2uHnzJubMmYPjx4+joKCgxuyLNmdv1UkTqb5DQ0Pg\n7++n1sfV5ZTkmoz9xetZBr791g0HDuzD0KFD1Xqshsalq/8vvtLE+4qP6DNQPepatcDX1QxE/1BH\njmiEfK9BdaMHAIphZCRGYWEhzpw5w2lsRDtMTExw6NAhBAQEwMzMDO+88w5mzpypKESr7NixY8jK\nykJFRQU8PDxQWFiIO3fuKBobWVlZeOutt7Bv3z6Ym5tz8GwaLz4+ARER09GyZfWyrejo9WrLLKjO\nxpgm49Q0TcceHByMp09LAYwD8G8A91FVVYkZM2YgKSkJy5YtU9ux6kuX/198R52culEHpaZbt24B\nwAslZZQLsBOiSQLG8TpngUDAuI6BqF9ubi6srW1RUnIC8gLZJia+SEtLRnFxsV6PdJKGW7t2Le7e\nvYuMjAzIZDL07t0bjx8/xrVr1/Dll19i8eLFsLKygpubG4qLi7F582auQ66Xut4HMlkmr85/XYlT\nFW3EfuPGDQiFfVFS8guASQA2wMDAF0eO/Ao/Pz+1HKMhdPn/RXTXw4cP4ebmhj179mDu3Lk4ceIE\n1yHxBs2OE3UQCARgjDVotIRm5IhGyDdUR0T4wsjIGuXlMkREjIer6wAaQW5m6vMBN2jQIIwaNQrJ\nycno0KED8vPzUVRUBB8fHwwfPhwTJ07UctTq8eLM9PMseHz6sNeVOFXRRuw7duyAlZUZ7tzpB8YA\nY+Oh6NGjFywsLNTy+A2ly/8vopt+//13DB06FOPGjcPYsWM5mYXmK5odJ1yirJVEY5QLZKelJSM6\neode1+chL6pv0Wo7OzssWbIEAQEBcHZ2RkBAAO7duweBQIC8vDxOMwU2ha5kwdOVOFXRdOzy2m1Z\nWVl48OAeXF1dkJCwDVZWVmp5/MbQ5f8X0T3x8Qno08cHBQWvYdWqjVi4cBFGjhzJdVi80BxqDxJ+\no44c0Sh5gezi4mK9r89DamroB1xwcDB+++03bNq0CYcPH0bfvn2xdOn/4OLS75UdQb7SlXIXuhKn\nKpqOXbl228OHD3HlyhVYWlqq5bEbS5f/X0S3UEfl5ZpD7UHCb7S0kmhFzRHk6j0dNIKs3xq6/Kv2\n8pRvvvkSs2d/UiNTYESEL/z9/XSqwaorWfB0JU5VNBl7YGAgvvvuOzg4OKB3797o378/AO6TPujy\n/4voDlrG+3LUtiFco2QnRGvkDXX5njlaR67fGpKQQdV9jY190LJlFxQVPc/+ZW4uxtGjG+Hu7q7N\np0IIIc0SJdZ5NWrbEHVpTLIT6sgRraLMTs1LfT/gJBIJBg+eioKCNMVtZmYiPH16C2Vlp0ENCMIH\ndP0izRF1VF6Nrg1EHagjRwjhnfp8wNU16itfXkkNCMI1ykxHmjMuOyo2NjZIS0tDu3bttHpcQrSN\nOnKEEJ1V16gvjXQSrtHyMkK40717d1y4cIE6ckTvNaYjR1krCSG8oFyuQibLVMx2yDOfUoNZNw0Y\nMAAymQxOTk5ch9JolJmOEPWQyWSws7NDeHg4evfujXHjxuHIkSPw9PRE7969ceHCBTx69AhDhgyB\nk5MTJk+eDBrsJ6RulLWSEMIbHTp0oA6bnklOToZMJuM8y2NTUGY6QtTn5s2b2LNnD+zt7eHm5oad\nO3ciJSUFBw8exNKlS9G1a1d4eXnh//7v/3D48GFs3ryZ03gXL16M2NhYWFlZoXPnznB1dcWhQ4fQ\nt29fnDhxAgUFBYiOjoanpyencZLmiWbkCCGEaIyZmZl8uQjXoTQan+q2DRgwQOvHJESdbGxsYG9v\nDwBwcHCAv78/AMDR0RHZ2dk4ffo0xo8fDwB4++23Oa3beOHCBezbtw+XL1/G4cOHceHCBcWgVGVl\nJc6fP49vvvkGUVFRnMVImjeakSOEEKIxujwTp4wvdduSk5M5OS4h6mJsbKz43sDAQPGzgYEBKioq\nYGRkVOP+XA4CpaSkYOTIkTAyMoKRkRFGjBgBxhgEAgFGjx4NAHB1dYVMJuMsRtK8qWVGTiAQRAsE\nggcCgSBD6TZLgUDwm0AgyBIIBL8KBIK26jgWIYQQwgU+7Nc0MzNDUlISBg4ciFGjRqFnz5745JNP\nsGPHDvTp0wfOzs64ffs2Z/ER8iqv6ph5e3sjNjYWAPDzzz/j8ePH2girXpRjl3dADQ0NUVFRwVVI\npJlT19LKGABDat32CYCjjLHeAI4DmKemYxFCVND1hBJEv+nLzBzX5MtUMzIy8P333+Pq1avYvn07\nbty4gdTUVERERODbb7/lOkxC6qR8Lah9XRAIBFi4cCGSkpLg5OSE/fv3o2vXrtoOUcHT0xMHDx5E\nWVkZiouLcejQIZVLxXV56TjRbWpZWskYSxYIBNa1bh4JwOfZ91sBnER1544QoiHUWCZ8RQ0d9ZAv\n63J3d4eVlRUAoGfPnhgypHos1cnJCSdPnuQwQkLqZm1tjYwMxeKtGolM5L/Lzc3FkiVLeFFyxs3N\nDSNGjICzszM6duwIoVAIc3NzlR1QQrigyWQnVoyxBwDAGLsPwEqDxyKEACgvL8f48eNhb2+PMWPG\noLCwELa2trh+/ToAYOzYsYiOjuY4StKcyBs41NBRr1ftMyLPBQUFwd3dHU5OTvjhhx9w8OBBiEQi\niMVi2NraokePHlyHSJ6Jj0+AtbUtBg+eCmtrW8THJ3AdEubOnYvMzEz88ssvyM7OhpubG44fPw6x\nWAwAaN++PW7dusVxlKS50mayExqOJUTDsrKyEBMTAw8PD0RERGDTpk1Yt24dJk6ciJkzZ+Lx48eI\niIjgOkzSDOTm5uLixYuwsLBA165da4zCk/pbtGgRzMzMMGfOnBq30wxn/cXExMDCwgKlpaVwd3dH\nUlIS0tPTAQAhISHw9fXlOEICVF8zIiKmo6TkBEpKqst8RET4wt/fj9OZuSlTpuDq1asoKytDWFgY\nXFxckJuby3niI0IAzXbkHggEgo6MsQcCgaATgJy67qictnXgwIEYOHCgBsMiRH917doVHh4eAIDx\n48fj22+/xZw5c7Br1y58+OGHuHz5MscRkuYgPj4BkyZNxdOnJTA0NER8fIKiwDtpGlUznDTb+XKr\nVq3C/v37AQB3797F9evX0adPH3z11Vdo3bo1pk6dynGEBACys7PRsmW3Z504ABDCyMga2dnZnHaW\n5IlX5OLjExARMR0tW1bXl4yOXk/XN9IoJ0+ebPJSeIG6RvUEAkE3AAcZY07Pfl4G4BFjbJlAIPgv\nAEvG2At75AQCAaORRUKaTiaTYeDAgYqMdSdOnMDatWuRmJgIHx8fyGQyHD58GA4ODhxHSvRZbm4u\nrK1tUVJyAvLi2SYmvpDJMmnkup6WLl2Kbdu2oWPHjujcuTPs7Oywd+9eVFZWIisrC5WVlbh9+za6\ndOnCdai8d+rUKXz22Wc4cuQIjI2N4evri0WLFuHp06f49NNPcfr0abRs2ZLrMAl049qhCzES3fUs\nkU6DRubUVX4gDsAZAL0EAsEdgUAQDuBLAIMFAkEWgEHPfiaEaJBMJsP58+cBAHFxcRgwYAC++eYb\n2NvbIy4uDmFhYaisrOQ4StIYdWUlXb16NUpLSzmISDX5qHp1IwdQHlUnryaVSrFr1y5kZGTgp59+\ngkQiQZs2bXD48GGUlZUhKCgIwcHBik5cbm4uJBIJcnNztRpnXl4ePDw84OrqipSUFK0euyEKCgpg\naWkJY2NjZGZm4ty5c8jOzsaMGTOwe/duvejE1d4DGBMTg9mzZyt+v2nTJsydO5fDCJ97WXblDh06\nIDp6PUxMfGFuLoaJiS+io9fzqoNE1zfCN2rpyDHGxjLG3mCMGTPGujLGYhhj+Ywxf8ZYb8ZYAGOM\nP4VACNFTtra2WLduHezt7VFQUIBBgwYhOjoaX3/9NTw9PeHj44MlS5ZwHaZOiI2NRd++fSEWizFt\n2jQcOHCA8wQJqpbQrVq1Cv/884/WY6lLt27Vy40A+Z64DJSXy9CtWzfugtIhp0+fRlBQEIyNjWFm\nZoYRI0YAAF5//XVs3rwZf/zxhyJhEZeJIY4ePQqhUIi0tDR4enpq7bgNFRgYiPLycjg4OGD+/Pno\n168fZDIZHj16hFGjRkEkEmHYsGFch9kkMTExkEgkkEgkWLNmDfz9/XHw4EHFoF1MTAwmTZrEcZTP\nvWwpcGhoCGSyTBw9uhEyWSbvlizS9Y3wjTaTnRBCNMja2hpXr16tcVtubi62bNmCJ0+eoHXr1lix\nYgVH0emWzMxMJCQk4MyZMzA0NMSHH36IwsJCzhMkyLOSSqVSODg4wNvbG3///Td8fX3x2muv4dix\nY1qPqTb5qHpEhC+MjKxRXi7j3ai6Lrp//z4mT56MgwcPonXr1lpPDLFt2zasXLkSBgYG6NKlCy5d\nuoSSkhJcuHABZ8+erZFFk09atmyJw4cP17gtNzcXgYGBepOoovYewHv37mHQoEE4dOgQbG1tUVFR\nwasl9crXMUdHR2zbtg2XL1/GrFmz8OTJE7Rq1QrHjh1DmzZtGvzYMpkMgYGB8PDwwJkzZ+Dm5oaw\nsDBERUUhLy8PO3bswOHDh3Hnzh3cunULf/75J2bOnInIyMh6PT5d3wjvMMY4/aoOgRCibnFxO5mJ\nSTvWtq2YmZi0Y3FxO7kOSWesXbuWvfnmm0wkEjEXFxdma2vLFi1axBhjbNmyZSwsLEzrMWVnZzOB\nQMDOnj3LGGMsIiKCrVixgtnY2LBHjx5pPZ5XycnJYampqSwnJ4frUHSKVCplzs7OrLS0lBUWFrK3\n3nqLLV++nA0YMIAdOHBAcb/U1FTWtq2YAUzxZW4uYqmpqWqP6cqVK6x3796K8yw/P59t2bKFRUZG\nqv1YmqZv18WTJ08yLy8vVlpayhhjbODAgezUqVPs/PnzbOTIkey///0v27BhA8dRPqfqOvbVV1+x\n7t27s7S0NMYYY0VFRayysrLRj29kZMSuXLnCGGPM1dWVTZo0iTHG2IEDB9ioUaNYVFQU8/T0ZOXl\n5SwvL4+1b9+eVVRUNOg4dH0jmvCsT9SgfpQm68gRQjiiPFpfUJCGkpITiIiYrvV9NLqKMYaJEydC\nKpUiPT0d165dw4IFC3D06FHs2bMHGzdu5CQu5ayk48aNQ3JysiJevunQoQPc3d1ppLqBRCIRQkJC\nIBQKMXToUPTp0wclJSVIS0vDwoULFct7TUxMtLbE6/jx4wgODoalpSUAwMLCQu3H0AZtXxeHDRuG\nwsJCjTy2nKo9gADQp08f/Pnnn4iPj0doaKhGY2io2texX3/9FW+88YaiLpupqSkMDBrfPLWxsYG9\nvT0AwMHBAf7+/gAAR0dHZGdnQyAQYOjQoWjRogXat2+Pjh074sGDBw06Bl3fCF9QR44QPUQbsmsa\nMGBAg+4/aNAgJCYmKhp4+fn5kMlknCdIkO8tkclkmDRpEqWd11Pz5s1DVlYWkpKSsGPHDnz22Wf4\n559/FAML8iVpfE8MwTfavi4eOnQI5ubmGnlsOVV7AOXGjBkDT09PtG3bVqMxNFTt65a6XyPlZb4G\nBgaKnw0MDFBRUaHyPvLbCdE1tEeOED1Uc0N29f6Z5rwhWz5zJVdZWQlDQ8M6729nZ4clS5YgICAA\njDEYGRlh+PDhigQJjDG8+eabOHTokKZDr0GelbRTp04oKCiAl5cXbt68icLCQrRr106rsRDtqav4\ncGhoCPz9/TRemNjPzw+jR4/G7Nmz0a5dO+Tn52vkOJr2quviokWLYGpqitdeew1DhgxBp06d6v3Y\nQUFBuHv3LkpLSzFz5kx88MEHsLGxQVpamkbfm6r2AMolJye/UEieD+TXsb59+yIuLg79+vXDxo0b\nceHCBbi5uaG4uBitW7du9KwcH1coEKIpNCNHiB7ShTTO2mRmZoakpCR4e3tj5MiR9dr4HxwcjPT0\ndFy8eBESiQTTpk3DTz/9hF9//RXp6ela78QBz7OS+vv7o6KiAqmpqXjw4AGEQiF8fX1hY2OD+fPn\nQyQSwd3dHVKpFEOGDMFbb73F2XLQpvr666/h5OQEoVCINWvWAKius9a7d294e3tj7Nix+PrrrzmO\nUnNelZlSG0u87O3t8emnn8LHxwcikQhz587Vydng+l4Xt2zZgr/++qtBj62cOXL16tV49OgRZ6/R\nzZs3YW1tjRYtWnCSlOlVlLMrP378GJGRkUhISEBkZCRcXFwQEBDQpJIqyq977f+Bqv+JLp7LhCg0\ndFOdur9AyU4I0RjakF3NzMyMnTx5kpmamjKZTNbgv29MgoTs7Gzm6OjYmHDr9dh1JT7ZuHEjY4yx\n2bNnM6FQyJ48ecJyc3NZx44dNRKLJqWlpTGhUMhKSkpYcXExc3R0VNwmTwbSs2dPtnLlSq5D1Yic\nnBxmYtKOAZeeJTS5xExM2nH6ftaHa4ryc1iyZAnr1asX8/LyYqGhoWzFihXM1NSU2draMpFIpEgi\n8ioLFy5kzs7OzNnZmVlYWLBz584xGxsb9vDhQw0/m5p0LZmLPpxPhKgLKNkJaQ7qKii6cOFCHD9+\nnIOI+Is2ZFdjjEEgEKBPnz7o2rVrg/62KQkSNDXS+/DhQ3Tq1ElRy0458cnw4cMBAE5OTvDw8EDr\n1q3x2muvoVWrVhpPvKBuycnJCAoKQqtWrdCmTRuMHj0aP/30k8o6a/qIb3tduaxbp07y6+Kff/75\nQvF1gUAAd3d3xMbGQiqV1quswqlTp3D8+HGcP38eFy9ehIuLS5NmlBpL15JccXU+5ebmQiKR8PZ1\nIaQhqCNHdJKqBvKiRYvg5+fHQTREVzSmLlFTGtMVFRWYMmUKHB0dERgYiNLSUvj6+mLWrFkQiUQQ\nCoWQSCQNiic+PgH9+/vh/v2HNRo/8veE8sZ+5UaoQCDQ+Q398g55c8Gn4sO61kmoj9rF10eOHNmo\n/VV1ZY7UNr51/F+Gq/NJXwYjCJGjjhzRSaoayOHh4di7dy/XoREea0wjrSmN6evXryMyMhK///47\nLCwssGfPHgBASUkJ0tPTsW7dOkyaNKnescgbP2VlO8FYOUpKViMiYjo2b94MLy+vBj83vvPy8sL+\n/ftRWlqKJ0+eYP/+/Rg6dCj27duHsrIyFBUV4eDBg1yHqTF82uuqS52ExmrM9QF4MXNk//79IRAI\nNDLwoLxndPXq1QCA2NhY9O3bFxERESguvgLg0rN78zfJFRfnkz4ORhBCHTmik1Q1kJvTSD1pGPm5\n0ZhzpCmN6e7duyuWAYvFYkUNI3ldJy8vLxQVFdV7yePzxo8dAFsAv6Gs7B/cu3cPU6dOfenf6uL7\nQyQSISwsDO7u7ujXrx8mT56sss6aPgsNDYFMlomjRzdCJstEaGiIyvvVteTcxsYGjx49anIcfJod\nVBdvb2/s37+/xqCAQCCAmZlZg5YhyzNHXrlyBXv37sWxY8fg6emJ4uJitabWl0ql2Lp1KyQSCc6e\nPYtNmzbh3LlzSEhIwJkzZ5CRkQE/Px8YGXly3vF/FS7Op+YwGEGaHyo/QHSSqgYyIapkZWXB1NQU\ndnZ2OHDgQKMeo7Fp3pWXNhoaGqKkpARAzU5VQ0btnzd+CgBcBZABY+OfEB8fDxMTE9y6dUtx37ff\nfhv29vbIzc1Fhw4davxOl8yaNQuzZs2qcdv8+fMxf/58ANVLqvVdhw4d6nXOaTIjn3xAIyLCF0ZG\n1igvl/G2k1BfyoMCHTt2VAwKhIWFYerUqWjdujXOnj1br31yQM0yEd7e3pg8efJLy5w0lPKeUQAY\nPXo0JBIJ0tLS4O7uDsYYSktLMXv2h3j33Xc1WpKiqbg4n6gsD9FH1JEjOqmuBjIhytbspJ+wAAAg\nAElEQVSv/w4zZnwIY+M3YW1ti+jo9XXOaLxKfRvTylQt1WKMISEhAT4+PkhOToaFhQXMzMzqHUN9\nGj/x8QmIiJiOli2rGy5Ned58pNxgJs+Vl5dj/PjxioLhW7duBWMMa9aswcGDB1FRUYHdu3ejV69e\njXp8bdWt06Z58+Zh3rx5L9weFBTUoMfh4j0nv76EhYVh6dKlGj2WJmj7fNLHwQhCaGkl0Ul1NZAJ\nkcvNzcV//vMpGEtHaekdTvZD1K5nJP9q1aoVxGIxpk+v3t/WEK9aaqfv+0BqJyvo1cuWl0WPuZCV\nlYUZM2bg6tWrMDc3x/r16wFUN2DT0tIwdepULF++vEnH0OdMuI3NZqit95yqPaNubm5ITExUHCs/\nPx937txR63E1SdvnU32XKhOiK6gjR3RSXQ1kolvqOxPVGFzvh7C2tkZGRobi5zlz5mDBggVgjClm\nTTIyMuDq6trgx35Z44fr561J+t5JbaquXbvCw8MDwPOSFAKBAKNHjwYAuLq6QiaTcRkibzUlm6G2\n3nOq9oz269cPS5YsQUBAAJydnREQEID79++r9bj6Rp8HI0jzQ0sric5R1UAmukmTnW8+7ofIzc1F\nUVER8vPzNXYMPj5vdZE3mEtKXmww63ujTCaTITAwEB4eHjhz5gzc3NwQFhaGqKgo5OXlYfny5fjr\nr7/w8OFDtG/fHowxHDlyBO3bt1csRTc0NNT5EhRr1qzBxo0bIRaLIRKJ1HL9Vx4gqD63MhAR4Qt/\nf796nVfafM+p2jMaHByM4OBgtR+LEMJ/NCNHdBYfi3rWlTmOPE+RLRaLMW3aNFRVVYExhjlz5sDR\n0RGDBw/Gw4cP1XY8PqVuB56P+N+4wTB8+BiN1S/i2/NWJ33MnNgQN2/exEcffYSsrCxkZWVh586d\nSElJwYoVK7Bu3TpUVFQo9kotX74cPXv2VGuyDT7YsGEDjhw5gp49e6rtMZs6o8b1e46Pn4WEEC1h\njHH6VR0CIQ0TF7eTmZi0Y23bipmJSTsWF7eT65AYY4xlZ2czJycnrsPgnWvXrrHhw4eziooKxhhj\n06dPZ9u2bWMCgYDFx8czxhj7/PPP2YwZM9R+7JycHJaamspycnLU/tgNicHEpB0DLjGAMeASMzFp\np9GY+PC8NUH+3jc3F/Hqva9p2dnZrFevXoqf33//fRYXF8cYY+zWrVvMwcGB9ezZk7Vr147Z2dmx\nLl26sL179zIbGxv28OFDxhhjFy5cYL6+vpzErw5Tp05lxsbGzMnJiVlYWLAJEyawfv36sV69erEf\nfvih0Y+rrvcnF+85vn4WEkIa7lmfqGH9qIb+gbq/qCNHGoqLRnF9ZWdnMzs7OzZ58mTm4ODAhgwZ\nwkpLS9kPP/zA3N3dmYuLC3v33XdZSUkJ16Fq1dq1a9mbb77JRCIRc3FxYba2tmzRokWsRYsWrLKy\nkjFW3RgViUQcR6oZqamprG1b8bPztfrL3FzEUlNTuQ5NJ+lrJ/Vlag8ShYWFsT179rzwu7fffpsd\nP36c9ejRg1VVVXESqybJO6ZRUVHMxcWFlZWVsby8PNalSxd27969Rj+uLg4Q8PmzkBDScI3pyNHS\nSqJz+J7MQblYedu2bbFnzx688847SE1NRXp6OmxtbREdHc11mFrFGMPEiRMhlUqRnp6Oa9euYcGC\nBS/cT18T1jT3JYHq1lyTFbB6ZOaNiIjA+PHjMWbMGAgEAr1edjdy5Ei0bNkS7du3h5+fH1JTUxv9\nWLqYzZDvn4WEEM2jjhzROXxvFCsXK3d1dUV2djYuX74Mb29vCIVCxMXF4cqVKxxHqV2DBg1SmSK7\nsrISiYmJAKr30A0YMIDLMDWG6z00RD/UztaryogRI/DkyROEhYU1KROjLlB+DRhjTR4I0rUBAr5/\nFhJCNI+yVhKdw/einqqKlYeFheHAgQOKIr2nTp3iMELts7OzU6TIrqqqQsuWLbF27VqYmpoiNTUV\nixcvRseOHZGQoF8NTWX6VEx527ZtWLlyJQwMDCAUCmFgYAATExOkp6cjNzcXmzZtwpYtW5CamgoP\nD48G18ojL6qdrVf5NbW2tsaxY8cgkUjw+PFjODs7w9LSskmZGPlKeVbyxx9/xLx581BUVIRTp05h\n2bJlHEamfXz/LCSEaB515IhO4nOjWNXyp+LiYnTq1Anl5eWIjY1F586dOYiMW6pSZBcWFgIAVqxY\nwUVIWtehQwdenauNcfXqVXzxxRc4e/YsLC0t8fjxY8yePRuPHz/G2bNnceDAAYwYMQLnzp2Dvb09\n3NzckJGRAaFQ+OoH15ABAwYgLi4OKSkpCA0N5SwOTYmPT0BExHQw1gplZfexcGGU3pZqUJ51EwqF\nGDhwIB4+fIgFCxagU6dOHEbGDT5/FhJCNI86ckRn8bVRXHt5j0AgwOLFi9GnTx9YWVmhb9++KCoq\n4ig6fsnNzaUGiI45fvw4goODYWlpCQCwsLAAAAwfPhwA4OTkhNdffx329vYAAAcHB2RnZ3PakUtO\nTsbJkycRFxendx055Rpo8hpmy5b5YsyYd/WynuCtW7cAAAsXLuQ4Ev7g62chIUTzqCNHiBrVXv40\nd+5cxff/+te/uAiJt+SzCC1bVu/ziI5erxMJBohq8iXFBgYGNZYXGxgYcF6E2tTUFEKhEFevXoVY\nLMbEiRMxc+ZMTmNSl7pm3oqLi/V62R0NAhFCCCU7IURtXlYMXB8yx6kq6G1mZqb4/Z49exAeHl6v\nx1KeRSgoSENJyQlEREzX6denufDz88Pu3bvx6NEjANWJa2qrT3ZFbTIwMMCXX34Jb29vSKVSvenE\nAS9PeKGLmRjrQ9+TuBBCSH1RR44QNVKVNU0fGh2ZmZlISEjAmTNnIJVKYWBggNjYWJXLSOuD0mbr\nLnt7e3z66afw8fGBSCTC3LlzX3oe8KGkBN86lur0qoyoupaJ8VVoEIgQQp6jpZWEqFFFRQWmTJmC\nM2fOoHPnzvjhhx8wfvx4VFXtQElJCIAkjBune5njjh07BqlUCnd3dzDGUFpaio4dOzb68WrOIujP\n/p3mYsKECZgwYYLK3ylnUOzWrRtlrNSC5pTwQl+TuBBCSGPQjBwhaqRcDNzCwgJbtmyBQNAKwFvP\n7uEAoAXnM0/yJZH37t3DmDFjXnn/+hT0Li0trffxqa6a/uLjDLRAIICZmZkiS6o+0reZt7roW+20\nly3JJ4SQV6GOHCFqpFwMXCwWo6SkBIyVArj+7B5XAFRw3uiQL3d7/fXXsWvXrlfev66C3p06dUJW\nVhaqqqqwb9++BsWgr/t3mjO+LnsTCAQQCoUwNDSESCTC6tWrOY2HNJ4+DgLxYfkxIUQ3UUeOEDWq\nXQzc2NgYDg52MDaeAnNzMVq1GonXXnuNN42O+o4GKxf0dnZ2RkBAAO7fv48vv/wSQ4cOxYABA/DG\nG280+PjNZRahueDj3sesrCy0adMGjx49wrFjx5Cenq5XyU60YcCAATh16pSixATX9G0QSL4k39HR\nEYGBgfj7778hEokgFoshEonQokUL/Pnnn1yHSQjhIdojR4gaqUqq4OHhgQkTJmDgwIH49ddfebdn\nqL6jwaoKevfp0wejR4/WRFhEB/Ft7+P69d9hxowPYWz8JqytbV9Z4kImk2HYsGG4fPmyFqPkv+Tk\nZJw6dYpXM0cvq522aNEimJmZYc6cOVqOqnGuX7+OhIQEfP/99wgJCcHJkyeRnp4OAFi/fj1Onz6N\nLl26cBwlIYSPaEaOEDVSlb3vP//5D3bs2IGpU6fi6dOnHEWmGfpQVoGoD5+WveXm5uI///kUjKWj\ntPROvZd58qmzwhdmZmYQCAQoKCjAsGHDYGtri+nTp3Mdlt5QXpLv6uqqmMFOSUnBpk2bEB0dzWF0\nhG/WrFkDe3v7OhNOkeaFOnKEqImqYuALFiyApaUlNm3ahF9++QWff/45bt26xWGU6sPHpBaEe3xZ\n9tbYZZ61l7mVlZUhLS0NLi4uEIlE+Pjjj3U6OUVjkmsIBAIwxpCamop169bh2rVruHHjBvbu3au2\nY9SXTCaDnZ0dwsPD0bt3b4wbNw5HjhyBp6cnevfuDYlEUuP+P/zwA4YOHYqysjKNxKMOtZfkV1RU\n4P79+5g8eTJ2796N1q1bcxgd4ZsNGzbg6NGj2L59O9ehEB6gjhwhGsTXzk5T62rxNalFU1D2OPXh\nw97HxmY3rJ15NjExEZMmTcK6desUy910fdauofEzxiAQCNC3b19YW1tDIBAgNDQUycnJajtGQ9y8\neRMfffQRsrKykJWVhZ07dyIlJQUrVqzAF198oeh4rlu3DocPH8b+/ftrdJb4pvb1+OnTpwgODsay\nZcvQo0cPjqJSLTY2Fn379oVYLMa0adPAGEN8fDyEQiGEQiHmzZvHdYh6bdq0abh16xb+3//7f1ix\nYgUCAgLg5OSEyZMno1u3bnj06BHXIRIto44cIRrC585OUws28zGphTroegOdPNfYZZ61M89mZ2ej\noKAAnp6eAKAXy5nKy8sxfvx42NvbY8yYMUhKSsI777wDAPjxxx/RunVrVFRUoKys7KUdiZe9X2rP\nbJaWlsLX1xeffPIJ+vbtC1tbW6SkpDQqfhsbG9jb2wMAHBwc4O/vDwBwdHRUXIO2bduGX375BYmJ\niTAyMmrUcbTl/7N352FNHesfwL8BAbmyW9yqLNoKsicssiuKiK3iiog7UlFavVq9Vu2vLpRqvZXW\npbW21hUFpKJQ0bZWoSCClV1oUUQxUesCiqC4IOD8/uAmDQiy5eSckPk8D89DCGTehOScMzPvvNP0\ndTx//jxycnKwbt06SdGTu3fvshTdPy5fvozY2FhkZGQgNzcXKioqOHToEFavXo2UlBTk5+cjMzMT\nx48fZzvULmvnzp1488038fvvv0MkEmHkyJEoLCzElClTaEEcJUU7chTFEC53dsT7ad2/fx8GBgbt\n/vuutpeTWNML3GfPnsHU1BQrV66EjY0NnJ2du0xqrDLoSJpn0zS3+/fvMxkiK4qLi7Fo0SIUFRVB\nR0cHWVlZuHjxIoCGwibW1tbIysrChQsX4Ozs3Ci1UiQS4eXLl4iNjYW7u3uLbTSd2Tx69CgAoL6+\nHhcuXMCWLVuwfv36DsUv/T9SUVGR3FZRUUFdXR0AwNraGkKhkPMXt01T8mfPno1NmzZBJBJJ9u3M\nzc1Fnz59WIyyQVJSEnJzc+Ho6Ag+n4/k5GRs3boVw4cPh4GBAVRUVDBjxgycPXuW7VC7PEIIzp07\nh2nTpgEARo8eDX19fZajothAO3IUxRAudnaki5Pk5ORgxowZWLp0absfh0tFLWSp6QXut99+CwDQ\n09NDQUEBPvjgA7mVrt++fTssLS0xa9YsfPXVV3Jpsytqb5pn0zQ3PT096OnpSWaPoqKiZB6jvBkZ\nGcHZ2RkAMGPGDJw7dw6DBg3C5cuXkZmZiWXLliE1NRVpaWnw8PAA0DBr5OTkhEWLFsHS0hKDBg3C\nxIkTW2yjuZlNHo8nqXJrb28PkUjUofjbkhouEAjw/fffw8/PD3fu3OlQO/LG1VR8MUII5syZI+lg\nXrp0CevWret0qj7Vfs3NhtP/g3KiHTmKYgjXOjtNLxKuXLmKy5cvY/z48R16PK4UtZCl5i5wxeuB\nACAwMBDnz5+XSyw7d+7E6dOn8dZbb8mlPapBc5Vn9+3bhw8++ABaWlosRSVbzT1HT09P/PLLL1BX\nV4e3tzfOnTuH9PR0WFtbo2fPnvD09ERKSgoSExNx6dIl7Nix47VtNFfAQ/rn0j/rTPzNPRcxV1dX\nREREYOzYsZxfO8TlVHyxkSNHIi4uThLTw4cPYWdnh7Nnz6KiogL19fWIiYnBsGHDWI60axN32Nzd\n3REb29DZ/+2331BZWclmWBRL6D5yFMWgwMAAeHuPgFAohImJCWudOOmLhGfPGvb3Cg72grf3iE7F\n9Lq9nBRRSxeF0j9XUWF+/Cs0NBTXr1+Hr68vbt68iTVr1gBoqMCXkJCAY8eOcbp4g6JqrvKsWH5+\nPoCGojgnT56Ue2yyJBKJcOHCBQwdOhTR0dHw8PCAQCDA7NmzMXfuXPTs2RMPHjzA7du3MW/ePKxY\nsQLl5eXtOo61ZXagIzMITf9H0vtyNr0PAHx8fODj49PuduRNnIrfcHwGpFPxuXKMHTJkCD777DP4\n+Pjg5cuXUFdXx44dO7Bp0yYMHz4cADB27FjObBzfVYnPR2vXrkVgYCAOHToEFxcX9OnTB9ra2ixH\nR8kbnZGjKIZxoYIfl9frcYn4AheA5AKXECIZ9Tx8+DBcXFwYj2Pnzp3o168fUlJSsHTpUoWqwNcV\niVOStbS0cOHCBVy/fh3+/v4YMmSIQhY/MTc3x44dO2BhYYHKykqEhoZi6NChKCsrg6enJwDAxsYG\nAoEAxcXF0Nfv2e6Uv6azZs2lgjFRXEhR97bkYip+c/z9/ZGXl4eLFy8iKysLTk5OCAgIQEFBAQoK\nCrBx40a2Q+zySktLYWBgAF1dXZw6dQoFBQUICgpC7969OV/Yh2IAIYTVr4YQKIpiUllZGdHUNCDA\nRQIQAlwkmpoGpKysjO3QOEMoFJIhQ4aQWbNmkSFDhhB/f3/y7NkzYmJiQlatWkVsbGyIk5MTuXbt\nmlziMTU1JQ8ePCDr168nNjY2ZOzYsaSurk4Sq5WV1St/M23aNGJra0u2bt0qlxiVQXT0YaKpaUB0\ndQUEAFmzZi3R09Mjt2/fJi9fviQuLi4kPT2d7TAZo0jHDun/laamAYmOPsx2SO0ijl9Hh69Q8ZeV\nlZHMzExOvifkLSUlhWRkZMilrZKSEmJtbU3efvttIhAISHZ2tlzapZjzvz5Ru/pRNLWSopSAeL1e\ncLAX1NSMUVsr6hLFSWTJ2NgYRUVFjX5WXl6OmpoaBAUF4fPPP2cpsobZkfz8fNy8eVMyQt90NuPu\n3bvIzs5GSUkJCxF2Ta+mJPfAf/+7Ba6u9ujbty8AwM7ODkKhEK6uruwGKyNNUyhlmfLX3vTM9j42\nE+nj8sSVVPz2iImJRXDw+1BXb5hR3LPn2y6xXrqjUlJSoKWlJZfMjaysHFy9+jfU1U1w65YQV65c\nhb29PePtUtxCUyspSkl0xeIkTBIXh7l79wFsbYfKvYIckVo/xOfzW6zAV1paCoFAgIEDB+L27dsQ\nCAQd3p+LauzVlGQVdOvWB/X19ZLf6UzRDq5prmqirFL+mK7I2FXSx7mQit9WilCgRZZEIhGGDBmC\noKAgmJmZYcaMGTh9+jTc3NxgZmaGrKwsfPfdd9i6dSvjx2Fle+2pltGOHEUpEUW6SGCT9EmSkBo8\nf54q95Nk0xk36Qp84upkV65cwZQpUxAZGYlLly5h0KBByM3NlWxe3VkikUhSQl4ZvdqJeYm6urvQ\n1NRkMSpmtHRhCKDT1XflcdGpKGvMuhJ5dZ7d3d2RmprKiSIq165dw4oVK1BcXIzi4mIcPnwY6enp\niIiIwMaNGxEaGooPP/xQpsfh5nSVgQuq82hHjqIohSGvilxcOEmKF7S///778PDwQHl5OXx8fJCT\nkwM9PT2UlZVhwoQJiI6OhpWVFWNxMFGQQlE03UIEeIqVK5c1KjbTVV6f173nOzubL4/PE9e2e1EG\n8uo8nzt3DgA3PmumpqawsLAAAFhaWsLb2xsAYGVlJdfzAx24oMRoR46iKIUhrxM5V06Sr0tH09XV\nhZGREdLS0uQSiziFs0ePHo1KvHt4eKCwsFAuMbBB3ImJi/svjIyM8Omn67Fnzx5JZcTt27dj9uzZ\njLQtzxnR1t7znZnNl9fniaaPy5e8Os/a2trg8Xh4/Pgx69VipQdxVFRUJLdVVFTkmmJNBy4oMdqR\noyiKkyZOnAhHR0dYW1tj9+7dABrWjX3yySews7ODq6srY6mOXDhJtpaOpqGhgfj4eERGRuLw4cMA\nOrYvV1uIUzgPHDiA7777Dvv27QMAlJSUoKampsunX9bV1WHRokVYuXIl42u9mpLX4AWT73l5fp5o\n+rj8hIeHIyxsHWxtB8PZ2RBLly7A1q1fSe4XiUSwsbF5zSO0DY/HAyEE+fn52L59O4qKinDt2jVk\nZGR0+rHbq7VjrLa2Nh49esRI203X6J04cRwHDnwHE5N69Ounj7feGoiwsDDMnTsXnp6eMDU1xbFj\nx7BixQrY2NjgnXfeabS+l+oaaEeOoijWuLu74+7du5g6deor9+3btw9ZWVnIysrCtm3bUFFRgSdP\nnsDV1RX5+fnw8PDADz/8wFhsbI/utyUdTVNTEydOnMCWLVuQmJjIyEW/dAqntbU1pkyZgpMnT6K+\nvh579+7F3LlzZd4m1/Tt2xfFxcXw9/eXe4GB2tpazJw5ExYWFpg6dSrS0tLA5/MhEAhgY2MDVVVV\nmbXF5Hue7c8TJVvZ2dmIj49HYWEhTp8+jevXr+ONN95AbW0tRCIRACA2NhbTpk3rdFuEEPB4PDg5\nOaFv377g8XiSarHy1nR/xKb3jRs3DvHx8YwVO2m6Ru/XX3/FxYsXsWXLFskefqWlpUhJScFPP/2E\nmTNnYtSoUSgoKED37t1x8uRJmcdEsYtuP0C1yt3dXZKjTlGyJH5f/fjjj6/ct3XrViQkJAAAbt26\nhZKSEmhoaOCdd94BANjb2+PMmTOMxmdoaMjayH7jdLSGcuridDRDQ0NJeqOurq5kE/NFixbJPA7p\nFE5zc3Noampi1KhRSEhIwJEjR5CTkyPzNrlKlqX426q4uBj79u2Ds7MzgoODkZWVhby8PADARx99\nJPk8yAqT73k2P0+UbKWnp2P8+PFQU1ODmpqapBDJ1KlTERsbi48++gixsbHNHts7SjqtkY1qscbG\nxo3Syvfu3QugIXuirKwMSUlJMDQ0xMWLFxmLobU1enw+H2PGjIGKigqsra1BCIGPjw8AwNramhZD\n6YLojBzVKtqJo5iira3d7Dqg1NRUJCcn48KFC8jPz4ednR2eP38ONTU1ye909kTeWinp7Oxs3L9/\nHz4+PrC2tsb8+fNhYmKCioqKDrfZHm1JRysvL5es1WKKdApnTEwMACA4OBj//ve/4eTkBF1dXcba\n5ho21k4aGRnB2dkZADBz5kzJ8Tg2NhZ5eXms7m9IUU2JO3IlJSVQUVHBoEGDZPbYTKWOd4a8U63b\nskZP/DMej9fonCnvdXyUfNCOHNUqcaXAOXPm4Pjx45Kfz5w5E4mJiWyFRXUBPB5P8iWtqqoK+vr6\n0NDQwOXLl/HHH38AkP2J/HWlpDds2ICwsDCMHDkShYWFmDJlCm7evCnT9lvzunQ0eV5AiFM4t27d\nihMnTkAgEEBHRwdBQUGMtclFbKydbC5966+//sKnn36K2NhYTlTyo5SPm5sbEhMTUVNTg+rqapw4\ncQIAMHDgQKiqqiI8PBwBAbJJnxW/x1+X1sgGNvZya+85kIudX0q2aGol1SrxATM4OBhbtmyBn58f\nHj16hPPnzyMyMpLl6KiuoOnJxtfXF9999x0sLS1hZmYGV1dXALI/ebeWpiISiRAfHw8AGD16NPT1\n9WXafls0l44mfQHRkOZXgOBgL3h7j5Bpp0I6lUicwlleXo6ff/4ZdXV1GDVqlMzaUhSBgQHw9h4B\noVAoSXNlkkgkwoULFzB06FBER0fDzc0NgYGBiIyMhIGBAaNtU1RLHBwc4OfnB1tbW/Tu3Rs2NjaS\n2fmAgAB89NFH+OyzzzrdzoMHD9CzZ094enpiyJAhyMrKgomJCbZv397px+4sNlKtW1uj97rfp7om\n2pGj2szT0xMffPABHjx4gLi4OEyePBkqKnRSt6sTiUQYO3YsoyXmm55s1NXV8fPPP7/ye9euXZOc\nyCdPnozJkyd3qt3W0lSk01IA7oxusnEBATTMAs6ZE4y6ulqoqakjJiZWKYtWyHOtl7m5OXbs2IGg\noCBYWlrC0NAQN27cwPz58yVFIHJzc+USC0VJW758OdauXYtnz57B09MT9vb2KC8vh6enJ+7evdvp\nz8idO3cwfPhwrFixAjExsQgOfh/q6g3pzXv2fMv6sed165iZ0NIavebuE5M+Z65bt46RuCh20Y4c\n1S6zZ8/GwYMHcfjwYezfv5/tcCg5YXpUry0dJCZO5K216+bmJlm4/9tvv6GysrJT7cmKvC8ggH9m\nAWtrMwDY4MULZmYBqX8YGxujqKjolZ+ztYdWV6StrY3Hjx+zHYZCCgkJQVFREWpqajB37lxculQM\nV9eRMjtGi6vFlpeXw9jYnPEMhPYSp1oHB3tBTc0YtbUiTu3lxsXOLyV7tCNHtUr6YnfOnDmSEsDm\n5uYsRkXJk7gEem5uLqysrBAZGYmUlBR8+OGH6NGjB9zc3FBaWtruNZPNrX1oDlOphK2lqaxbtw7T\npk3DoUOH4OLigj59+kjWjLKJjQsItmYBqcbKy8vlltapDGjqWcdFRUVJvmeys8XlY4+8U63bSl7p\n9xT7aF4c1SrpE12vXr0klf4o5VFcXIxFixahqKgIOjo6+PLLLxESEoKTJ08iOzsbd+/ebdcFUXl5\nOU6fPg09PT0YGRk1mxIirS17qrVXc2kqkyZNanSfjo4OTp06hYKCAgQFBaF3796vpFuyRd77crFR\nsZFqTN4V8pRJamoqhg0bhrFjx8Lc3Bzvv/8+2yEpFCaO0WJcP/ZwcRN6Jv8fFLfQjhzVqkePHkm+\nf/r0Ka5evYrAwEAWI6LkTboE+owZM5CdnY2BAwdi4MCBABoqmLZVTEwsjIwGw9d3HO7cud+mi1G2\nTuR5eXmwtLSElZUVlixZwugG5B0hzwsINio2Uv9go0KessnKysKOHTtw6dIlXL16FceOHWM7pHZp\nbisXADhw4ADu3r3LaNtMHqPpsaf9uN75pWSHduSoNjt69CjeeustzJs3jxPpZZT8NLc9QEeIL0af\nP0/Fy5fPUVt7vk0Xo2ycyGNiYuHl9Q7u3u2B0tI7WLp0Gezt7RlrTxHIexaQ+v3BfpQAACAASURB\nVAcdYWeek5MTjI2NwePxEBgYqJB7qDaXGbF//378/fffjLbL9DGaHnvah3Z+lQddI0e1ifSi2fDw\nCJiYDFSaA6k8qjZyXdMS6KNGjcL333+P69evw9TUVLJRdGs6s9ZBnmsR6PqClsmzYiP1DzYK3Cib\ntpRz57q6ujqEhIQgIyMDb775JmbNmoXs7GzMnDkTmpqaOH/+fKNqvbLE9DGaHnvah6vr9yjZojNy\nVKtoSo9intBlSVwC3cLCApWVlfjwww/x/fff45133oGDgwN69+7dpsfpbLqHvFIJ6ewHxTV0hJ0Z\n0sW8MjMzIRKJ8PLlS8TGxsLd3Z3FyDqmpKQEixcvxp9//gl9fX3weDw4OjoiKioKubm5jHXixLi4\nXkyZ0f9H10c7clSr6EXtP6OcVlZW8PX1RU1NDfLz8+Hi4gI7OztMnjy5w+mGXCcugR4ZGYmioiIc\nOXIE3bt3h0AgQGRkJH755RdMmDChTY+lKBejyrC+YPv27bCwsKCl7OWgpbVL7UXTy2RPepDOwcEB\nixYtgqWlJQYNGoSJEyeyGFnHDBw4UPJeEwgESnWepihlRDtyVKuU4aK2NdKjnHp6eoiLi8OcOXOw\nefNm5Ofnw8rKCuvXr2c1xrCwMHz11Vdyaatp9bykpOQ2/60iXIwqSoezM3bu3IkzZ87g4MGDbIei\nFGQ1q09H2GVLXMyrsrIShBDs3bsXly5dwo4dO1iOrGOkZ9xUVVVRV1fHYjTyQ9ftU8qKduSoVinD\nRW1rmo5yXrt2DVVVVZLUmzlz5iAtLY3NEOWmuVTbiIhvsGfPnjY/hiJcjCpCh7OtvvrqK1hbW8PG\nxgbbtm1DaGgoSktLMWbMGGzbto3t8JRKaWkpBAIBzp49i4CAAFhZWWHSpElwdnZGbm4u2+EppZiY\nWEydOgvnzxco/LYO0qmiYtra2o2qT3dFyr78gVJetCNHtUlXuqjtiKajnJWVlSxG848NGzbAzMwM\nnp6eKC4uBgDJBbqjoyOGDRuGK1euyLRNZUq1VYQOZ2tyc3Nx4MABZGVl4fz589i9ezcWLlyIN998\nEykpKViyZAnbISqNK1euwN/fX/L/MDAwwJ9//onw8HDaiZMiz7Rf8cDUixfnUFdXqfBrwJsr2DJ3\n7lwsXLgQAoEANTU1LEUmH6mpqfDy8oK/vz+GDBlCU8epLo9WraTaTJkrRjUd5dTV1YW+vj7S09Ph\n5uaGgwcPYtiwYXKNKTc3Fz/++CMKCgrw4sULCAQCODg4ICQkBN9//z0GDRqEzMxMhIaGIikpSWbt\n0up5iuXcuXOYOHEiunfvDgCYNGkSzp49C6D50XuKGWVlZZgwYQLi4+NhZmaGtWvXYunSpQAAS0tL\n2NjYtPIIymPnzp1ISkpCv379GG+rM5V0ucbY2BgFBQWS28uXL5d8r4jr/ToqPz8fRUVF6NOnD9zc\n3JCRkQFXV1e2w6IoRtAZOYpqg+ZGOQ8cOID//Oc/sLOzw8WLF7F27Vq5xpSWloaJEydCQ0MD2tra\nGD9+PJ49e4aMjAz4+/uDz+djwYIFuHfvnkzbpam2iq29nTdZFepQdrq6ujAyMpJ0opuineoGoaGh\nuH79Onx9faGhoYEHDx4AaHh93n77bcltWemqa8DLy8uRlZWlsDOLneHk5IS+ffuCx+PBzs6uS2aL\nUJQY7chRVCuaG+UMDQ1FTU0Njh8/jvz8fBw7dgy6urosRtlwofPy5Uvo6+sjNzcXeXl5yMvLw59/\n/inztpQ91VaReHh4ICEhAc+fP8eTJ0+QkJAAT0/PdnUc6PqTztPQ0EB8fDwiIyMRExMDNzc3xMY2\nrMUqKipi5HOqiHbu3Il+/fohJSUFn3zyCQ4dOgQAOHPmDOzs7NCzZ0+ZttcVB6aaFqNS5DV/HaGs\nBV8o5UQ7chTVTlw5SXp6eiIhIQE1NTV4/PgxEhMT0aNHD5iamiIuLk7ye9KdUFnqCuvHmMKlCmp8\nPh9z586Fo6MjXFxcMH/+fNja2rarc1ZbW4uZM2fCwsICU6dOxS+//NIoVevMmTOYNGkSE+F3KZqa\nmjhx4gS2bt0KMzMz3L9/H1ZWVli7di0sLS1ZHwziCvEgQ1BQkKSq6t69exEUFMRIe4o2MHXgwAHc\nvXu32fuUdd9XOqNNKSvakaOoduDSSZLP5yMgIAA2NjZ499134eTkBACIiorCnj17YGdnBysrKxw/\nflzusSk7rs1gLV26FIWFhUhKSoKzszPKy8tRWloKAwODNv19cXExFi1ahKKiIujo6OCvv/5CcXGx\nJM1t3759CA4OZvIpKCSRSAQLCwts2LABL1++hK+vL7p3747vv/8eYWFhyMvLg4mJCdasWYOqqioY\nGxszFktUVBSGDh0KgUCA0NBQ7N27Fx9++KHk/t27dzdaU8UF/fv3R+/evfH7778jKysLY8aMYawt\nRRqY2r9/P/7+++9m71OmYlTSWjrmcu1YTFEyRwhh9ashBIpSDJmZmURXV0AAIvnS0eGTzMxMtkOj\nOERbW5sQQkh8fDwZOXIkIYSQ27dvk8GDB5N79+6xElN09GGiqWlAdHUFRFPTgERHH27T3wmFQmJs\nbCy5nZycTCZMmEA2btxItmzZQiorK8nAgQNJfX09Q5ErLqFQSNTU1EhBQQEhhJCAgABy6NAhYmRk\nRNTVtYmq6r8IwCPa2trk1KlTjMVx6dIlMm7cOFJXV0cIIeT9998nkZGR5K233pL8zNXVlfz555+M\nxdAeJiYm5MGDB4QQQo4ePUr69etHVq9ezXJU7Pj000+JmZkZ8fDwIIGBgSQiIoJoaWkRc3Nzwufz\nyfPnzxv9fllZGdHUNCDAxf+doy4STU0DUlZWxtIzoCiqrf7XJ2pXP4rOyCmJlgoWhISE4PLlyyxE\npJgUYWG8Mi9y55oJEyagX79+2LFjB0JCQhAeHo5evXrJPY7OziQ3HdVWUVFBUFAQDh06hJiYGPj7\n+0NFhZ5OmmNqatpoD8qCggLcvHkLL16cQ339EwAnUF39FHw+n7EYkpKSkJubC0dHR/D5fCQnJ+P6\n9esYMWIETpw4geLiYtTV1cHS0pKxGNpD+v3m5+eHJ0+eYO7cuewFxJLs7GzEx8ejsLAQP//8M7Kz\ns8Hj8eDo6IioqCjk5uY2Wg8GdM01f+1Bz3+UsqHbDyiR5lIMdu3axUIkikt8kgwO9oKamjFqa0Wc\nOknGxMQiOPh9qKs3dDj37PmW8+s9urrt27fDysoKLi4umDp1KisxdLbEukgkwoULFzB06FBER0fD\n3d0dffr0Qb9+/bBhwwacOXOG2SegwJoWXrh58yYaTr3i/4U5eDx1RsvdE0IwZ84cbNiwodHPMzMz\nsXHjRpibmzO2/qwjSktLJd/n5+fD1tYWgwcPZjEidqSnp2P8+PFQU1ODmpoa/Pz82rQWLDAwAN7e\nIyAUCmFiYsKZ8xPT6PmPUkZ0CFWJNC1Y8OzZM3h5edGNaNuJqwvjubR+j/rHzZs3oaKiIvNtINqj\nszPJ5ubm2LFjBywsLFBZWYnQ0FAAwIwZMzBgwACYmZkxEXaX0PTCu3///gDqARz4308ioKJCGJ3V\nHzlyJOLi4iTHgocPH+LGjRtwcnLCzZs3ERMTg8DAQMba76g1a9bAz88PK1euZDsUxokLJIlEIsTE\nxDT7O23pxIkp0po/WaDnP0pZ0Y6cEmlasODbb7+lC4E7iIsnSWVd5M5F4guuuro6BAcH4/Dhwxgy\nZAi+/PJLVuLpTLqVsbExioqKEBkZiaKiIhw5cgSPHz9GVlYWTp8+jfnz58vhGSiupsdYLS0tfP75\nRvB4wVBR+RdUVHbju++YndUfMmQIPvvsM/j4+MDW1hY+Pj6SqodTp06Fm5sb5ypmxsTE4ssvv8XT\np30xZcqsNlUHdnd3l0NkzBC/T65fv47o6GgAgJubGxITE1FTU4Pq6mqcOHECPB4P2traePToEZvh\ncg49/1HKiteeER5GAuDxCNsxKAORSIRhw4ZJDmq///47tm/fjqqqKkREREAgELAbINVp5eXlMDY2\nx7Nnv6PhZFYATU0viESXOdXhVAY6Ojp49OgRwsPDJZ+x6upqODk5IT4+nrUZrPLy8k6nW4nTl168\neA5CXmD//v2YNWuGjCPt2srLy5GXlwegofosm5/PcePGYdmyZfDy8mIthqZkeSyrr6+HqqoqI3HK\nkviY4eLigsuXL8PU1BRz5sxBVVUVoqOj0bt3b/Tq1Qu+vr4wMDDA6tWr8a9//Qvnz59/ZZ2cMqLn\nP6or4PF4IIS0b4alvdVRZP0FWrVSLoRCITExMZHcTk5OJhMnTiReXl4kJyeHxcj+IRQKiZWVFdth\nKLTo6MNETe1fREWlOwF4JDBwOtshUV0IrYjXeR2tHiprV69eJUZGRmTChAmstP86Ha0OrKWlRQgh\nJCUlhXh4eBA/Pz9iZmYmj5A7TVzpNiUlhYwbN07y8+rqakIIIU+fPiUODg4kLy+PlfgUgfizpaPD\nZ/WzRVEdBVq1knodccECAIiOjoaHhwfnNtGkqZ6dExgYABOTN5GYeBRlZfcQHR3FdkhKratVUGMi\nfUmZ1ulyZR1PTEwsrK2dUFX1Bk6dOtumtEV56uiaTunzR15eHr7++muFr8ocEhICPp8Pe3t7jBkz\nBrW1tV3meCJrXF2/TlFMoh05JSJdsKCqqgqhoaGc6zg1Lcjy/Plz5ObmYvjw4XB0dMSYMWNYLRrB\ndaGhoRCJRFi9ejWioqKwePFitkNSWjExsTA2NseoUQthbGzOuYvljlCE7Te4jAvreLjSmXwdWZTQ\nd3JygpGREYNRykdUVBTy8vKwZs06RETs6FLHEyZwcf06RTGJbj+gJMQFC5pKTk5mIZqWFRcXY9++\nfXB2dsZ7772Hb775BvHx8Th+/Dh69uyJH3/8ER9//DH27NnDdqictHPnTpw6dQopKSk4fvw45zrq\nykL6Yrmh5H8BgoO94O09QqEvMDq7/UZ4eDiioqLQq1cv9O/fH/b29pL3KCEE8+bNw4ABA/Dpp58y\n+TRY07gj3PC+kHdHuLNbUchLZ0vo9+jRg6HImCHOjtHW1sbjx48b3ddVjycURXUe7cgpGVkUO2CS\nkZERnJ2dATSUNt+4cSP++usvjBo1CoQQvHz5Ev369WM5Su7jWsosm8LCwqCtrY1ly5bJrU1FuVju\niI5eYEtvblxTUwOBQAAHBwcADTPxM2bMgLW1NVavXs1k+Kziwj6UXOhMtpWhoWG7XhtFPu6JBzRs\nbGygoqICPp+PuXPnYsmSJV36eEJRVOfQjpwSUYTNMpvOIGlra8PS0hLp6eksRURR7adIF8sd0d4L\nbKDlzY0JIViwYAECAgK6dCdOjO3NmrnQmWSKImcgiLcT6NatG5KSkhrdx/TxJCoqCtu3b0dtbS2G\nDh0KLy8vXLhwAV9++SW2bduG7du349q1a7h+/TpmzZqFc+fOyaRdiqI6j66RUxKKsC4CeLUgi4uL\nC8rLy/HHH38AaNiXq7kUUeofijwqLSsbNmyAmZkZPD09UVxcDAC4ePEiXFxcYGdnh8mTJ6OqqqpD\njy3euPd1ZLHGp6uTfp+6ubnh999/R01NDYsRyQ/b63i6alEIcWdo2LBhOH78OMvRtM/rCiMxeTy5\nfPkyYmNjkZGRgdzcXKioqKCmpkbSWTt37hzeeOMN3LlzB2lpaRg2bFin26QoSnZoR05JcGGRfVtI\nF2SprKzE4sWLERcXh5UrV8LOzg58Ph/nz59nO0xOkx6VVuQR6o7Kzc3Fjz/+iIKCApw8eRJZWVkg\nhGD27NnYvHkz8vPzYWVlhfXr13fo8dv6mnbVi+WOamlzYx6Ph+DgYLzzzjuYOnUq6uvrGYtBJBLB\n2tqascdXJGx3JpmgqFVi21IYianjSVJSEnJzc+Ho6Ag+n4/k5GRcv34d1dXVqK6uxs2bNzF9+nSk\npqYiLS0NHh4eMmmX6hyRSAQLCwuEhITAysoKvr6+qKmpQU5OjuRa6aOPPqLHOyVAO3JK4p/UjF8B\nDAEwHo8fX8SmTZtw+vRpuLm5wczMDNnZ2aioqMDo0aNhbW2N+fPnw8TEBBUVFYzHKC7IEhkZiaKi\nIhw5cgTdu3eHjY0NUlNTkZ+fj8LCQgQHBzMeiyIrLS2FgYEBHjx4AAMDA7bDkbu0tDRMnDgRGhoa\n0NbWxvjx4/HkyRNUVVXB3d0dADBnzhykpaU1+/eRkZGwtbUFn8/HnDlzIBQK4erqCltbW6xZs6Zd\nsXTFi+WOcnBwgJ+fH2xtbfHuu+/CxsYGOjo6kvuXLl0KPp+P2bNnMxqHMg5uKANFrRLbnmwZJo4n\nhBDMmTMHubm5yMvLw6VLl7B27Vq4urpi3759MDc3h4eHB9LS0vDHH3/Azc2t02221AnJysqCra0t\nBAIB7YS0wdWrV7F48WL8+eef0NPTQ1xcHObNm4cdO3YgLy8PAD3eKQPakVMS4tQMDY1pAIqhoZGK\nqKhoiEQiHD58GOnp6YiIiMCGDRsQFhYGDw8PFBYWYuLEibh58yarsSvqKCsbxK9VREQEDhw4gJkz\nZ7IdEuvak2paVFSEjRs3IiUlBXl5edi6dSuWLFmCDz74ABcvXkTfvn0ZjLTrW758OS5fvoxff/0V\nQqEQDg4OSE5OhkAgAACsX78eUVHM7n1YV1f3ygVkaWkpxowZA0dHRwwbNgxXrlxhNAZKthRl6UBz\n2M6WGTlyJOLi4iSv1cOHD3Hjxg24u7sjIiICw4YNg52dHX7//XfJ4JgstNQJ+eGHH5CbmwtVVVXa\nCWmFqamppLMrEAggFApRVVUl6WzPmjWLzfAoOaEdOSUSGBiAjIxkGBkNwM2bJQgMDIClpSW8vb0B\nAFZWVhAKhUhLS5N0AN555x3o6+uzFrOijrKyQfq1Wrv2c6xa9TEGDRrEdlhy5+npiYSEBNTU1ODx\n48dITExEjx49oK+vLymac/DgwWbXeiQnJ8Pf31/ynhf/zbRp0wDQE2NnSW9u7O/vDzs7O7kP1JSU\nlLxyARkSEoJvvvkGWVlZ2Lx5M0JDQ+USCyUbbHeGOoPtvRmHDBmCzz77DD4+PrC1tYWPjw/u3r0L\nDw8P3Lp1C56enlBRUYGRkZFM0yqb64RUV1fDyckJADB9+nSZtdVVaWhoSL5XVVXF/fv3WYyGYgut\nWqlkevbsCV1dXUlqhoqKiuRgoKKigrq6OqipqTX6G7aKZ9C9c9qOi69VZGQkvvzyS6ioqEhKauvo\n6CA7Oxv37t3DF198gUmTJsm8XT6fj4CAANjY2KB3795wcnICj8fDgQMHsGDBAjx79gwDBw7Evn37\n2vR44nVcAC0k01lNZ9vYqKQ7cODAVy4gMzIy4O/vL/n/1tbWMhoDJVttrero7u7OuYqLXKgi6u/v\nD39//1d+Lr1e9ddff5Vpm007IXfu3JHp4zNJJBJhzJgxcHd3R0ZGBvr374+ffvoJt27dwsKFC1Fe\nXo5u3brhyJEjMDU1ZSyOpucjPT096OnpIT09HW5uboxnN1DcQGfklFBrF6Oenp6SA8Avv/yCyspK\neYT1CkUeZZU3rr1WTVMUt23bBgC4e/cu0tPTkZiYiJUrVzLW/urVq1FcXIyzZ8/i0KFDWLZsGWxs\nbHD+/Hnk5+fj2LFj0NXVfeXvRowYgSNHjkjWhFZUVMDNzQ0xMTEAXu2IUM1rS1ERttLhml5AVlRU\nQF9fX7JGKC8vD3/++SejMVCy1daqjlzrxIlxuTASUzPmzXVCtLW1kZmZCQA4fPiwTNuTteZSQ2fM\nmIHFixcjPz8fGRkZjKfiN0095fF42LdvHz744ANJujrV9dGOnBJ6XVVDHo+HdevW4ezZs7C2tkZC\nQgKMjIzkHSIA9lNOFAnXXqumKYp6enoAgAkTJgBoSOcpKyuTa0xtuSCxsLDA//3f/2HYsGHg8/n4\nz3/+g23btmHHjh2wtbVVqFFjtr1ufUt9fT1rgw9NLyB1dHRgamqKuLg4yc8KCgqa/lmHTZw4EY6O\njrC2tsbu3bsBAKdPn4arqyscHBwQEBCAp0+fyqw9ZdWWzpB4fVdCQoJkScGdO3dgZmYm9+NRU1ws\njMTk0obmrj327NmD+fPnQyAQ4OnTp80OtnFF09TQ69ev4/bt2/Dz8wMAqKuro3v37oy1b2xs3Og4\ntXz5cqxduxZ8Ph/5+fnIzc1FaGgozSJRBuINWdn6agiB4pqysjKSmZlJysrKiImJCXnw4AErcURH\nHyaamgZER4dPNDUNSHT04Tb/rVAoJFZWVgxGxy2dea1k7euvvyaffPJJo5/NnTuXHD16VHJbW1tb\nbvGIXxtdXUGHXhvpzwPVOqFQSKytrQkhhFy7do3w+XyyefNm4ufnR0aMGEGGDx9OysrKiKamAQEu\nEoAQ4CLR1DRg9DWWjosQQiIiIkhYWBgRCoXE19eX2NraEktLSxIeHi6zNh8+fEgIIeTZs2fEysqK\n3Lt3j3h6epKnT58SQgj573//Sz799FOZtUe1TPqYM2vWLPLNN9+QsWPHktjYWBaj4iY2Pp/V1dWS\n7zdt2kSWLl3KWFud0dxxZPny5WTAgAEsRtVA+lzVNE6K+/7XJ2pXP4qukaNe0XTdipYWe2+TwMAA\neHuPgFAohImJSbtHK5Wp6lVnXytZGjFiBCZNmoQPP/wQBgYGePjw4Su/Q+Q0UtjZ9YNsrOPqKq5c\nuYJp06YhMjISOTk5yMvLQ2FhoWSkXd5rg5obxS4vL4dQKERkZCQjbW/duhUJCQkAgFu3buGHH35A\nUVER3NzcQAhBbW0tXFxcZN4u9Xrbt2+HlZUVXFxcMHXqVLbD4RzxjHnDMROQnjFn6jMaExODzZs3\nQ1VVFYMGDcL+/fsZaUcWmp6/tLW1JWvlxo8fjxcvXqC+vh6amppyi6m5c5UsswsobqIdOaqR5i56\nAS9GN+ltjaGhYadPHKWlpZgyZQp69uwpqez0999/Y/Hixe3eG4zLZPFayYJ0imK3bt3A5/ObTaWR\nh85ckHCxiIyiKCsrw4QJE3Ds2DGYm5sjJycHo0aNapQuxfbgA9Od9NTUVCQnJ+PChQvQ0NCAl5cX\n7Ozs4OPjQ9dbsuzmzZtQUVHBvXv32A6lXYKCgjBu3DhGCkVJa2sBGVmJiYnFv/+9UvJZXLNmHXr2\n7MlIW7LQ3Pns4MGDCAkJwdq1a6Guro4jR47IbXkDPVcpsfZO4cn6CzS1klMyMzOJrq7gf6kUDV86\nOnySmZnJdmjtJk4rKC4uJnw+nxQWFkruE4lExMLCgty4cYOxts3NzcncuXPJ4MGDyfTp08lvv/1G\nXF1dyeDBg0lWVhZ58uQJmTdvHhk6dCgRCATk+PHjjMTCJWykKHYmRYirn4fhw4eTnJwcVmN4HaFQ\nSAYPHkxGjx5Ndu3aRQghZP/+/WTx4sUsR/YPplPH3NzcyE8//UT8/PwIIYRcunSJqKqqkv/85z/E\n2NiYXL16lRBCyJMnT8iVK1dk0ib1elpaWoQQQmpra4mjoyNJT08nISEhJCIiguXI2q5pijqT5JWu\nz0YaZ1fD1XMV1T6gqZVUZ8l7FI5pTWcFAOD58+fw9/fHN998gwEDBjDW9rVr13D06FFYWFjAwcFB\nsvF6YmIiNmzYAAsLC4wcORJ79uxBVVUVnJyc4O3tLddUDHliK0WxM+W9u9rnQZ40NDQQHx8PHx8f\naGlpsR3OK5hOHTt37hxevHiB7777DpaWljAzM4OhoSF0dXWxf/9+BAYGoqamBjweD5999hnU1dUx\nduxYFBYWdrptqnniWZTPP/8cnp6ecHV1hY2NDZycnDB27FiYmZmxHOGrmm7joqqqKrlv7dq1uHXr\nFvbs2cNIhoO8ZszZSOOUNXGKNlvLGui5SnnRjhzVCBf2tJElXV1dGBkZIS0tTdKRCw0NxZQpU+Dl\n5cVo26amprCwsACAZjdev3XrFhITE7F582YAwIsXL3Djxg1OXkx0FttpHx29IOHC5yE8PBxRUVHo\n1asX+vfvD3t7ewANF3jBwcGor6/Hnj174OjoKLeY2kpTUxMnTpyAj48P5zZTZ/rCR1tbG48fP8bA\ngQNx/fp1VFdXQyAQwMLCAsOHD5eUWRcTiURKtaaXDY8ePQKARun0WlpaKCoqYiuk1xJv43L+/Hno\n6+ujsrISH374IQgh+Oijj1BdXY29e/cyGoM80vUVvRPChXXUXDhXUexgvCPH4/F8AWxFw1YHewgh\n/2W6Tapz2F63IktNZwUqKipQXV2NFStWyKVtseY2Xu/WrRuOHj2Kt99+m/FY2MaFEdeOXpCw+XnI\nzs5GfHw8CgsLUVNTA4FAAAcHBwDAs2fPkJeXh7S0NMybN49TMznSRUV0dXVx4cIFliN6FdMXPjwe\nD/Hx8SgpKcGlS5dw584dWFhYIDg4uMXR+7q6OoSEhDTaZLigoADvvfceVFVV4e3tjV9++YVT/2tF\nxPbsSVu1tI1LeHg4nJ2d8d1337EZnswocieE7UFKaV3p2o1qO0b3kePxeCoAvgEwGoAlgEAej2fO\nZJvKqi0b8LYHF/e06SjxrMCWLVvw1VdfobCwEHw+HwKBALt27WKsXdJKVcbRo0dj+/btktv5+fmM\nxcI2ru1z115sfR7S09Mxfvx4qKmpQUtLC35+fiCEgMfjITAwEADg4eGBx48fS2YbuIapDYVlgcmN\nmAkhSEtLk/yf+vbtixEjRiAtLb3FvblKSkpe2WR43rx5+OGHH5CbmwtVVVW5ztpFRkbC1tYWfD4f\ns2fPxsCBAwEAlZWV6Natm2SD7WHDhuHatWtyi6szmNwbTV6cnJyQk5PTbDVgRcXlTdFfh639MFvS\nla7dqLZhekbOCUAJIUQEADwe7zCA8QAuM9yuUqJpOY01nRVomsrEtNY2Xl+zZg2WLFkCGxsbEEJg\namqK48ePyzVGeVHkEVcukR4ckH5PiTt3rZFeb2NtbY3CwkLweDwQQlBcfD+HiwAAIABJREFUXIxT\np07Bw8NDZvFyIeWoNfKs9FpTU4Nvv92FFy/SXxm9B4CBAwc22mRYKBSiuroaTk5OAIDp06fj5MmT\ncom1aVrfw4cPMWPGDFy6dAmlpaWwt7dHWloanJyccOvWLQwaNEgucXUGl2ZP2qKlbVx8fX0xevRo\nvPvuu/jtt984uQa1I7hSdbk9FD0tlFJ8THfk3gRwU+r2LTR07igGNJeWc+vWLSxcuBDl5eXo1q0b\njhw5AlNTU7ZDZYU802ma7lclvY5B+r6ukhrTFjTto/3c3NywcOFCrFq1CrW1tThx4gQWLFgAQghi\nY2MxbNgwnDt3Dnp6etDW1n7tYzW33kacqnXixAls3rwZrq6uMotd0S6ameDp6YnvvvsOs2fPxr17\n95CWloZu3Qzx4sWro/e9evVqlI6tqqqKO3fusBM4Xk3r09fXh4eHB1JTU3H9+nWsXr0au3btgqen\nJyfXZzaHCyne7fG6bVwmT56MR48eYfz48fj5558bvXco+aGDlBTbOFHsZP369ZLvhw8fjuHDh7MW\niyIrKSlBbGwsdu3ahWnTpiEuLg5ff/01Pv74Y/j5+eHFixd4+fIl22GygoszA4qyTkOWFHHElU0O\nDg7w8/ODra0tevfuDRsbG+jq6oLH46F79+4QCASoq6vDvn37Wn2sltbblJSUYMWKFUhNTW1UEa+z\nFO2iWdZUVFQwYcIEJCUlwdLSEkZGRhg6dChSU8+judH7p0+fvpKOLe6gZ2ZmwsnJCYcPH2bjqUh4\neHhg586duHPnDsLDw/HFF18gJSXltbO4IpEIY8aMgbu7u2SQ8auvvsKsWbOQk5MDALh69SoCAgIk\nt5miiLMns2bNarFQUFBQEIKCguQcEdUUHaSkOiolJQUpKSmde5D27lfQni8AzgB+lbq9CsDKJr8j\n0z0YlJV43yax//73vyQ8PJwMGDCAxai4gYt71Ij359HVFTC6Pw+l+KqrqwkhhDx9+pQ4ODiQvLy8\nDu3H9/XXX5NPPvnklcfm8/kkPT1dpjETws3Pnbzcv3+fmJiYNHtfS3tzife9FIuIiCBhYWEkMzOT\n2NjYED6fT5YuXUrc3d3l8hz++usvYmZmRh48eEAIIaSiooLU1NQQExMTMnLkSEIIIaGhoWTAgAGk\noKCgxccRCoVETU1N8jsBAQHk0KFDZMSIEeTixYuEEEI+/vhj8s033zD8jBrIam+0uro6GUfWOjb2\n4aQoSn7QgX3kGC12AiALwFs8Hs+Yx+OpA5gGoGsuAuKApmk5lZWVLEbDHVxbjCydclZVlYNnz35H\ncPD7nCwGQbEvJCQEfD4f9vb28Pf3x6VLxR0q1jBixAgcOXIEFRUVAICKigoEBQVh3rx5Mk2pFBOn\nHGlqekFHRwBNTS+lSDm6c+cOXF1dsWLFimYLvbRU1KFpOvby5cuxdu1aGBoaYvfu3Th16hT69Okj\nqVrKNOm0Pj6fj+XLl0NdXR1GRkZwcXEB0DBDV11d3WqhLVNT00Zr/0QiEd577z3s3bsXL1++RGxs\nLKZPn874cwJafv1FIhGGDBmCoKAgmJmZYcaMGTh9+jTc3NxgZmaGrKwshIWFYfbs2XB3d8fs2bPl\nEq9YVyjSQlEUA9rb82vvFwBfAMUASgCsauZ+xnq2ykQoFBIrKyvJ7YiICLJ+/Xri4uJCEhISCCGE\n1NTUkKdPn7IVImu4NjOQmZlJdHUF/4ul4UtHh08yMzNZiYdSHJ19L0dGRhIrKytiZ2dHvLy8iKqq\nKuHz+cTOzo7w+XySk5PDSMzKOIsgi1n36OjDRE2tB1FR0SQ8nirh8wXk/v37DETLnJZmGp8/f04G\nDx5MfvrpJxIQEMBihA3EM4d//fUXIYQQe3t7Mm/ePEIIIcePHycTJkwg69evJw4ODqSmpkausbX2\nudfS0pI8B+nrAIqiFAs6MCPH+Bo5QsivALreDscc1FxlxIMHDyIkJARr166Furo6jhw5wun1AEzg\n2mJkRVynQXFDZ9edSa+3kdcaTWVcFymLQi/ix6itzYD4OHH5shdn1jm35/1DmtmKRUNDA6NHj0Zo\naCjjm1q3lampKSwsLAAAlpaW8Pb2BgBYWVlBKBSCz+fDz88P6urqco2rtc/96yokUxTVtXGi2AnV\nec2l5YglJSWxERKncGkxMtc6lpTikNUgABeL/3Qlsij0wuViMe19/7TUuZgxYwYSEhLg4+PDVKjt\nIr08QUVFRXJbRUUFdXV1AIAePXrIPS46+EdRVEuYXiNHsYjLG/GygUsbZSrq5qcUu2Sx7oyu0WRe\n4wtvoCMX3rJ4DCa09/3T3CBjaGgosrKy8OuvvyIoKIgzs0jNzRxygbKuN22rdevWITk5me0wKIoV\ndEaui6Ij7tynjClnVOd1dnaZyzM9XYUsZt25OnPf2feP+NxUW1uP+vpqfP/998wG3A6vS1Fku7PJ\npawSrgkLC2M7BIpiDY/tESgej0fYjqGrKS8vh7GxOZ49+x3iNAxNTS+IRJfpwZ+ilBzXjg8ikQi+\nvr5wdnZGRkYGHBwcMHfuXKxfvx7379/HoUOHEBwcjHPnzkFHRwdvvPEGtm7dipkzZ2LOnDmYPXs2\nRo4cKfe420IW6xC5tt9kZ94/XHvvdRU6Ojp49OgRRCIRxo0b12gGtCtpbk/ChIQEhIaGYty4cZg0\naRLbIVJUp/B4PBBC2jVqRFMruyCuldunKIo7uJimde3aNaxYsQLFxcUoLi7G4cOHkZ6ejoiICGzc\nuBHu7u5IT0/HX3/9hUGDBiEtLQ0AcP78eUa2TpAVWaRzcyklHOjc+0cRz02KsERBejC8qw+MX716\nFYsXL8aff/4JPT09HD16lPXZUopiE02t7ILowmiKol6Ha2larVULnDp1KlJTU2FsbIyFCxfihx9+\nwO3bt2FgYABNTU02Q1dKHX3/KNq5SVGWKChT1cqmexJyeRCAouSBzsh1QVwccacoilu4NNPzumqB\n9fX18PT0RFpaGs6dOwcvLy+88cYbiIuLg4eHB1shK72OvH8U6dykSEWBHj16BODVwjJdkfSxQlVV\nVVJNlKKUFe3IdVEdrYro7u6O1NRUjBs3juEIKYqiGrSWDvbmm2/i/v37KCkpgYmJCdzd3REREQFP\nT085RUjJiqJU7FWUNFBFSP2Upa6eOkpR7UU7cl1YR0ZMz507B6Drp2dQFMUdbUkNGzp0KMzMzAAA\nHh4euH37Ntzd3eUSHyVbXJoNbglXt3+QFhMTC2Njc4watRDGxuaIiYllOyTGNT1WiG/TaxZKWdGq\nlVQj2traOHnyJCIiInD8+HFkZWVhwYIFOHr0KExNTdkOr1VtqYD3888/Q1tbG8uWLQMAWFtb4+TJ\nkzAyMmI5eoqiKIorxGvkpLd/4MoMIq0A+g8/Pz8sX74cw4YNYzsUiuqUjlStpMVOqEbEo1qEEJw/\nfx7//ve/kZiYiDfffJPlyNru2rVrOHr0KCwsLODg4CCpgJeYmIiNGzeCz+c3+n06kkdRioFrpfip\nro1rRYGkKfN+kNLHgVWrVuHZs2d0dp5SWjS1kmpEPDt66dIlLFiwQOE6cUDrFfCaojPClKIRiUQY\nMmQIgoKCYGZmhhkzZuD06dNwc3ODmZkZsrKysHTpUoSHhwMATp06heHDh7MbdCcpYxoZxT6upoFy\nIfVTJBLBwsICISEhsLKygq+vL54/fw4vLy+sWrUKQ4cOhbm5OdLT02XWZtPjgLe3D06fPg1VVVWZ\ntUFRioR25Khm9e3bF927d0dubi7bobTb6yrg1dXVoVu3bnj58qXkd54/fy73GCmqs1rae23z5s34\n/PPPsWnTJvz4449ISUnBkiVLsH//frZD7jBFqiBIUfLAlQqgze3rBgD19fW4cOECtmzZgvXr18uk\nLXocoKhX0dRKqln6+vrYs2cPvL290aNHD4XKPW9ths3ExASJiYkAgNzcXFy/fl0eYVGdFBYW1mht\no7JraebZ2toaQqEQ3bt3x65du+Dp6Ylt27ZxqkhDeylzGhlFtYQLqZ/N7evG4/EwadIkAIC9vT1E\nIpFM2qLHAYp6FZ2RoxqRXi9maGiIEydOYNGiRcjKymIxqvaRfg43b95EQkJCo/smT56MiooKWFtb\n49tvv5VUwgNaT1nLzs7Gw4cPMXHiRNja2sLV1RWFhYVyfX4UBbQ+8wwABQUFeOONN/D333+zEqOs\ntDeNbOLEiXB0dIS1tTV2796NGzduYPDgwaioqAAhBJ6enjhz5oycoqco5rCd+tnSvm7in8tyrzcu\npJNSFNfQGTlK4sGDB+jZsyc8PT0l+zMNGDBAoToqTTdElb5YE99XXl6Ozz77rMURzNcVS9mwYQMG\nDBgAgUCA+Ph4/P7775g9ezby8vLk8vyUzYYNGxAZGYnevXujf//+cHBwwO7du7Fr1y7U1tbirbfe\nwsGDB9G9e3dG2ufyLGBrM883btzAli1bkJeXh3feeQfjx4+Hk5OTnKKTLXEaWXCwV6MKgi1dvO7b\ntw96enp4/vw5HB0dMXnyZKxatQoLFy6Ek5NToxlMiqI6ri1rzGW1Dr29xwFlUVxcjGnTpkFFRQVx\ncXEKUWGckh06I0cBAO7cuQNXV1esWLFCYTYYbUvBhwMHDmDx4sWSv2lLwYTWiqWkp6dj1qxZAAAv\nLy9UVFSgurpaDs9YueTm5uLHH39EQUEBTp48KZkVnjx5MjIzM5GXlwdzc3Ps2bOH5UjZ8bq91wgh\neO+99/Dll1+iT58+2L17N+bPn48XL17IO0yZac9G0lu3boWdnR2cnZ1x69YtlJSUYN68eXj06BG+\n//57REREyDFyiuq6WtrXraXf6SxF2VBenhISEuDv74+cnBzaiVNCdEaOAtBQ3KS4uFjS0VFXb0hh\n4NK+Oc1pafbs+PHj2LhxIyZOnCg5iUgvlG7IsS9AcLAXvL1HNBrRay1lTV1dvVEMtOolM9LS0jBx\n4kRoaGhAQ0MDfn5+AIDCwkJ88sknqKysxJMnTzB69GiZttvcLOC1a9ewcOFClJeXo1u3bjhy5Air\nJ8ymM8979+5tdF/TWXSBQICLFy/KLT6mGBoatjr6npqaiuTkZFy4cAEaGhrw8vLC8+fP8ezZM9y6\ndQsAUF1djR49erSrbZFIhLFjxypUhgJFManpcWjZsmUQiUQ4fPgw6uvrUV5eDkNDQ5SWlsq03bYc\nB7oikUiEMWPGwN3dHRkZGejfvz/+/e9/Y+vWrejWrRuSkpKQlJTEdpiUnNEZOUpCEStCtVbwQZp4\noXTD5qmA9EJpaa11zDw8PHDo0CEAQEpKCgwNDaGlpdXJZ0K1BSEEc+fOxbfffouCggKsXbtWplVH\nW5oFnDFjBhYvXoz8/HxkZGSgb9++MmuTKYoysy5rVVVV0NfXh4aGBi5fvow//vgDALBy5UrMnDkT\nn376Kd57770OPTbdc5J6ncjISNja2oLP52POnDmIi4uDtbU1+Hy+wm//0VY//ZSIy5ev0G1CGNK0\nSmhlZSVCQ0Px4Ycf0k6ckqIdOUqirR0dLmlLwQexti6Ufl3KGo/Hw/r165GTkwNbW1t8/PHHOHDg\ngKyeDiXF09MTCQkJqKmpwePHjyWVRqurq9GnTx/U1tYiKipKpm1KzwJqa2tj/PjxePr0KW7fvi2Z\nEVRXV2dsTZ6sKPOea76+vqitrYWlpSU+/vhjuLi4IDU1FdnZ2Vi5ciUCAwOhoaHRoc9tXV1ds3tm\nLVu2DI6OjrCwsEBWVhYmTZoEMzMzrFmzhoFn2CA8PBzm5ubw9PTE9OnT8eWXX+Lrr7+GpaUl7Ozs\nMH36dMbapl5VVFSEjRs3IiUlBXl5edi6dSs+/fRT/Pbbb8jLy8Px48fZDrHTmts3rqamBjk5ObCz\ns4O1tTWWL18BQgYqzGCwomlaJfT69et4+PAhvvjii0b/l8uXL2Po0KGSvxOJRLCxsWnpYSkFRlMr\nKYnGHZ2G1EOuV4RqT1pjWxZKt5ayJr4vPj5eBtErj46kpfH5fAQEBMDGxga9e/eGk5MTeDwewsPD\n4eTkhF69emHo0KF4/PgxY3ErYtpsW1OIuyp1dXX8/PPPr/xculMVFxfXoccuKSlBbGwsdu3ahWnT\npkn2zNLQ0EBWVha2b9+O8ePHIz8/H3p6ehg0aBCWLVsGfX39jj2ZFmRnZyM+Ph6FhYWoqamBvb09\n7O3tsWnTJgiFQqipqeHRo0cybZN6veTkZPj7+0v+1/r6+nB3d8ecOXMwdepUSTl+RXf16tVGn4G4\nuDh88cUX+Pbbb6Gurg5397EAxANddHsAWWtaJVSckXL//n0sXrxY8n/JyclBbW0tRCIRjI2NERsb\ni2nTprEVNsUgOiNHSXBlg9H2aG32rKnOLpRW1nQ1WehIWtrq1atRXFyMs2fP4tChQ1i2bBkWLFiA\n0tJS/PHHH9i2bVujznZnNTcL+K9//Qv9+/fHTz/9BAB48eIFnj17JrM2ZY2rM+vNjeZXV1fDyckJ\nZ8+eBdDw/2ZqFksWn92BAwc2u2eWeLbW2toa1tbW6NWrF9TV1TFo0CDcvHlTJvFLS09Px/jx46Gm\npgYtLS2MGzcOAGBra4vp06cjKioKqqqqMm+Xap5IJMLGjRtf+XldXR3mz5+Pmzdvwt7eHg8fPmQh\nOtlqbt+4qqoquLm5wcTEBDzecwDidHfuDwYrmqaDi+LbBgYGjf4vIpEIU6dORWxsQzZGbGwsAgK4\nW++A6jjakaMaUaSKUM3Nnk2aNAnl5eUoKytDUlISZs+eje3btzf6u47uu6PM6WqyVFpaCoFAgIiI\niEYVRceNGye5oH8dJjvT0rOA7777rqRc/8GDB7F9+3bY2trCzc0N9+7dk3nbssLlvZak13fo6uri\n+PHjOHDgABYuXIikpCT89ttvWLdunczbldVnt7U9s6TTu4GGwQtZ7aH1OoQQ8Hg8nDx5EosWLUJu\nbi4cHR3x8uVLxtumGmhpaeHIkSOoqKgAADx8+BCrVq2Cv78/wsLC0KtXL0Y69fLW9DNw//59yW1D\nQ0OEhX0CHq9UYQaDFU1zA9Y8Hg/duv2TYCc+Nok7ciUlJVBRUcGgQYPkHS4lB7QjR72C7Q1GO4Op\nzpYiFoLhoitXrmDKlCk4cOAADA0N2z1LJ4/OdHOzgIMGDUJSUhIuXryIrKwsTnSKWsLlmXXp0Xx7\ne3sIhUIMGTIEs2bNwtixY7Fv375GFySyIMvPLldSbd3c3JCYmIiamhpUV1fjxIkTePnyJW7cuIFh\nw4Zh06ZNePToEd0WRY5UVVVhYGCAfv36QV9fH0uWLMHQoUPx9ttvw8bGBm5ubl1ijVLTz4Cenh70\n9PSQnp4OAHj48AHMzQcrxGCwPLVlu6SwsDDMnj0brq6uMDMzw+7duxs9hvTgdXl5OTw9PREaGool\nS5agZ8+er7Q5cOBAqKqqIjw8nM7GdWF0jRzVLiKRCL6+vnB2dkZGRgYcHBwwd+5crF+/Hvfv30dU\nVBQIIViyZAlqamqgqamJffv24e2332Y8NibXBonT1RoeF6C5/+1XVlaGCRMm4NixYzA3N0dubm67\n/p6ttV/l5eUQCoUtbiDPRYGBAfD2HsG5uFta31FQUAB9fX1GZjpl+dlty55ZLf2+LDk4OMDPzw+2\ntrbo3bs3bGxsoK+vj5kzZ6KqqgoAsGTJEujo6DDSPvWq4uJiZGRkwNnZGe+99x6GDBmCW7duISIi\nAgKBgO3wZKa5GaF9+/YhKCgIKioq8PHxQbdu3eDo6MhShNzV0nZJiYmJ2LhxI+zs7FBYWIgLFy7g\n8ePH4PP5GDt2LPr06dPocWJiYhEc/L5km6hNm8JaPNYEBATgo48+wmeffSaPp0ixgRDC6ldDCJSi\nEAqFRE1Njfz111+EEELs7e3JvHnzCCGE/PTTT2TChAnk8ePHpL6+nhBCyJkzZ8jkyZPlEltmZibR\n1RUQgEi+dHT4JDMzs9OPXVZWRjQ1DQhw8X+PfZFoahqQsrIyGUTe9QmFQjJ48GAyevRosmvXLkII\nIYcOHSIffPCB5He8vb1Jampqi4/B5P+3JdHRh4mmpgHR1RUQTU0DEh19mLG2ujqhUEisrKwktyMi\nIkhYWBg5duwY8fX1JSUlJWTw4MGkqqpKpu3K87NbVlZGMjMz5XJcqK6uJoQQ8vTpU+Lg4EDy8vIY\nb5NqnlAoJMbGxpLbycnJZMyYMcTe3p6cOXOGvcCaSEhIIJf+n707j2rqTP8A/g2IgLLXpS4FcQMi\nARJAUVnEulZcR6QotCpqtdVal9qqtRVpnVbF9aeOtagg4C62UmsVK7KobAFBUUQx0VFnQEABBUR4\nf3/QZCKEJZDkZnk/53AOkOTmSXLvzX2357l9m+kwtJLoO1Dko48+ItHR0YQQQgoKCoiTkxMJDg4m\n3333nfg+H3/8Mfn111/f2k5rz2fKPBdR8vN3m0imdhSdWknJrLnabUKhEM+fP8f06dPB4XCwbNky\n5ObmKiUuRa4NUuXpaupCX18fMTExiIiIwJEjR9CnTx9kZmaCEIJHjx4hNTW12ccre+0XnU4rfw1H\ntJ49e4bVq1cjLCwM/fv3x5IlS7B06VK5Pqeyjl1lr6FdsGABuFwunJ2d4evri169etFETAyS3Lcv\nXbqMP/+MQ1ZWHiZMmKYy66nPnDmDW7duyX27NAlY67SmXJLkfkT+XvsqqTXJrOh6fi0ja8tP3j+g\nI3JqRSAQEA6HI/579uzZ5NSpU+Lb7O3tyezZs8muXbvE/7O2tlZafKIRFBMTrkJGUGgvV9tI7jfP\nnz8ngwcPJr/99hsJCAggdnZ2ZNq0acTb27vZETlCFP/5SmJiBJBSHEUeu0yP2NORY2YJBALCYrHI\n9evXSWFhIdHV1SfASgJ4E+AoY7M3wsPDiYODA3FyciKjR48mFhYWpG/fvoTL5ZKCggK5PAfd91qn\n4YwEaddO69evJ1wul1RXV5Nnz54RKysr8vTp07e209K5hulzEdU+aMOIHF0jR8mMtLDgv6ysDL16\n9QIAHDx4UBkhiSl6bVDXrl3pKFwbSC7SNjU1RUpKCgCI06a3ljLXfjVXV9HY2Fih9eu0hTLXHyry\n2GVyDa221w1UFba2tti9ezeSkpKgo2OI2toQAOMBDGRkPbWoQPm1a9dgbm6O58+fY/ny5fDx8ZFb\nTTu678mmpXJJLBYLDg4OGDFiBIqLi/Htt982Wh/XUj1cup5f+9CplZTMWjoZrVq1Cl9//TWcnZ0Z\nSX+tzlk3VYm7u7vCn6MtU3KU9fk2NyVPUUkstIkmTf9hsuSDqtYN1CZWVlbIzc1FaGgo/vWvf6FD\nBx0AdwEUA/gvI+U/GhYoNzMzk3vWVbrvtV5T5ZIkb6uoqICZmRl+/fVX5OXlYe7cuVK31VyZKFUu\nP0MpiKxDePL+AZ1aSVEK03A6hypRlyk50qbkGRsbi3/fvHkzcXV1JY6OjmT9+vVMhKh2NHH6jzKn\n/UrSxPdSHUmez/T0jAiL1YHo6Zkzdm7btWsX+eabb976n+R0Pnmg+578REcfJR06GBIDg17t3meY\nOhdR7Yc2TK2kDTlK7ug6MtXRcE2jLIyMjEh8fDzx8fER/2/x4sUkPDy83XGp+wWAqCF34cIFsmDB\nAkIIIXV1dcTHx4ckJiYyGZpa0NT1h0yd++iFG7Oknc8MDMzIn3/+ydg57datW8TGxoYUFxcTQggp\nKSkhS5YsIQcPHpTr87R232tNp2LD7xttoYjvQ3odpp7a0pCja+QouWpY3yQsbA8tCMqwmpoaBAQE\ngM/nw97eHuHh4diyZQtiY2NRWVmJYcOG4V//+lejx4nm7CtiGqGmzOO/cOECLl68CB6PB0IIXr58\nifz8fKVMS1Vnza0/VGdMraFV1bqB2kLa+axjR2uYm5sz9lmw2WysXbsWXl5e6NChA7hcLubPn495\n8+Zh165dOHnyJKytrdv9PLLse635LtHGaeuK+D6k6/m1B10jR8kNTdeumvLy8rB48WLk5ubC2NgY\ne/fuxZIlS5CSkoLs7Gy8evUKv//+e6PHETmvp5CkKfP4CSFYvXo1+Hw+MjMzcffuXcyZM4fpsGSy\nc+dOsNlsBAYGKu05aTkP+aNrg5mjquezwMBA5OTkIDMzEwcOHMDQoUNx69YtZGRkyKURJ9LafU/U\nqchmszFjxgxUVVXh/PnzsLOzg4uLC06fPi23mNSJqu4/lHqgDTlKbujCZ9VkaWkJNzc3AEBAQAAS\nExPx119/wc3NDQ4ODrh8+XKTtYU6dOiA2tpa8d9VVVVyiUndL+RFjdyxY8fiwIEDePnyJQDgyZMn\natdxsXfvXsTFxeHw4cNKfd7mFuy3lmjk8+nTp5gxY4a8Q6SoVlH185mq1HmT7FQ0MTFBaGgoFixY\ngLFjx2LmzJn4z3/+I76vUCgEh8NhMFrlUfX9h1JttCFHyQ3tVVJN0jKLfvbZZzh9+jSys7Mxb948\nqQ00FoslzsZ27949sNlsXLp0SW5xyeNCnimi93T06NGYOXMmhg4dCgcHB/j6+qKiooLh6Fpv0aJF\nePDgAcaNGwdzc3N8//33AIA///wTI0aMUPjzt3cUKSkpCQDQo0cPHD9+XJ6hKYyxsTHTIVAKoKrn\nM1XKDivZqThr1iykp6ejb9++4syaAQEBb91fm6ZZqur+Q6k+2pCj5Ib2KqkmoVAortsWHR0NDw8P\nAMA777yDiooKnDx5UurjWCwWevXqhRkzZmDMmDF49OgReDyeXGNT1+lgZWVlAOp7ut3c3HDp0iVk\nZ2cjOTlZrlOWFG3v3r3o2bMnrly5gidPnuDYsWOIj4/H0qVLcejQIabDa5GoUaROvffadHGqbVTt\nfKZqyx0k9/3IyEj8+eefyMrKQl5eHgDg/v37uHLlCrhcLnbv3s1IjExStf2HUg+0IUfJFe1VUj2i\nQrVsNhsvXrzAokWLMG/ePAwaNAjjx4/H4MGDGz2muLgYFhYWAIAVk2mtAAAgAElEQVSVK1di+/bt\neO+993Dy5Em4u7uDx+MhKSkJc+fOhYODA5ydnREfH6/kV8YsVerplqY1jRtvb29UVVWBEAJDQ0P8\n/PPPGD16ND7//HO1GElvrqalorW38fjdd9+By+WCx+Ohd+/eCAoKkmN0FKV6yx1EnYp8Ph9nzpzB\n6tWrYWJigqtXrwIAgoODweFwkJmZyUh8FKWOaNZKSu5otiTVIZoa2VBISAhCQkKkPubp06cYMWIE\nvvzyS3EW0g4deqCi4i62bt2GqKhIhIeH4+LFi9DR0UF2djby8vIwZswY5Ofno2PHjop+WYyT7Omu\nzzSWjaAgb4waNVKl9n1ZGzfZ2dno0qULHj9+rKCINEt7Go/BwcEIDg7Gixcv4OnpiSVLlsgxMopS\nveywok7FCxcuoHv37vjyyy8xePBgzJw5E9u2bUNtba24AzEwMBDnz59nJE6KUie0IUdRWqioqKjJ\ndNE9evRAXl4eioqKYGVli8rKywBMAThj5cqVSE5OAofDwXfffYfPP/8cAGBjY4M+ffrg7t27sLe3\nV/4LUjJ1K59QUFCA6dOnY+fOndi9ezeys7NhY2Pz1tpIoVCIbdu2ITMzEx988AEmT54sdbSWakz0\n/s6cORPJycl4+fIl7t27h+XLl6O6uhpRUVEwMDDAuXPnGj02ICAAK1asgJOTEwORU5pMtNwhKMgb\nenpWqKkRMrbcQbJTcceOHSgtLUV5eTksLCwwY8YM9O/fH7t27cJvv/2m9NgoSp3RqZUUpWVaOyWw\n8bScd6Cj0xmxsbFS76/IcgWqRp0S+9y9exe+vr4IDw9HWloaOnfujFu3biE4OBjp6elgsVgghGDe\nvHkIDQ3Fu+++i19++QXz58/H69evmQ6/WZL7HFP73927dzF9+nRERESga9euuHXrFs6cOYPU1FSs\nXbsWJiYm4PP5cHNzQ0RExFuPXb9+PSwtLfHRRx8xEjul+VRxuYOnpycOHToES0sbjBo1H/v2/YzM\nzCyYmZkhOTkZABAVFcVwlBSlHmhDjqK0iCyL3xs3Vgj09Drg0qVLOHLkCDw8PBAZGQmg/mL20aNH\nsLGxUdprYZK6JPYpLCzElClTEB0dDQ6Hg4SEBHFmOA6HA0dHR5w9exbvvPMOLl68iAkTJgAAeDwe\nbty4ofLTZJlcIwe8/f6KRqK9vb3RqVMndOnSBebm5vDx8QFQ/34LBAJxg/Ps2bOIi4vDjh07lB43\npV1ULYlG79698fhxIaqqTFBWZgJCxuHYsVPYunUrPvvsM7kn1aIoTUYbchSlRWRZ/C7ZWDEymgAW\nqwAHDuzFn3/+ie3bt2PAgAEghMDBwQH+/v4IDw+Hnp6eEl9N20lLVBEeHi7TOiVV7OluyNTUFJaW\nlkhISJB6u+QolqrUmpKFKHuolZUVsrOzW7i3/Ine38TERPH/9PX1xb+zWCzx3zo6Onjz5o24wblt\n2zY8efIErq6u4PF4WL9+vVJjpyimCAQCdO5sB0AAIAHA7+jUyRbGxsbIysoCn8/Hjz/+yMgxTVHq\nhq6RoygtIuvid39/P4waNbLRerqUlBQUFRWhe/fu+PHHH1Wmp1cW0kZwZB3VUfXEPvr6+oiJicGY\nMWNgZGQET09PREVFYcSIEbh586b4QkmU1KZjx/r9Iyxsj0o2TCU1t85TWRq+v61RVlaGoqIi/PTT\nT4zGTlFMaep7yMjICGlpafS4oCgZ0BE5itIibZkSKG1ajqqn3pdFQUEBeDwenj17hsePH2P8+PGw\nsbHBV199xXRocmFoaIjY2Fhs374d/fv3x8uXLzFo0CCsX78eLi4uKC0tValaU63R1v0vODgYW7du\nlWssku9veXn5W7dJ6xjQpGOHotpC2vdQUFAgnJ3d6XFBUTJiMZ2ggMViEaZjoCht057RjLezWdb3\nphoaekMovNPqbQmFQvj4+CAnJ0f8v4yMDBw+fBjbt2+XKZ62EAqFmDhxIk6ePIkPP/wQERERyMjI\nQEhICLKysqCnpwcbGxskJyejV69eCo9HWaR97mlpaRg9eiFevMgQ38/EhIe4uH1wdXVlKtQmtWf/\nCw4OhrGxMZYvX66UOBu+1/I4dihKU4iOESMjIzg7u9PjgtJ6fycfk2lqEB2Roygt1J7F7/IqMttw\ntMLZ2VkpjTgRaYkq3n//fRgZGUFfXx9sNhtCoVBp8ShaUyNB6pSBE5B9//vhhx9gY2MDT09P5OXl\ngRACb29vfPHFF+ByuXBwcEBaWppcY2zqvVa1As2UajA2NmY6BEaIvocqKirocUFRbUQbchRFyUTe\nF/6iqY1btmzBxIkT5RRly1pKVKGrq4s3b94oLR5Fai5bqbpk4BSRZf/j8/k4fvw4srOz8fvvvyMt\nLU3cgVBZWYnMzEzs3r0bc+fOlVt8zb3X6tZoppSDiYyrqoQeFxTVdrQhR1GUTOR54S9Zg8vV1VWp\nFzSiRBURERE4cuRIu7a1b98+cSkGVdTSSJA6ZOAUkWX/S0xMxNSpU6Gvrw9jY2NMnjwZhBCwWCz4\n+/sDADw8PFBeXi7OgNlezb3X6tZoppTrypUrGDFiBKZMmYL+/fvj66+/RmRkJAYPHgxHR0c8ePCA\n6RAVgh4XFNV2NGslpfWkrdcSUeaaGnXSVDZLWYimNp4+fRq2tra4cuWKAiJtnihRxZgxYxAYGPjW\nbbI0Kj/55BN5hyZXrclWquoZOCW1df+TXI8t+fmKGnfy0NJ7LY9jh9Jc2dnZuHPnDszMzGBtbY35\n8+cjNTUVO3fuxK5du+SerEdV0OOCotqGjshRFOjUlrZob5FZaVMblUWy7pipqSlSUlLg5+eHwMBA\ncbbG3377DZ6enlIfHxERAUdHR3C5XHz88ccIDg5GaGio0uKXlSb2eLdm//P09MSZM2dQXV2N8vJy\nnD17VrSYHMeO1a9bS0pKgpmZmdzWKbXmvVa1As1tIa0WY0ZGBr744guGItIMrq6u6NatGzp27Ij+\n/ftj7NixAP5XUF6TacJxQVHKRkfkKApATU0NAgICwOfzoauri+rqahgYGODf//43DAwMsGXLFpw5\ncwaDBw9GcXExXFxcNHaai7I0rMHVs2dPxmKRpY5abm4uNm7ciGvXrsHc3BzPnz/Hjh07VL4zQBt7\nvLlcLvz8/ODg4IDu3btj8ODB4tsMDAzA4/Hw5s0bHDx4UK7Pqy3vtbSERc7OzgxFoxkk1+nq6Og0\nKihPURQliY7IURSAvLw8LF68GJGRkXjy5AlmzZqFkpISmJiYYOXKlSCE4MSJE+L7q/pFu7porgaX\nsjSXnEKav/76C76+vjA3NwcAmJmZKTPcdtHGHu/Vq1cjLy8PCQkJiIyMFE+TFnXcZGdnK6TxoU3v\nNVMJizQFLcFEUVRb0YYcpVA7d+4Em81utP5I1VhaWsLNzQ2JiYnw8fHBlStX0L9/f0yfPh2EELz7\n7rvg8/lMh6kxRFMbi4qKcPfuXcTGxsLHxwe//fab0mOhKeG1D4vFQmlpKdLS0lS68Lk6YDJhkaZo\n6j2j76V87dq1C2w2G++99x6uX7/OdDgUJRe0IUcp1N69exEXF4fDhw8zHUqzGn5hikZZRD2lurq6\n4tuqqqqUF5gGa6rWlrLJmvp65MiROHHiBEpKSgAApaWlygiTkqP58z/BxIkzGN/32iI4OFhlEl5I\nq8VIyU6UMdXLy0vcmVVUVISffvoJ7733XqPbqLbZs2cP4uLiMG/ePCQnJzd73507d2LQoEGYNWsW\nRo8eDR6P99asHIpSFbQhRynMokWLUFBQgPHjx2Pjxo0ICgqCm5sbnJ2dcfbsWabDe4tQKERKSgo8\nPT0RGxsLLpeLgoICnD59GiwWC+Xl5ejSpQsA0JO5HMg6nVGRZE0EwmazsXbtWnh5eYHL5WLFihVg\nsVgK7T2fOnUqXF1dweFw8MsvvwAADh48CBsbG7i5uWHBggX4/PPPFfb8mkSV9j1pJBPpTJs2DX37\n9kVtbS0AoLy8HNu3b0ddXR3DUdZjMmGRJlOVTi5NIroesbW1xY8//oiNGzfC0NAQAwYMwMcffwyB\nQIBhw4bB0dER69atwxdffIGLFy/i888/B4vFAp/Ph6+vL9Mvg6IaI4Qw+lMfAqWprK2tSXFxMVmz\nZg2JiooihBDy/PlzMnDgQPLq1SuGo6snEAiInZ0dCQwMJHZ2dsTe3p4MGDCAcDgcYm5uTnr06EGm\nT59OOBwO4fF4ZN26dcTa2prpsNVaamoqMTXlEYCIf0xMuCQ1NZWxmAoLC0lqaiopLCyU+bFLliwh\nhw4dUkBU9UpLSwkhhFRWVhJ7e3vy+PFjYmlpSYqLi0lNTQ0ZPnw4WbJkicKeX5Oo4r4ncuvWLWJj\nY0NKSkoIIfWf+9y5c8msWbPIwIEDyYABA4itrS0JDQ0lO3fuJGw2mzg6OhJ/f3+lxyoQCAiHwyGv\nXr0i7u7uJDo6msTHx5OJEycqPRZVYGRkJPX/s2fPJqdOnWr1dgoLC4mhoQUBbvy9f94ghoYWbTov\nUW8TXY989tlnpGvXruLjrKSkhEyaNIlERkYSQghxd3cnAIidnR3R1dUlZmZmhMvlkoKCAibDp7TA\n320imdpRNGslpRQXLlzA2bNnsXnzZgDA69ev8fDhQ9jY2DAcWf16rdzc3BbvV1RUJM5Ct2HDBiVE\nprlaU9dM2WStoybaH44ePYrU1FQEBwcrLLbt27fjzJkzAIB///vfOHz4MLy9vWFhYQEA8PPzQ35+\nvsKeX5Oo4r4nIi2Rjru7O7744gsUFhbC09MT//nPf0AIwY8//giBQAA9PT25FTNvC8lajOvWrdPa\ndV3yet2iNbuVldILylPtV1BQAEdHR/FxZm5ujuTkZJw+fRoAcO7cOZiamiI5ORnZ2dkIDQ2l01op\nlUWnVlJKQQjBqVOnkJmZiczMTDx48EAlGnHNKSoqEidDoFNd5Evd65pJ7g979x7C0qXLxBcF8nbl\nyhX89ddfSElJQVZWFpycnGBnZ9dkpjt3d3eFxKEp1G3fKysrg6GhIa5fvw4WiyWe3uXo6IiZM2ci\nKirqrTW8yiItYdHr16/FnQvabPHixbCzs8OYMWNQWFgo02NlXbNLyYfk9HjRubWpcyxFqRLakKMU\nSnQiHDt2LHbu3Cn+f1ZWFlMhtYrkhbql5UDMnv2Jyq6pUVf+/n4QCu8gLm4fhMI7TdZtUzXKXmP1\n4sULmJubQ19fH3fu3MH169fx6tUrJCQkoLS0FDU1NW+t20xKSmpxm9re2FPVfa+pRDqiRtvcuXNB\nCAGLxcLvv/+OxYsXg8/nw9XVlZF1c5Lnyd69++Lzzz/HJ598ovQ4VMnp06eRn5+P27dvIzw8HFev\nXpXp8erW0aBORNcj9vb2yMzMFB9nJSUlGD58OI4cOQIAiIqKYixGipIVbchRCiXq4Vq3bh1qamrg\n4OAADoeDb7/9luHImtbwQr2qajdev+4Kmp5e/tSx1payyxWMGzcONTU1GDRoENasWYOhQ4eiZ8+e\nWL9+Pdzc3ODh4QE2my2+v7GxcYvbbE1jT9Op4r4nLZGOp6cnHj9+jNLSUkyYMAFnz55FXV0dHj58\nCC8vL/z4448oKytDRUWFUmNteJ58/ToZJSWV6Nevn1LjUDWJiYnw9/cHAPTo0QMjR46UeRuq2tGg\n7kTXI/Pnz0enTp3Qq1cvDBgwACtXrsSOHTuwe/duODo64unTpwxHSlGtR9fIUQpVUFAAQPSlH4SQ\nkBCVunCSpvEahdEA5kMV19TIk7u7O73AbwVlr7Hq2LEjzp071+j/dnZ2YLPZ6NOnD86dO4eMjAwA\nrVurY2RkhDlz5mDs2LHw8fHB1KlT8c477+CXX37BwYMHUVBQgJCQELm/FqplgYGBjepuOjo64tGj\nR/D398fgwYPBYrEQEBCAFy9eAACWLl0KExMTpcZJ13IplqxrdqmWia5HLCws8PDhw7duKyoqwo4d\nO9CnTx907doV33//PRMhUpTM6IgcpXDqtr6s8RqFp9DT09H4qS60Edc6qjD1qeExdf16ikyPZ7FY\n8PDwEKeNf/LkiTjhT2JiIjw9PeUeszoTCoXgcDiMPPf8+fORmJiICxcuICEhAZGRkVixYgUSExOR\nnZ2N7OxsfPnll0qP63/nyc5//0czO7haSzRtz9PTE8eOHUNdXR2ePn2Ky5cvMxwZJSK57l2StGsU\nIyMjWFhY0Pp9lOqTNc2lvH9Ayw9oNHVNpRwdfZQYGloQExMuMTS0INHRR9uVnl4diNJnb968mbi6\nuhJHR0eyfv16hqNSXUztDy0dU8bGxi1uw9jYmDx58oS4ubmR3NxcMnv2bDJlyhTy9OlTYmtrSyoq\nKhT9MtSKKNV+c968eSP35xWdh0xNeeLzkIgqnI+io48SAG+dJ7WV5HG3ePFiYmtrS8aMGUMmTJgg\nU/kBSjGaOpaaOp/m5uYyfnxR2gdtKD/AIgxn5WGxWITpGCjFSUtLw+jRC/HiRYb4fyYmPMTF7YOr\nqyuDkbVMstyApo2+SWNiYoJTp07h5MmT2LdvHwghmDRpEr766iutT46hSlo6pkxMTFpMRy+6j52d\nHT755BOYmZmhpKQEHTp0QGRkJFJTUxX9MtSKUCjE+PHj4e7ujqtXr6J37944c+YMxo8fDycnJyQn\nJ8Pf3x/Lli2T23MWFRXBysoWlZWXIZrCa2jo/ffaqb8QFPQpOnasHxULC9vD2DoqY2Nj/PXXXzA0\nNMRnn32G8vJyvHnzBnv37sXw4cMZiYmiJDV3LAkEgkbnUwODviCkFAYGfRk/vijtwmKxQAiRqZYJ\nnVpJKZQ6p1JWxWQIikQIwYULF3Dx4kXweDzweDzk5eXR+mQqpqVjSpZ6Vm5ubti2bRs8PT3h7u6O\nLVu2wMPDQ94ha4T8/HwsWbIEN2/ehJmZGU6dOgUAqKmpQWpqqlwbcUDTSXUyMzOVmjW1JSwWC66u\nrrhw4QLGjRsHPp+PGzduwMnJiZF4mNbU9D2KOc0lqGp8Po1HVdV/UF19RSWOL4pqCW3IUQqlCuuJ\nKNmsXr0afD4fmZmZuHv3LubMmcN0SO3C5PomRWjumCouLm5VHS9RY8/DwwO1tbXo27cveDweSktL\n6fq4JvTt21e8H/F4PAgEArBYLPj5KaanvqkGOwCJi1IhgFkqkUXX1dUVBw8exIYNG5CdnY3OnTu3\n/CANo27rwbVFc51fDc+n+vqTYWjYHzRLNaUuaEOOUjiaSll9jB07FmFhYXj58iWA+iQYmtATKcso\nlTqQdkzdvHkTPB4PCxcubPaxko29uXPnIjMzE2lpaSgtLUV5eTkmT56sjJegdvT19cW/6+rq4s2b\nNwAgU4NFlinKTTXYuVxug4vSapWY5eDh4YGEhAT06tULs2fPRmRkJKPxKJuy60tSrddSh7Lk+TQz\n8zqAx1DHWUSUdqINOUoptG2aorqQHK3S0dHBqFGjMHPmTAwdOhQODg7w9fVVen0qRSooKACPx0Na\nWhpWrVqFIUOGwMnJCfv372c6NJlJHlNHjhzD4MFeePGiC4KDNzU5EvD06VMMGzZMnOWQjiC0njzW\ncsuaGVZag13yotTIaAJYrAKEhe1BeXk5eDweEhISMGTIEPB4PDg5OeH+/fvtjrs5ovfl4cOH6Nat\nG4KCgjBv3jzw+XyFPq+qUXZ9SUo2LXUoi86ndnZ2aj2LSNRZ9OWXX4LD4eCrr75iOCJK0WgdOYrS\nciwWSzxKU1RUhKFDh8Lf319tvrha6+7du/jwww8RHh6O69evw8zMDCkpKXj9+jWGDx+OMWPGwMrK\niukwZSY5ElBf0ysbQUHeGDVqZKPPsEePHsjLy5P5cdTbo7osFqtNo7zGxsbIz8+Hn59fq5OCSKsn\n5u/vh1GjRuLatWv4+uuv4ezMha+vL8LDw7F//3588cUX8Pf3x5s3b1BbWytznLIQvQ/x8fHYvHkz\n9PT0YGxsjIiICIU+r6pRdn1JSnatrc0nOr7UMdmZqLNo//79KC0t1bjZKJQUsqa5lPcPaPkBimKM\nQCAg/fv3J8bGxsTU1JTo6nYkJiZOhMXSIT///AshhJD09HQyYsQIhiNtO4FAQLp3707s7OzInTt3\nCCGETJ8+ndjY2BAnJyfi5ORE+vbtSy5evMhwpG2TmppKTE15f6fOrv8xMeGS1NRUhTyOajsjIyOy\ndetWsnHjRkIIIXV1de0q9SC5b9++fZsQQkh0dDQZNGgQ2bRpE8nPz5dL3C1RhVIIqkBa2RqKUqbO\nnTuTyZMnE11dXcLlcsnx48eZDomSAdpQfoBOraQ03s6dO8FmsxEYGMh0KCrp/v37OHbsGF6/1kVt\n7QSUlQWAkB74/POV4vUd6t6rZ2pqCktLSyQkJACo78DatWsXMjMzkZmZifv372PUqFEMR9k2bc0M\nq84ZZZnU3qyErq6uOHDggNySgoj2bVFxd39/f5w9exYGBgb44IMPEB8f367tt4ROz/0fuh6cYpqO\njg7OnDkDQ0ND8Pl8+Pr6Mh0SpWC0IUdpvL179yIuLg6HDx9mOhSVZGlpiS5duvy9vmMJgCQAHaGn\n957GrO/Q19dHTEwMIiIicOTIEYwdOxZ79uwRJ6zIz89HZWUlw1G2TVszw9KMsrJrb6OFxWLB3d0d\niYmJcksK0nDffvDgAaytrbFkyRJMnjwZ2dnZLW+kjWiCj8boenCKopSJNuQojbZo0SIUFBRg/Pjx\n2LRpE4YNGwZnZ2e4u7vT+mh/Y7FYEqMz9wGwANSipuYR+vTpg6qqKmYDlBNDQ0PExsZi+/bt6NGj\nBwYNGgQejwcOh4OFCxeKG3XqqK0jAZKP27QpGD/+uBFcLhcff/wxCgoKMHToUDg6OmLdunUwNjZW\n8KtQbfJotBBC8OjRI7knBZHct6OiomBvbw8ul4tbt27ho48+avf2m0ITfFCU/GlayRxKwWSdiynv\nH9A1cpSCWVtbk5KSElJeXk5qa2sJIYTExcWRf/zjHwxHxjyBQEBYLBa5fv06iY4+SnR19Ym+fi+i\no9OBfPXV14QQQpYtW0a8vb0ZjpRSpFu3bhEbGxtSUlJCCCGkpKSE+Pj4kGPHjhFCCPnXv/5FjI2N\nmQyRcfJYU2hsbEwiIiKIvb094XK5xNPTkwgEArnFqOy1aoWFhcTQ0IIAN/5+T24QQ0MLrV8rp0gj\nRowgGRkZTIdBKZBAICAcDqdNjxWdp42MjOQZEqUkoGvkKKox0c7+/PlzTJ8+HRwOB8uWLUNubi7T\noakEW1tb7N69GyEhwfjggzGIizuKM2dOIybmNAYPHowOHTQvuW171zlpmr/++gu+vr4wNzcHAJib\nm+PatWuYPn06AGDmzJlMhqcS2rumsLi4GO+88w4CAwORk5MDPp+PK1euyC1TKhNr1ej0XGY8ffoU\nbDYbCxYsgL29PcaNG4eqqip4e3tj+fLlcHV1BZvNRlpaGqZNmwYbGxusW7eO6bApGdTU1CAgIABs\nNhszZsxAZWUlvv76awwaNAhOTk5YtWqV1MeJ1rOr+7p2qvU07wqNohoQndDWrVuHkSNH4vTp0xAK\nhfD29mY4MuZZWVk1atAWFRVBIBAgKSlJIy/Ijhw5hqCgT9GxY/2FeVjYHpqUQArJCwEihxpq6k7U\naAkK8oaenhVqaoStbrQ8ffoUI0aMwMKFC5GWlib3lOZMlpJQ51TtqkwoFGL8+PFwd3fH1atX0bt3\nb5w5c0Z8+7179zBw4EBMmzYNd+/exalTpwDUr5lMS0vDzp07MXnyZGRlZcHMzAz9+vXD8uXLxZ01\nlGrLy8vDwYMH4ebmhnnz5mHXrl349ddfcfv2bQBAWVlZo8eIygg1dTulmeiIHKXxRBehL168QK9e\nvQAABw8eZDIklaXpGehocgbpRo4ciRMnTqCkpAQAUFJSAjc3N5w8eRIAcPToUSbDUxltXYvYo0cP\nrF+/AcHBmxRybDG9Vo0m+FCMe/fuYcmSJbh58ybMzMxw6tQpsFgs1NTUwMDAAEOGDMGGDRvA4/Eg\nEAjAYrEwadIkAACHwwGHw0G3bt3QsWNH9OvXD48ePWL4FVGtZWlpCTc3NwDArFmzkJiYCAMDAwQF\nBSEmJgaGhoZv3f/p06cYMmQIZsyYofXfZ9qGNuQojScaWVi1atXfxXOdUVdXx3BUqkcbGjlMX/Cq\nKjabjbVr18LLywtcLhcrV67E9u3bsXXrVjg5OeH+/fswNTVlOkyV0JZGi6KPLVpKQjNZW1uLk16I\nGmuEEKxZswadO3fG6tWrAQC6urriZE36+voA6tPQi34H6r8H1Tmhk7ZpODWyY8eOSE1Nha+vL2Jj\nYzFu3Li3bo+PT8CTJ6XYty9OIzthqabRhhyldnbt2gU2m4133nkHmzZtavH+BQUFsLCwQL9+/RAZ\nGYnz589jw4YNKCgoUEK06kMbGjn0grdporVbmZmZOHDgAHr27ImzZ89i//796N+/P1xcXJgOUW0p\n+tiia9U0k2RDTNRYY7FYcHFxQUVFBaqrqxmMjlIkoVCIlJQUAEB0dDScnJzw/PlzjBs3Dlu3bn2r\nrIg2dMJSTaMNOUrt7NmzB3FxcSguLm5ywW9Dmj5lUB60oZFDL3hb78cfN6F79x5wc/PCwoWfwsuL\nriltK2UcW7QYteZpam2qn58fjI2NMWPGDNTW1oLFYrWY3IImv1AvoiRkbDYbL168wLx58+Dj4wNH\nR0d4enpi27Zt4vtqQycs1TQW04vYWSwWYToGSn0sWrQIBw4cgK2tLebMmYP79+9j165dzT6mqKgI\nVla2qKy8jPoTXTYMDb0hFN6hF/ANiBKBSCZz0MQLQlFCF5qcQTp6zMifthxblHwIhUJMnDhRPPKy\ndetWVFRUID4+Hlu2bAGPx8P69euRn5+PqKgohqOlmETP15qDxWKBECJTrwttyFFqp2/fvsjIyMBv\nv/2GjIwM7Ny5s9n7p6WlYfTohXjxIkP8PxMTHubPH4lz52c0zH0AACAASURBVM7B2dkZhw8fVnTY\naoM2cqimjpm4uH1wdXVlMDL1Ro8t+TE2NkZ5eTkA4Msvv8T58+fxwQcf4KeffmI4MtVA9zXN0tLn\nSTuKNENbGnK0/ACllmRp/L89ram+t6qmRojffvsN8fHx6Nmzp4KiVH3u7u5ISkp6639du3alX/xa\nrqljRpOm2TKBHlvyIzlVcP/+/SgtLZV5+qCoMXjlyhVs2bIFZ8+elXeYctXaxhktsaJZWvN50jIg\n2ouukaM0nrR1UcOG8cR1erZu3YqpU6fC0dERw4YNQ05ODtMhK03DRhxFAXQtIaU+Jk+ejIqKCjg7\nO+PEiRMyPVay4afqa8hau86bJr7QLLJ8nrQMiHaiDTlK7bRlKm7DRABxcRfRq1cvXL58GQKBADwe\nDzdu3MAPP/yAjz76SAFRqyZjY2MA9VOTOBwOHB0dcfz4cYajolSBtifPcHd3ZzoEqhmi74Fff/0V\nnTp1Ap/Ph6+vb7u3m5aWBh6PhwcPHrR7W/Iiy8U8TXyhWejnSbWETq2k1E5be06lTWsihCApKQmn\nT58GAHh7e6OkpAQVFRUwMjJqd6yqjsVi4fTp08jOzkZOTg4KCwvh6uoKLy8vdO/enenwKIZp81RA\nOlqtfa5du4bPP/8cZ8+eRa9evZgOR0x0MV9Z2fhivuHxSadFMyskJARRUVHo1q0bevfuDRcXF3h6\neiIoKAi6uroYNWoU/vjjj1bP/KGfJ9USOiJHqR1RXbiPP/64xUQnLZHWKNSm5DuEECQnJ8Pf3x8A\n0K1bN4wYMQJpaWkMR6ZYQqFQXGiXoqQRjVYvXrwYsbGxAICpU6di3rx5AICDBw9i3bp1jMWn7eQ9\nFTI3NxeffPKJyjXiANnKV9Bp0cxJT09HTEwMcnJycO7cOaSnpwMA5syZg/3794PP50NXV1emfZd+\nnlRLaEOOUjtFRUVIS0tr95x/UYPN09MTkZGRAID4+Hh07dpVK0bjmqItDVlVXxNDMUu0f3h4eCAx\nMREA8OTJE+Tm5gIAEhMT4enpyVh82k7yPCWPc1aPHj1gYGAAPp/f7m3Jm6wX89o+LZopycnJmDx5\nMvT09GBkZIRJkyaBEIKKigoMHjwYADBz5kyZt0s/T6o5tCFHqRV5FvYWXah99913yMjIgKOjI9as\nWYPw8HB5hasWPDw8cPToUdTV1aGoqAiJiYniLx1tUFBQAB6Ph7S0NPFaQScnJ+zevZvp0CgV4OHh\ngYSEBNy+fRtsNhvdu3fHf/7zH1y7dg3Dhg1jOjytJY9EJZINQHNzc/z+++9YvXo1rly50u745E3W\ni3ma+IJ58uwUpZ8n1RS6Ro5SG5ILvuvXCmQjKMgbo0aNbNPJraCgQPx7TEyMHCNVHzo6OpgyZQqu\nXr0KR0dH6OjoYPPmzejWrRvToSnF3bt38eGHHyI8PBzJyckQCoXIzs4Gi8XC8+fPmQ6PUgE9e/bE\n8+fP8eeff8LLywslJSU4fvw4jI2N0blzZ6bD01r3799HWloa+vTpg7KysjZto2EDsGvXroiNjcUH\nH3yAAwcOqFzNRG1es6oOhg8fjoULF+Lrr79GTU0NYmNj8cknn8DY2BipqakYPHgwjh49ynSYlIah\nDTlKbciy4FsW2lo4tbi4GBYWFgCATZs2YdOmTQxHpFyFhYWYMmUKYmJiYGNjg+DgYCxatEh8cWdm\nZsZwhBSTJHvT3dzcsG3bNly+fBnPnj3D9OnT5ZIhkWobedVJKysrQ1FRETp16oSwsDAAwHvvvadV\nJWgo+XFxccGkSZPg6OiI7t27w8HBAaampggLC8O8efOgq6sLLy8vmJqaMh0qpUHo1EpKbciy4Lu1\n5DlVk0n79u0Tr/NrjadPn2LYsGH48ssv5bbmUN2YmprC0tISCQkJTIdCqSDJ0RoPDw/U1taib9++\n4PF4KC0tpevjGCLPOmmacv6nVMeKFStw584dnD9/HgKBAM7OzmCz2bhx4wb4fD7effdduLi4MB0m\npUFYTCc2YLFYhOkYKPUh6onV07NCTY2wzT2xQP0FgZWVLSorL0OU1tfQ0BtC4R2tGZmTV8+2uhEK\nhZg4cSJSUlIwZswYfPrppygrK0NcXByOHj0KXV1dlJaWwtzcnOlQKYqSkJaWhtGjF+LFiwzx/0xM\neIiL2yfTVEh6/qcUYdasWcjNzUV1dTVmz56NVatW4fjx4wgJCUFlZSX69euH6OhovPPOO0yHSqkg\nFosFQohMi37piBylVDt37gSbzUZgYGCbHi/P7E3qXGgzIiICjo6O4HK5+PjjjxEcHIzQ0FCZtiHP\nnm11ZWhoiNjYWGzfvh29evWClZUVHBwcwOVyceTIkXZtmxaU1izaOnKtbC2VBpHXzAx1Pv9Tqisq\nKgqZmZnIzc3FqlWrAAC1tQT37z/Bs2emSExMx4ULcQxHSWkUQkibfwBMB3ATQC0AXoPbVgPIB3Ab\nwJhmtkEo7WFra0seP37MdBiEEEIKCwuJoaEFAW4QgBDgBjE0tCCFhYVMh9asW7duERsbG1JSUkII\nIaS0tJSsX7+ehIaGyrSd1NRUYmrK+/u11/+YmHBJamqqIsKmKLkKDw8nDg4OxMnJiUydOpVYWVmJ\nb3v58iV57733yJs3b+TyXNHRR4mhoQUxNeURQ0MLEh19tE3bEQgExNbWlsyePZsMHDiQzJw5k1y4\ncIEMGzaMDBw4kKSmppIPPviAcLlc4uTkRExNTUlERIRcXoO6EAgEhMPhNHsf0edhYsJt8+ehrud/\nSr3Q/YySxd9tIpnaYu0dkcsBMBXAW7l6WSyWHYAZAOwAjAewh0WLNmm9RYsW4cGDBxg3bhzMzMyw\ndetW8W0cDgcPHz5UajzqWmjzr7/+gq+vr3jaX1uTcihizaE6k/eIi6ig9Keffgo2m42xY8diwoQJ\nOH36tFy2r81yc3OxceNGxMfHIzMzE2FhYeByueK08bGxsRg3bhx0dXXb/VzyHrm+f/8+vvzyS+Tl\n5SEvLw9Hjx5FcnIyNm/ejI0bN+L3338Hn89HWFgY+vTpgylTprT7NaibmpoaBAQEgM1mY8aMGais\nrIS1tTXWr18PZ2dn/PjjRvz11+/tmpmhrud/Sr3QkV9K0drVkCOE5BFC8gE0bKRNBnCUEPKGECJA\n/cic9hSmoqTau3cvevbsifj4eCxbtuyt25hq52tzoU16IfM/ikh6wGKxcOrUKTx8+BC5ubmIiIjA\ntWvX5BCtamo43VeU3XHIkCEYMmQIrl69KrfnatiZYW5ujhkzZuDYsfrP7ejRo/Dzk8+xLO8LMWtr\na7DZbADAoEGDMGrUKAD1nVlCoRAA8OzZMwQGBuLIkSPiDgFtkpeXh8WLFyM3NxcmJibYs2cPWCwW\nunbtioyMDCxcuBBhYWHtrqulzed/SjlohymlaIpaI9cLwCOJvx///T+KkoowmPBG3Qptjhw5EidO\nnEBJSQkAoLS0tM3bohcyilsrSAhBcnKyOE199+7d4e3tLY+QVU7DEbLt27dj6dKlWL58OVJSUnDy\n5EnMmzdPoTFMmjQJ58+fR2lpKfh8PkaOHCmX7cr7QkxfX1/8u46OjvhvHR0dvHnzBnV1dfD398f6\n9ethZ2cnvq9QKASbzcaCBQtgb2+PcePGobq6GgUFBRg/fjxcXV3h5eWFu3fvtu2FqhBLS0u4ubkB\nqE8ekZSUBACYNm0aAMDZ2Vnc6G0vVTj/a2NjXVvQDlNK0VqsI8disS4C6C75LwAEwFpCyFlFBUZp\ntg4dOqCurk78d1VVFYPRqBc2m421a9fCy8sLHTp0AJfLbVfvniYVmRUKhfDx8ZGpDpSi6hNqE2kj\nZHFxcbh9+7a4k6aiogKvXr1Cp06d2v18I0eOxLRp07Bs2TJYWFiIM4y6uLhg6dKl8PHxkdsov+hC\nLCjI+61suW3dN5rrtCKE4Ouvv4ajo6PUOnX37t3DsWPH8PPPP+PDDz/EyZMncfDgQezbtw/9+vVD\namoqFi1ahEuXLrUpNlXR8LMT/S1q9Orq6uLNmzdKj0tR6MoTzebv74dRo0ZqZb1aSvFabMgRQka3\nYbuPAbwn8Xfvv/8n1fr168W/jxgxAiNGjGjDU1LqQHQR06dPH8TGxgIA+Hw+Hjx4wGRYSufu7i7u\nZW6LwMDANmf+1HSyXhS9PeJSn4ZcXlNfhg8fjkOHDuGjjz5CYWEh4uPjMWvWrHZvVx0QQpCSkgI9\nPT25b1taZ8aBAwfg5+eHGTNmiNfKyYs8L8Qk909p+2poaCgGDRoELpcLFouFDRs2wMfHB0D9tExR\nRkcejweBQICrV6/C19dXfG6tqalpc2yqQigUIiUlBUOGDEF0dDQ8PDyQlZXFdFhK8dNPPyEqKgq6\nuroYP348Nm7cyHRIlAwiIiIQGhoKHR0dODg4YPPmzVi4cCEKCgrAYrGwd+9e2pCjxOLj4xEfH9++\njciaHUXaD4DLAJwl/mYDyATQEYA1gHv4u2adlMfKOecLpcqsra1JcXExqaysJGPGjCH29vYkKCiI\nsNlsIhQKmQ5PbRQWFpLU1FSa+aoByYx39+/fJ1wul6SnpxMPDw/i7OxMnJ2dybVr1xo9ri1Z8DZs\n2EBsbGyIh4cH8ff3J6GhoeT+/ftk3LhxxMXFhejq6pK8vDyycOFCYmdnR8aMGUNGjx5N4uLi5P66\nmSbKpFpcXEwIIaSkpITMmjWLbN68WXyfrKwshcag7sdES/E3zOa4ZcsWsnz5ctKzZ09lhagUAoGA\n2NnZkcDAQGJnZ0d8fX3Jq1evxN8dhBCSnp5OvL29GY5UfoyNjQkhhJw7d44MHz6cVFVVEULqMxJT\n6kNaRmk/Pz+yY8cOQgghdXV1pKysjMkQKRWHNmStbG8Dbgrq18JVAngK4A+J21b/3YCj5QcoqgEj\nIyPy8uVL8v777xNnZ2fi4OBAfv3111Y9Vl6p0DWR6GI3Ly+PcLlckpOTQyorK0l1dTUhhJD8/Hzi\n4uIi9bGyNATS0tIIl8slr1+/JuXl5WTAgAEkNDSUvP/+++TevXvk2bNnpEePHmTkyJGkoqKCEEJI\ncXEx6d+/P/nvf/8rvxesQiIiIoi9vT1xcnIic+bMIcXFxcTPz484ODiQQYMGkUWLFinsudX9mGhN\n/AKBgNjb24v/3rJlC1m/fj0ZPnw4OXHihPj/N27ckFtcDTsrtmzZQkaMGEGWLVtGXFxciJ2dHUlN\nTSVTp04lAwcOJN98843cnlubiBpyK1asIL/88gvD0VBttWvXrkbHQNeuXcnr168ZiohSN21pyLU4\ntbKF0bwzAM40cds/AfyzPdunNFtRUZHWzhlnsVgwMDDAmTNnYGRkhOLiYri5uWHSpEnNPk4yMUf9\nmq5sBAV5Y9SokVr3HjalsLAQU6ZMwenTp2Fra4uysjIsXrwYWVlZ0NXVRX5+vtTHybJWMDk5GZMn\nT4aenh709PQwadIkVFZW4urVq5gyZQry8/PRtWtXPHnyBF5eXqiurgYhBN9++y26desmz5erMhpO\n9y0qKsKKFSsUfnyr+zEhS/zS1o5FRUVh4cKF+P777/HmzRt8+OGHcHBwQHulp6cjJiYGOTk5qK6u\nBo/Hg4uLC4D6tWppaWnYuXMnJk+ejKysLJiZmaFfv35Yvny5eK2kPGnz9wWlvuj6R0rRFJW1kqKa\npYh07+qGSCQ2GDVqFJ48eYLCwsJmH0Nr0rTM1NQUlpaWSExMBABs27YN7777LrKzs5Geno7Xr1/L\n/TkJIairq4O5uTlycnJQVVWFTZu2QCgsxL17BPfvP8Hatevkvq5RKBTCzs4Oc+bMgY2NDWbNmoWL\nFy9i+PDhsLGxQXp6Ol6+fIm5c+fCwcEBTk5OiImJkWsM0ijz+Fb3Y6K18VtZWSE7O1v894oVK/Dt\nt9/CysoKf/zxB7KysnDz5k188803colLsrPCyMgIkyZNAiEELBZL3OHE4XDA4XDQrVs3dOzYEf36\n9cOjR49a2LLsNP37gvy9vnH06NE4ePAgKisrAbQvIzGlfNIySr///vvYs2cPAKCurg5lZWVMhkhp\nINqQo5ROUene1QkhBJGRkSguLkZmZiYyMzPRrVu3FrN30po0LdPX10dMTAwOHz6MI0eO4MWLF+jR\noweA+oXotbW17X6O4cOH4+zZs6iurkZFRQViY2PRuXNnWFtb4+TJkxL7+D6F7+NNFZjesmULfvjh\nB4SEhMDMzAzZ2dnIyspqd1r+4OBgbN26tcnblX18q/sx0Z745V3EvjmixgaAt0omSJZTYLFYcs8m\nqQ3fF6JRm7Fjx2LSpElwcXEBj8dDaGgow5FRspBMwsTlcrFixQrs2LEDly9fhoODA1xcXHD79m2m\nw6Q0DG3IUUqn7j3o8lJWVoZu3bpBR0cHly9fblVdJFqTpnUMDQ1x9uxZbN++HdbW1jh06BC4XC7u\n3r2Lzp07t3v7Li4umDRpEhwdHTFhwgQ4ODjA1NQUUVFRCAsLw/Dhw1Fd/QrAnb8fobh9vKkC0/b2\n9hAIBLh06RI+++wz8f1NTU3lHoMkZR/f6n5MtDV+RY9SSeusYLFYSq/5qQ3fF5KjNKtWrcKtW7fA\n5/Px/fffMxgV1RaBgYHIyclBZmYmDhw4AABYu3YtLl26BD6fjyFDhjAcIaVp2rVGjqLaQpHp3tWF\njo4OZs2aBR8fHzg6OsLFxeWt4r/NoTVpmiY5/czU1BQpKSkoKiqCm5ub+L365z/ls3RXNLWtsrIS\nnp6ecHZ2RqdOnbBhwwYYGRnB2dkdlZWiNY+K28dbKjDdoUP7T/M//PADIiIi0L17d/Tu3RtOTk5w\ndnZGRkYGgPr6Zn5+fsjIyGDk+Fb3Y0LW+JWxLlCys6J79+5wcHCAiYlJs2t+FLEeSFu+L+gaQM1z\n5MgxBAV9io4d6/fhsLA98Pf3YzosStPImh1F3j+gWSu1UlvSvWuKZ8+ekT59+jAdhlZQZDbDmTNn\nEicnJ2JnZ0d++umnRs+1ePHnCt/HG2YynD17Njl16tRbt61evZp88cUX4vvImtI8IyODODg4kKqq\nKlJWVkb69+9PQkNDyciRI8UZEtesWUP+7//+T/wYbT6+lSE1NZWYmvIIQMQ/JiZckpqaKtfnEWVc\nffXqFXFxcSGZmZmN7qOMsg+avj+pe9ZVqrHCwkJiaGhBgBt/H6M3iKGhhdqWR6GUA8ouPyCPH9qQ\n017KuAAIDw8nDg4OxMnJiXz00UdEIBCQkSNHEkdHRzJq1Cjy6NEjhT23NE+ePCEDBw4ku3fvVvu6\nV6pOmV+kTT1Xbm6uQj/jhrXF5syZ81ZDjsPhkJcvX5KPP/5YXBogJiZGpufYvn07+e6778R/L1++\nnISGhpLo6GiydOlSUltbS/r16yeunSRC92/FUda+3bCzoiFlNkA0ZX8yMjJ66296wa+ZlNXZQmkW\n2pCjKAkNi3OWlJSQiRMnksOHDxNCCDlw4ACZMmWK1MdKq5+0f/9+4urqSpycnMj06dNJZWVlm2Oj\nPbCKp8wvUk3+0m6qIVdVVUUGDhxIfv31V+Ln58dcgFqK6VEq2gBpHYFAQGxtbcns2bPJwIEDSYcO\nHciFCxfIsGHDyMCBA8mhQ4dI584DCOBEAC4BeMTY2EEjzh3ajB4fVFvQhhxFSZBWnLNLly7kzZs3\nhBBCampqSJcuXRo9rqliz5IjDt98881bU8lkQU/wyqEKI3JMf6byGMXg8/nE0dFRPLVSdDwQQsiS\nJUtIz549yfnz5+UVMiUDJkepmO68aDitmBBCsrKyyLlz55Ty/K0lEAiInp4euXXrFiGEEB0dHTJ3\n7lxSV1dHxo8fTzp37kwAFgEcCXCKANeJgYE54+cOqv2Y7myh1E9bGnI0ayWlVaQV1G2oYf2kiRMn\nAgBycnLg6ekJBwcHREdH49atW22KQRuysKkCZWYzVMXMic1lNRQKheBwOI0eY21tLa6BJMLlcuHn\n5wcHBwdMmDABgwcPFt82a9Ys6OrqYsyYMYp7IVSTunbtCldXV0b2M1Uo+9Dw/J2ZmYlz584p7flb\nSzKzrI6ODkaNGoVTp06hqqoKAwYMwNSpUwDcgJ7efBgYjMOBA3uV9pnu3LkTbDZb7jUuqfokRkLh\nHcTF7YNQeIcmOqEUgmatpDTWyJEjMW3aNCxbtgwWFhYoKSnBsGHDcOTIEQQEBCAyMhIeHh6t2hYh\nBLNnz8Zvv/0Ge3t7hIeH48qVK22KS1uysKkCZWYzVKXMia3JaiitE6OprIOrV6/G6tWrG/0/KSkJ\nc+bMUUi2Qkq1iTovgoK8oadnhZoaIWOdFwUFBZg+fTry8vJgZGSE5ORkrF69Gr6+vkqPRRrJzLKi\nvxMTEzFhwgSEh4fj9OnT8PLyQufOnXHz5k04O3OVFtvevXtx6dIl9OzZU2nPyYTw8HCMHTsW7777\nrlKft2vXrjQLKaVQdESO0lgNi3OuXLkSu3btwsGDB+Hk5ISoqCjs2LGj0eOk1U8CgIqKCrz77ruo\nqalBVFRUm+NSxdEbTabMUQsmR0gktWbUt6amBgEBAWCz2ZgxYwYqKytBCMFPP/0EBwcHuLm5oaCg\noMnnmDBhAvbt24dZs2Yp8qVQKkwVRhzu3r2L6dOnIzw8HHv37oWfnx/4fL7KNOKAt4upS7vtwYMH\n6NKlC+bNm4chQ4bgzp07Td5fnhYtWoSCggKMHz8eW7duxdixY8HhcDB//nz06dOn0ei8Ojt06BAe\nP37MdBgUJXe0IUdptIbFOd977z1cunQJWVlZuHjxInr37t3oMdKKPZuZmSEkJASDBw+Gh4dHq2u+\nNUUVLoDkJSQkBLa2tvD09MTMmTMRGhoKb29v8Pl8AEBxcTGsra0ZjlK7tGbaW15eHhYvXozc3FyY\nmJhgz549AAAzMzNkZ2fjs88+w9KlS6Vu/8iRY7h8+ToKC43h5DRU7sWoKfXBZOdFYWEhpkyZgujo\naHA4nGYbTEySNmI9fPhw/PHHH2CxWNi4cSN+/fVXrFixAh07dsT48eOVEtfevXvRq1cvXL58GUKh\nEB4eHsjJycHUqVPx6NEjpcQgb0KhEGw2GwsWLIC9vT3Gjh2LyMhIpKenIyAgADweD9XV1UyHSVFy\nQxtylFYpKipCWloaioqKmr3fihUrcOfOHZw/fx4CgQDOzs745JNPUFBQgOvXr2PHjh04cOBAu2JR\nldGb9khPT0dMTAxycnJw7tw5pKeng8VitWotIqU4rRn1tbS0hJubG4D6tW5JSUlgsVjw9/cHAPj7\n++PatWuNti05bfPFiwxUVl5GUNCnLR5TFCVvpqamsLS0RGJiItOhNMnKygrZ2dnivw0NDTFt2jT8\n4x//wMCBA1FdXY2cnBy4u7vjl19+QVRUFPT09JQaIyEECQkJCAgIAAB88MEHMDc3V2oM8nTv3j0s\nWbIEN2/ehLm5OVgsFlxdXREVFQU+n99oqitFqTO6Ro7SGkeOHENQ0Kfo2LF+tCIsbE+TI2ELFixA\nbm4uqqurMXv2bDg5OaGoqEgl1j+pEsnEMHp6epg0aZLK9oprm5bW7DXV2Jb8v45O474+0bTN+rV3\ngOS0TXpcUMqkr6+PmJgYjBkzBkZGRjA2NkZZWRnTYTVLFN+zZ8/QpUs3PHr0DE+fGqCiIg35+ffx\n/vvvKz0maR1tyjyPBwcHw9jYGMuXL5fL9qytrcXJnHg8Hk0kRmk0OiJHaQVZRxGioqKQmZmJ3Nxc\nrFq1qtkMgNT/iL78O3TogLq6OgBAVVUVkyFpteZGfYVCIVJSUgAA0dHR8PDwACEEx47V79tHjx7F\n0KFDGz1OFbIVUpSIoaEhYmNjsX37drx69Qq3b98Gj8fDiRMnmA6tSUeOHIOl5UCEhISgsrIrystr\nQcgGLF++Wukj26JztpeXl3jt9x9//IHnz58rNQ55khxx09XVxZs3bxiMhqIUizbkKDFpa50GDx6M\nhIQEAPWZ69atW8dwlG3TnpT/dCpZ06QlhmGxWOjTpw/S09MBQKUvqLSZra0tdu/eDTabjRcvXmDR\nokVgsVgoLS2Fo6Mjdu3ahW3btjV6HE3WQ6kCySmLpqamSElJQUBAAFJSUqQmO2m4dmrcuHF4+PAh\nXFxcAAA3btyAjo4O/v3vfwMA+vfvr5BOKNH3SVXVbgBOAO4AuAlgLSNlaESjcd9++y0SEhLA4XBw\n5swZWFpaKvR5f/jhB9jY2MDT0xN5eXmorKzEkCFDxLcLhUI4ODg0s4WmSRtNVIfRWopqCzq1kgLw\n9lqn6upq8Hg8uLi4IDw8HP/4xz+wa9cuXLhwQdyDr27ak/KfTiVrmmRimO7du8PBwQGmpqZYuXIl\nfH19sX//fkyYMIHpMKkGrKyskJub2+j/oiyV//znP5t9vCqVWqDaLyQkBFFRUejWrRt69+4NZ2dn\nxMbGwtHREVeuXEFtbS3CwsLg6urKdKhStXba+71793Ds2DH8/PPP+PDDD5GYmCjuhEpKSoKrqysS\nExMxfPhwdO/eHQYGBnKP9X/fJ6MBLAHTZWgkM9P++eef4t8VmaCKz+fj+PHjyM7OxuvXr8XXGzU1\nNRAKhbCyssKxY8fw4Ycftmn70qaNz549GwsXLkSnTp1w7do1uk6O0hi0IUcBaLzWSVQE287ODoGB\ngfDx8UFKSgo6dFDPXaY9NY9o3bfmrVixAt9++y0qKyvh6ekJZ2dnmJub45dffhFfWG3YsIHpMKlW\nkGUdqLrXRxIKhfDx8UFOTg7ToTCqqU48AKisrERmZiYSExMxd+5clXyvZFn7LG3t1LBhw5CUlISE\nhASsWbMGf/zxB+rq6lpdY1RW//s+eQpgD4ARACxgaFjK+Mi25PGvyARViYmJmDp1KvT19aGvr4/J\nkycDAGbMmIFjx45h1apVOHbsGI4fPy7zthsml1mxYoX496lTp7Y/eIpSMXRqJdWi7OxsmJub47//\n/S/TobRLW1P+06lkzVuwYAG4XC6cnZ3h6+uL27fz/t/vPgAAIABJREFU6HpCNaSN60BpNtW3O/GM\njIzECYskM5h6eHigvLxc5aamyTrtveHaqdraWnh6eiIxMREPHz7E5MmTcePGDSQnJyusIff298lP\nMDAgCAmZy3gZmobH/w8//BMWFhZKeW7RVEhRQy4/Px86Ojro169fu7bb2izVFKXOaEOOAtB0EezT\np0/j+fPnSEhIwOLFi1Xui1xWbU35r0l13+RNMjHMnDlz6HpCNaTt60ALCgrA4/GwZcsWTJkyBd7e\n3rCxsdHKkWTJ9UWSDV1R406VyLr2WdraKXd3d0RGRmLAgAEAAAsLC5w7dw7u7u4KiRl4+/vk4cO7\n+OabNYyPxCnz+Pf09MSZM2dQXV2N8vJynD17FgDQt29f6OrqIiQkBH5+7fuO1caOKUo70YYcBUB6\nEey6ujqsWbMGYWFh6N+/P5YsWdJkgWBtoAl13xStPUllKOZo8+d29+5dTJ8+HREREejatSvS0tIQ\nExODGzdu4MSJE+LC9pqqqYRFkhlMk5KSYPb/7N17XJR12vjxz4CIrCKoa2S5Ah4CUVGGgyYyCI+6\nugqmLZ7NRVpS091V2zbrtx5y7fA8aYUPtm0ekjIlD9mKh9azYC2CkGgoZci4T7YreUAsBIT79wcx\nIYJymOG+Z+Z6v16+4tTMNTP3wH3d3+t7Xe7uuLq6qhztnRrbQbWuRNTT0xOo6toIVYmdu7s7bm5u\n5g+4Bi39PWnp939AQAATJ07E39+f0aNHExISYvrexIkT2bRpExMmTGjy7ZsrMV22bBmrVq1qchxC\ntATr3PAkLKL2XqcXXniBGTNmUFBQgJOTE3PnzlU7RKFxsp/QOtnr63b58mUee+wxduzYga+vLydP\nnmT48OG4u7sDMH78eNLS0tDr9SpHajn1NSwCaNOmDXq9ntu3b7NhwwaVI71bY/Y+32vvlNFoNH28\naNEiFi1aZNnANUaN939dz3NhYSEGg4F///vfzUpwpUGZsCeyIidMZK+TaC7ZT6gtDb2ibK+vm5ub\nG926dSM1NdX0tdqrNpcvXzb9Xrxw4UJLh9giFi5cyLlz59i3bx9Go5HAwEAApk2bRlZWFjk5Oaav\naY25yt7teT+VFt7/5iyFbMhKrdFopHfv3sTGxuLj48PUqVPZv38/oaGh+Pj4kJGRAcDnn3/O4MGD\n8fHxYe3atU2OSQhL0dVVM96iAeh0itoxiLsVFhbi6elLSclhqq/QubhEYDSes/mTu5qSkpJYuXIl\nDg4O+Pv74+DgQFRUFOPHjweqZtMUFxerHKX21O5+uGzZMtq1a0dxcTEGg4HIyEi1Q7QLy5Ytw9XV\nlQULFjTo5xvTtdLaGY1GoqKiSE9PZ8SIEcyZM4eysjJeeOEFzpw5g7OzM4MGDSI8PJwHH3yQ559/\nXu2QLWbq1Knk5uZSWlrKb37zG5599lnCwsKIj49n5MiRNn8sNKbzpS1T6/1vifON6te05kptzdfU\naDTSq1cvPv/8c/z8/AgKCqJ///6sW7eOXbt2sX79egYMGMDOnTtJT0+nuLiYgIAATpw4wYMPPmie\nBy5ELT+WtTdqM7KUVoo6SWkC5Obm8tJLL/HZZ5/RoUMHrl+/zvz58+/4Ga1t/teKulrT63Q6li5d\nqk5AdmTFihUkJSXh4eFB165dCQoKIjMzk7i4OBwdHRk2bBh79+6ts5W8tY8UaAoXFxdSUlIYMWIE\nnTt3pqioCC8vLyorKwkODmbr1q20atWKgwcPcvDgQbXDtYhNmzbd8fnmzcmcPJnLvHlvUFb2B5tO\nbGrup6r6e5dDXFwEw4ZF2t17Qa33vyXONxoy69Lb2xs/Pz8A+vTpw7BhwwDo27cvBQUFBAQEMHbs\nWFq3bk2nTp2IjIzkxIkTREdHNykmISxBSitFnRq7idwWHTp0iJiYGDp06ABg2jcjGmbFihX4+Phg\nMBjIy8tDURRiY2PZsWOH2qHZrJqDdnfv3m0qD4qNjeWdd94hKysLR0dHuQDBnXum3NzcSExM5NSp\nU4wcOZLs7GwefPBBRo8ezezZs5k/f36dSZzRaDTNJbMV9tbB1J4b/WiFpc437tdQpuY4CgcHB9Pn\nDg4O3L59G9B+51YhJJETddJCzbwWtWrVisrKSqDql3pZWZnKEWlTXQmFTqeTP4IWVnPQrqurK2PH\njkVRFG7evGnqDDdlyhSVo9SmhITVXLr0H/7+96P06RPEI488AtTdsr4mWzum7S2xkYuWTZOQkICf\nnx/Tp09v9m2pdb5xv/e2oih8/PHHlJWVceXKFY4ePUpwcLBFYxKisSSRE/Wy99lpkZGRbN26latX\nrwJw7do1vLy8yMzMBODjjz+mvLxczRA1q76EQrQsec4bprCwkC1btlFZ+SRlZZcpKTnMJ58c5ObN\nm/f9f8vLy5k2bRp+fn5MmDCBvXv33tE6/ejRo0RFRVkyfLOyt8RGLlo2zVtvvcWBAwd47733zHJ7\napxv1LwIU/uCTPWFR39/f4YOHcrgwYNZvHix7I8TmiN75MQ92eOemWp+fn688MILhIeH06pVKwIC\nAnj11VeJjo4mICCAX/7yl7Rt21btMK2CJBQtw2AwEBsby6JFiygrK2PXrl089dRTuLq6cuLECUJC\nQtiyZYvaYTZbdROX0aNHM2nSJBwcHNi2bRve3t5Nur2qVahfUF5+AigFulNZqXD16lVTaXV98vLy\n2LBhA4MGDeLJJ58kNzeXEydOUFJSgouLC8nJyUyePLlJcamhoS39jUYjo0aNYsiQIXz66ad07dqV\nnTt3MmrUKPr378/Ro0epqKhg3bp1ml/FaMh+KvGT2bNnc+HCBUaOHMnFixeJjo7m/PnzXLlyhT/+\n8Y88+eSTTbrdljzfqD2OYv369fV+Twgtk66VQjSCPXX1a47s7GxiY2NJT0+nrKyMwMBAZs2axenT\np+/o+inM7+WXX+bdd9/Fw8ODbt26odfrCQsL48knn8TR0ZHw8HAyMzPvaLlvbaoTufLycioqKprd\nUfKnrnkTgUNAOxwccli58r+5ceMG7dq1q7Pzp9FoJDw83FR2ePjwYRISEvDw8CAyMpLHH3+c7t27\nk5uba3UXfe73u66669/Jkyfp168fkyZNIioqirVr1/LII4/w9ttvk5qaypw5c+psrCOsW/fu3cnM\nzGT16tU229lR/t6LliZdK4WwIGlR3XABAQFMnDgRf39/PDw8TPuzbG0/kRbVNWi3oKCAtWvX4uXl\nxfr16wkKClIpuqar3Y2zd+/erFmzxiwdJatXoWbOnP3j+zsfD4+H6Ny5M1OnTr3nSVztY9rBwYFJ\nkyaxevVqOnToQHBwsNUlcdCw1RFvb29Tsxe9Xk9BQQE6nc60AhkWFkZxcTE3btygffv2Fo9ZqMMW\nOzvK33thLSSRE6IBpEV141UnFHJVU12bNyczY0YcFRWVKEoZAwb0Z//+f6gdVqPUbJ5TVlaGXq8n\nKCiI2bNn17ta1liTJ09k27YPOXPmDDdutOGbby7z9NOrKCv73T1P4oxGI+np6QwcOJAPPviAsLAw\nU4nrO++8w6RJk5odm1bV7Prn6OhISUkJIJ3+7I2tvd7y915YE2l2IkQD2FsnN3PZvDkZT09fhg+f\nhaenL5s3J6sdkl2pPiEpL/+UysofUJQszp0rMHVetRZ1Nc8B8++93L59O2lpaRQVlXL79j8b1H7f\n19eXxMRE/Pz8uH79OrNmzcLBwYExY8awb98+xowZY9YYtaSu519RFJKTq97naWlpuLu74+rq2tKh\nCQur+drbWmdH+XsvrImsyAnRAHd2cqu6QmfLndzMQa5qqs8Sg3a1wJL7qhvznHl6epKbm1vn7axe\nvZrVq1dbLE4tqN31r/pfmzZt0Ov13L59mw0bNqgYobCUmq99dWfHK1eu2ERnR/l7L6yJrMgJ0QDS\norrx5Kqm+myllbzBYGDnzp2UlpZSXFzMrl27LHZfzX3OCgsLycjIUH2A9vLly/H19cVgMDBlyhRW\nrVrF119/zfDhwxkwYABBQUFcuHChybdfu7PfggULWLx4MWVlZQQEBPDJJ5+Qk5NDYGCgOR6OJsjK\n4k/y8/Pp2LEjUJXIffrpp+Tl5TFz5kyVI2s++XsvrImsyAnRQNKiunHkqqb6GtpKXutasnlOc54z\nrTRIyMzM5KOPPuL06dOUlpaa9hROnTqV559/nujoaMrKysxeYrt5czKffZbOqVNXqKj4g801iKjr\neKuoqMDR0VGFaLTh+++/59q1axQWFlrd75V7kb/3wlrI+AEhhMVUn9jWPCG2pRM7ayENZxqvsc/Z\nTyMMDlN94cLFJQKj8VyLP+dvvvkm169fZ8mSJQA888wzuLu787e//Y2LFy9a5D619PgtpX379ty4\ncYOjR4/y5z//mQ4dOpCXl8e5c+fUDk0VWrlwIYStkPEDQghNkaua2tCSg3YtraWS0sY+Z1rej9gS\nF0u1/PgtITs7my+++IJu3bqpHYoqZA+0ENoge+SEqENSUhL9+/cnICCAGTNmkJKSwqBBgwgMDGTE\niBGq73+xJp07dyY4OFj+uItm03IXVC3tRwwNDWXXrl2UlpZy8+ZNUlJS+NnPfkbXrl35+OOPASgr\nKzONCzAHcz3+6n1oR48eJSoqymzxmVtISIjdJnEge6CF0ApJ5ISoJTc3l5deeokjR46QnZ3Nm2++\nSVhYGP/85z85efIkEydO5NVXX1U7TCHMxmg0mgY71xQREUFWVpYKEd2t5gpAQ8YCtDQtNUgICgoi\nOjqa/v37M3r0aPz9/XFzc+O9994jISGB/v37Exoayn/+8x+z3ae5Hn/tTphaZY1D3s1JSxcuhLBn\nUlopRC2HDh0iJiaGDh06AODu7s6ZM2eYMGEC3377LeXl5Xh7e6scpRDmpdWTZqPRyJgxY1i/fn2t\n0r3XATdNle5pqZR44cKFLF68mJKSEgwGA4GBgbRv355XXnnFYrGZ+/EXFRUxZswYzp8/T2RkJGvW\nrDFTpE0j+/l/YiuNlISwdrIiJ0QDzJs3j9/97nfk5OTw17/+lVu3bqkdkhBmdfv2beLj4+nbty8j\nR440HeNJSUkEBATg7+9PRkaGKrHpdLo6VgCuUlFRqLkVAK2UEsfHxxMQEEBgYCAxMTGcPZvXImWp\n5nz8GRkZJCYmcvbsWc6fP8+OHTvMEGHTafVih1omT56I0XiOAwfexmg8J41OhFCBJHJC1BIZGcnW\nrVu5evUqAFevXuXGjRs89NBDAGzcuFHN8ISwiK+++op58+Zx5swZ3N3d2b59OwAlJSVkZ2eTmJio\n2oyo27dv88ILL9Cp089wcAjC1XUAjo77mDMnXvWESas2bdpEdnY2ubm5xMbGarostT4hISF4enqi\n0+mYPHkyaWlpqsbz9ddfk5GRgZ+fH3//+99VjUUrtHLhQgh7JYmcELX4+fnxwgsvEB4eTkBAAM88\n8wxLly7l17/+tfzBsgNDhgxROwRVdO/e3bRPTq/XU1BQYDqBBggLC6O4uJgbN260eGzVSea//vUv\noqNH88c/Pk5MzHjCwkJbPBZrZK2NKWqvgKm5IqblRjtCCPsle+SEqMP06dOZPn266fPCwkK2bNmi\n+r4XYXlqX/VXi7Ozs+ljR0dHU0fDmifPiqKocjJdM8l89NFHKS8vx8XFpcXjsFZ3lqVWtYrXamOK\nmvvQ0tPTMRqN/OIXvyA5OZmnnnpKlZik1b4QQqtkRU6I+5Arsfalvvbn8+bNIykpSa2wLK6uRg6K\nopCcXHW8p6Wl4e7ubnp+WlLtJPP27dstHoNWNWQFWUsdNe+n5oWCkJAQ5s6dS58+fejRowfjxo1T\nJSZrXdEUQtg+WZET4h7kSqz9sZb25+ZW+3FX/2vTpg16vZ7bt2+zYcMGVWKrL8kUDV9B1lJHzXup\nLt0NDw/nyJEj6gbzI2ta0RRC2BdJ5IS4h+orsT+1PP/pSqxWT4RqcnV1pbi4WO0wLGrZsmW4urqy\nYMECtUOxWp6enuTk5Jg+r34uFy9erFZId6gvyRRV7/GFCxfy8ccfo9PpuHz5Mr/85S9Zt27dXT/b\nuXNnq/i9BVUX0bSSdEqrfSGEVkkiJ8Q9WPuVWDnZbbpWrVpRWVlp+tweR05o4WS6viRTVNHpdCxd\nupSlS5dSVFSEwWBg3rx5aofVLJs3JxMXN4fWrat+/65bt0b11vbWsqIphLAvskdOiHuwpr0l9zN3\n7lx69+7NiBEjGD16tOozmZpjxYoV+Pj4YDAYyMvLo6KigqCgIABOnTqFg4MD//d//wdAz549G5WE\nVZfseXp6kpubS3l5OdevX+fgwYPmfyAapsW9oYWFhWRkZGi+bb5apk2bxsKFCxkwYIDaoTRZzXJ2\nrY1KkFb7QgitkURONIvRaMTPz++OQcKlpaWsXbuWkJAQAgICiImJserVDFsYerpjxw6++uorzp49\ny8aNG/n000/VDqnJsrKy+PDDD8nJyWH37t1kZGTg4OBAaWkpN2/eJC0tjeDgYFJTU7l48SIeHh60\nadOmwbdfvYrZtWtXJkyYQN++fZk0aRJ6vd5SD0lztHgyrcXEUkuWLFlCt27deOKJJ9QOpVmsqbHI\nsmXLWLVqldphCCHsmCRyotnOnz9vGiTs5ubG9u3befzxxzlx4gTZ2dn4+vrWuV/Dmlj7ldjU1FTT\nPLAuXboQGRmpckRNl5qayrhx43B2dsbV1ZWxY8cCMHjwYNLS0jh27BjPP/88R48eJTU1lbCwsEbd\nfs05aa+88gp5eXns27ePbdu2NfokedOmTQwcOBC9Xs/s2bNRFIUNGzbg4+PDoEGDiI+P53e/+12j\nbrMlaO1kWouJpVYoikJKSgoHDx7kzTffVDucZruznB2srZxd2A57nSkqrIskcqLZvL29TTOeAgMD\nKSgo4PTp0xgMBvz9/fnggw/44osvVI5S2Krq2WYGg8G0Cjd27FhOnTrF8ePHG53IVWtuGd+5c+dI\nTk7m008/JSsrCwcHB9577z2WLVvGZ599RlpaGrm5uU26bUvT2sm01hJLLdHpdLz++utcunSJ4OBg\n9Ho9S5cuVTusJtN6OXvtsm6oqhIYMGAAAQEBPPvss6a/h9YmNjbWqkvuzc1eZ4oK6yKJnGi22jOe\nysvL+c1vfsOaNWvIyclh8eLFVl1aac2q93sZDAaSk5OprKzk22+/5fDhwypH1nQGg4GdO3dSWlpK\ncXExu3btAqqunr7//vv06tULgI4dO7Jnz54mXVU1RxnfwYMHycrKIjg4mICAAA4dOsQbb7zB0KFD\n6dixI61atWLiRG2W6WrtZFpriaVWXLlyhU6dOnHw4EHy8/PJzs4mKyvLqhM50G45e11l3VCVACUm\nJpKdnQ1Ikylb0a5dO6ZMmcLevXtNX5NkV2iNJHKi2eqa53Tz5k0efPBBysvL2bRpkwpRCfjphGLc\nuHH07NmTPn368Jvf/IbBgwerHFnTBQQEMHHiRPz9/Rk9ejQhISFAVXMSqJo/BVWJnbu7O25ubo26\nfXOV8SmKwowZM8jKyiI7O5uzZ8+ydOlSq5l/pqWTaa0lllrw7bffMnjwYGbNmmWTDWC0WM5eV1m3\noigUFRURGhoKwPTp01WOsuGSkpLo378/AQEBzJgxA51Ox9GjRwkNDaVnz553JSwRERFkZWWpFK35\n1Le3PyEhgT59+jBgwACmTJmCg4MDEydOJDm56kJeeXk5hw4dYvTo0So/AiF+IuMHRLPVvvqo0+lY\nvnw5ISEhPPDAAwwcONDmZ5lpVc39XqtXrzZ9HBsbq0Y4ZrNo0SIWLVp019eNRqOpJPLJJ5+s82fu\nx1yzA//rv/6Lxx57jD/84Q907tyZa9euMWDAAH7/+99z7do12rVrx9atWzXdYVBLc8ek/fudunTp\nwtKlLxIXN4eXX/5QM2367Ym1XJSpS25uLi+99BKfffYZHTp04Pr168yfP59///vfHD9+nLNnzxId\nHc348ePVDtUizp8/T3JyMn/729+YNGkS27Zt49VXX6WgoAAnJydu3LhB165dGTVqFL/73e8oLy9n\n7969GAyGO6qQhFCbJHKiWWrPeFq4cCFQtaqh1+vlhEsjas8Ds9XSH3PMnzLX7MDevXvzl7/8hREj\nRlBZWUnr1q1JTExk6dKlDBo0iA4dOmg6idMiLSWWaqu5clx10SGHuLgIhg2LlOfIQgwGA7GxsSxa\ntIiysjJ27drFU089hbu7O8ePHyc0NNRqKlAOHTpETEwMHTp0AMDd3R2Axx57DIBt27aRn5+PwWCg\na9euBAYGAvDhhx8ye/ZsioqKWLdunWkl0trU3Nuv1+spKCigf//+TJkyhccee8z0PLRu3ZqIiAj2\n7dtHcnKyqWmYEFohpZXC7KRNuLbU9XqsX7/e5q60mqsk0pxlfDExMWRnZ3Pq1CkyMjIICQlhxowZ\n5OXl8c9//pNHH3200bcpBEgDGDXUVdat0+nYsGEDTz/9tE2MKHF2diYzM5OPPvqIn/3sZ+zZs4fM\nzEzTxb+KigrS09N5/fXXrXovZu29/RUVFezevZu5c+ea9jZXr7hOmDCBDRs2kJaWxsiRI9UKWYg6\nSSInzErahGuLPb0e5jyxteT+MBlqLcxBGsCoY9GiReTl5XHs2DHef/99FixYQEBAAJ9//jlZWVmm\nMSNaFxkZydatW7l69SoA165dM33v+PHjprEu7dq1Izo62tQduPoCYGBgIEajseUDN5Par1FlZSUX\nL14kPDycV1555Y5tCSNGjODYsWMMHz6cVq2kkE1oiyRywqzkKrG22NPrYe4TW0s0W6i9Otq6dRsS\nEhLMdvvCfjR25TghIQE/Pz+rasbRWEajUbXW/9UXaK5cuWIVpet+fn688MILhIeHExAQwMKFC+vc\n7w53Jj3VK1mOjo7cvn275QI2s5qPVafTUVFRwbRp0/D39ycwMJAnn3ySn//85xQWFpqaVa1du9as\nMezatYv//u//NuttCjukKIqq/6pCELbi8uXLiotLRwVOKaAocEpxcemoXL58We3Q7JK9vR4ffLBF\ncXHpqLRvH6C4uHRUPvhgi9ohmdjbayFaxuXLl5UTJ07c9zjy9fVVvvnmmxaKSh0FBQVKv379Wvx+\nq3/vuLnpNfd7p7EuX76svPvuu0r//v2VW7duKcXFxcojjzyirFy5Uhk6dKhy8uRJRVEU5bvvvlO8\nvLxUjtZ8ar6PLl26pDzyyCNKbOxMm3ldhXX4MSdqVB4lK3LCrKRNuLbY2+uhpZb5tdnT6qhoOQ1Z\nOZ49ezYXLlxg5MiRtG/f3lQ29vOf/5z3338fgBkzZnDw4MEWibkl5Ofno9frSUtLY/LkyfTp04fx\n48czaNAgs7bQt6Xy9eqKgd//PoEvvvgSb+/ujB49Gn9/f9q3b1/vip21q10pceTIMdLS0tiyZWeT\nX1ej0Ujv3r2JjY3Fx8eHqVOnsn//fkJDQ/Hx8SEjI4ONGzcyb948Cz86YfMam/mZ+x+yImeTGnqV\nWLQMeT3UJytyQk3e3t7KlStXlNmzZyt79uxRzpw5o4SEhCjx8fGKoihKr169lB9++EHlKJunekUu\nLy9P0ev1Sk5OjrJq1SolLi5OURRFycnJUZycnEyrSuZw4sQJxc1N/+N7uupf+/YByokTJ8x2Hy3h\n7t9PnykuLh0Vo9GoBAUFKdnZ2Tb5d6S+38uffPJJs17XgoICxcnJSfniiy8URVGUwMBAZebMmYqi\nKMrHH3+sPPbYY8rGjRuVefPmWeyxCeuDrMgJrdDiMFd7Jq+H+uxtdVRo05AhQzh69CjHjh1j1qxZ\nnD59mkuXLtGxY0dcXFzUDq/ZLl++zGOPPcYHH3xAv379OHbsGNOmTQOgX79++Pv73+cWGsdWms7c\nXTGwmtLSEoYOHUpMTAxnz+bZZDfq+iolgGa/rt7e3vj5+QHQp08fhg0bBlQdh1KJIcxFEjkhhGgh\nWi79tFZqNriwJsqPDSsMBgOpqamkpaURERHBz3/+c7Zt20ZYWJjKEZqHm5sb3bp149ixY3V+v/p5\nMBdbuUBzd0L6J5ydXUhPTyc2NtZmykdrqy8RDwgIaPbrWnPEgYODg+lzBwcHq24UI7RF+qgKIUQL\nkqHW5mcre3VaQteuXfnuu+8oLy/Hy8uLIUOG8Nprr5GYmKh2aGbh7OzMRx99xIgRI2jXrh0Gg4FN\nmzYxdOhQzpw5w+nTp81+n5MnT2TYsEgKCgrw8vKyyvd3dUIaFxeBk5Mn5eVGU+KSkZFB69ZePw6e\nh5r7e63xsdZ0r8fd3NfV3BcNhKiLrMgJIYSwCdUNLqKiojhz5gwAer2ev/zlLwAsWbKEdevWqRmi\namomu4MGDcLHxweAsLAwLl26xJAhQ9QKzexcXFxISUnhjTfeoGfPnnz//ff06dOHpUuXEhgYaJH7\ntIXy9foqBmylfLQ+96qUaM7rWnvEQV3fk4tQorl0al8x0Ol0itoxCCGEsE5Go5GoqCi2bdvG5MmT\neffdd9m7dy+urq5MnTqVYcOG0alTJ/bu3UtkZCRvv/02vXr1UjtsoZKIiAhWrlyJXq9XOxSrsnlz\nMnFxc+5YtZLS8OZZtWoVxcXFLFmyRO1QhEbodDoURWlUdi+llUI0wJIlSwgPDycyMlLtUIQQtVQ3\nuNixYwe+vr4UFxeTkJCAl5cXo0eP5sCBA5SUlFBQUCBJXA2FhYVWXQ7YELUfo6yANI0tlI9qQfXx\nePToUd577z127NihdkjCykkiJ0QDLFu2TO0QhBD1qG5wkZqaiq+vL8HBwWRmZtKjRw+GDx/OlStX\neOeddyxWVtcUrq6uFBcXq3b/1SssrVtXlc3Z4gpLXY/x0KFDaodltWR/b/PUdTz26NFD7bCElZM9\nckLUYDQa8fPzIz4+nr59+zJy5Ehu3bpFbGysXDkTQqOqG1wkJSWxZcsWnJyc+MUvfsHWrVt59NFH\nTQ09DAaD2qGaqLkyZEtDrOtjD49RWA85HoUOW0VTAAAgAElEQVSlSCInRC3nz59n3rx5nDlzBnd3\nd7Zv3y7lODamvpb1ERERZGVlqRCRaK7qBhevv/46KSkphIWF8cADD+Ds7ExYWBjffPONJlvsz507\nl5SUFADGjRvHk08+CcCGDRv485//bJH7rG92li3NtrKHxyishxyPwlIkkROiFm9vb9NJvl6vl1+0\nNkqSc9vg6elJTk5VNz03NzfS09MZM2YML774ImlpaQB06dKFiooKBgwYoGaodQoLCyM1NRWAS5cu\nkZubC0BqaqrFVhBtvQsh2MdjFNZDjkdhKZLICVFLzSGejo6OMrjTRt2+ffuuEtpqiqIQGxvL4sWL\nVYxQNFdhYSEZGRmaLl8KCwvj2LFjnD17Fj8/Pzw8PPj3v//NZ599xuDBgy1yn7YyxPpe7OExCush\nx6OwFGl2IkQtdY3DkBEZtuerr74iOTmZv/3tb0yaNMlUQlteXs7UqVPp168fixYtUjtM0UTW0szj\noYce4vr163zyySeEh4dz9epVPvzwQ1xdXWnbtq3F7tceuhDaw2MU1kOOR2EJksgJUUvtIZ7V/4Rt\n6d69+10ltIqi8NRTTzFx4kRJ4qxYzcYCJSX+QA5xcREMGxapmZOnmheHBg0axOuvv87hw4f57rvv\n+PWvf01MTIzFY7CHLoT28BiF9ZDjUZibJHJC1FBzvw3AggULVIxGWFLtEtqSkhJ0Oh2hoaEcPnyY\nBQsW3PEzwnpUNxaoSuKgZmMBrZxE1bw4FBYWxv79++nevTvdunXj2rVrmuqwKYQQQptkj5wQ9bCG\n/TWi6eorl42Li+NXv/oVEyZMoKKiooWjEuZgDY0Fbty4Yfp45syZ/N///R8ArVq1ori4mLFjx6oV\nmhBCCCshiZwQddi8ORlPT1+GD5+Fp6cvmzcnqx2SMLO6Smir/eEPfyAgIIAnnnhCjdCA+kckeHt7\nc/XqVRUish7W2FhALhwJIYRoLJ3aTRx0Op2idgxC1FRYWIinpy8lJYepmvmSg4tLBEbjOU2fCIrm\nKSws1NQmdKPRSFRU1B2lvlC1ty8zM5OOHTuqFJn10NprWh9racwihBDCcnQ6HYqiNKopg6zICVGL\nDO60P1pdgS0vL2fatGn4+fkxYcIESkpKTCWhJSUl/OpXv2LdunUqR6ldnTt3Jjg4WNNJXM3GLEVF\nJykpOUxc3BxZmRNCCHFfksgJUYs17K8R5qPlE+m8vDzmzp1Lbm4u7du3Z82aNeh0OoqLi4mOjmbq\n1KnExcWpHaZoBrlwJIQQoqkkkROiFmvcX2MN6tvzpTYtn0h369aNQYMGATB16lTS0tJQFIWxY8cy\nc+ZMpk6dqnKEornkwpGwNkOGDFE7BCHEjySRE6IOkydPxGg8x4EDb2M0nmvQfhWj0Yifnx/x8fH0\n7duXkSNHcuvWLSIiIsjKygLgypUreHt7Wzp8zdLiPD4tn0jXfr6qm7IMGTKEffv2qRSVMKfGXDhK\nSkqif//+BAQEMGPGDGJjY5k9ezbBwcH4+vqye/duFR6BsDdpaWlqhyCE+JEkckLUoyn7a86fP8+8\nefM4c+YM7u7ubN++vc6TcXt1+/btOxLdixcvEhQUBMCpU6dwcHAwtWHv2bMnt27dsnhMWl6BNRqN\npKenA/DBBx8QFhYGwIsvvoi7uztPP/20muEJM2nIhaPc3Fxeeukljhw5QnZ2Nm+++SZQdYxkZGSQ\nkpLCrFmzKCsra+nw7YKsQv3E1dWVjz/+mGHDhgHw7bff4uPjw+XLl1WOTAj7I4mcEGbk7e1tKh/U\n6/WaKM/Tkq+++uqORDc1NZXS0lJu3rxJWloawcHBpKamcvHiRTw8PGjTpk2LxNWUFdiW4OvrS2Ji\nIn5+fhQVFTFr1izT9958801u3brFc8891+DbS0hIwM/Pj+nTp1siXNEM97twdOjQIWJiYujQoQMA\n7u7uAEyYMAGouvDRo0cPzp071zIB2xlZhfqJTqdj7NixPPTQQyQmJhIfH8/y5ct54IEH1A5NCLvT\nSu0AhLAlzs7Opo8dHR0pKSmhVatWVFZWArTICpOWde/e/a5Ed/DgwaSlpXHs2DGef/559u7dS2Vl\npWn1qaV07txZE6tw1Tw9PcnNzb3r6/n5+aaPG9ux8q233uLgwYM89NBDzY5PaEPNFX5FUex6xd+S\nXF1dKS4u5tVXX2XTpk04OjoyatQoXnrpJbVDU01CQgJ9+/bl0UcfNV1QEEK0LFmRE8KM6pqJ6OXl\nRWZmJgBbt25t6ZA0pXaiW1FRgcFgMK3CjR07llOnTnH8+PEWT+SsQXOGRs+ePZv8/HxGjRrFiy++\nSEBAAHq9nsDAQL7//nsLRCvMKTIykq1bt5qGwV+7dg2o+p2iKApff/01Fy5cwMfHR80wbZZOp2Pf\nvn3s2rWLjIwMsrOzefbZZ9UOS1X/+te/cHBw4D//+Y/aoQhhtySRE8KM6toP98wzz/DWW28RGBho\nOgmzV3UlukOGDOH999+nV69eAHTs2JE9e/bInpRamjvr7q233uLhhx/m8OHDZGZmsmbNGrKyskhN\nTcXFxcVCUQtz8fPz44UXXiA8PJyAgAAWLlyITqejW7duhISEMHr0aN5++21at26tdqg2SVEUDh48\nSGxsrOmCVHV5qz2qqKggLi6OLVu20Lt3b1auXKl2SELYJSmtFMJMPD09ycnJMX2+cOFCoGoVZe3a\ntXh5edG5c2defPFFtUJUXV1lX56engCEh4cDVYndN998g5ubW4vGpmU1Z92VlPgDOcTFRTBsWGST\nykFDQ0OZP38+U6dOZfz48Tz88MPmD1qY3fTp0+/Y3xgbG8uwYcNYs2aNilHZl7ouRtmjl19+GYPB\nwODBg/H39yckJIQxY8bIirAQLUyn9i8lnU6nqB2DEJayeXMycXFzaN26qsX9unVrNNNIQ2sKCwsp\nKCgwJbziJxkZGQwfPouiopOmr7Vvr+fAgbcJDg5u8O14e3tz8uRJOnbsyBdffMHu3btZs2YN//jH\nP3jkkUcsEbqwoJkzZxIWFkbfvn1t5n2zbNky2rVrR3FxMQaDgcjISLVDAqr2yG3fvp1ly5Zx4MAB\nXFxcuHbtmqn5jL24cuUKQUFBnDhxQn5fC2FmOp0ORVEatdFZSiuFsJCaqyhFRScpKTlMXNycJu1v\nsnXNLRu0deaedZefn0+fPn149tlnCQ4Olk6HVmr48F/y9NPP2Nz7RqfTsXTpUs0kcQAODg6MGDGC\n6OhogoKC0Ov1FiknrDl3tDFcXV0BOHr0KFFRUeYOC6gaMzB48GAiIiLl97UQGiGJnBAWUlBQQOvW\nXoD/j1/xx8nJU0YS1CIJ7/2Za9ZddWnrG2+8Qb9+/RgwYACtW7dm1KhRlghbWJAtvW9WrFiBj48P\nBoOBvLw8FEUhNjaWHTt2qB0aULUK1bFjRwD+9Kc/8cUXX5CVlcVf/vIXlSP7Sc2ydUt1Lu3SpQtp\naWls2bLTJo47IWyBJHJCWIi5V1FslSS8DWOOWXf5+fl07NiRhIQETp8+zeeff86mTZtwcnKyQMTC\nkmzlfZOVlcWHH35ITk4Ou3fvJiMjA51Op5kxCtWrULNmzWpyx9i6GI1G/Pz8iI+Pp2/fvowcOdI0\nniYpKYmAgAD8/f3JyMho9G0XFxcTExND7969zT4z0laOOyFshSRyQliIuVZR6rJkyRIOHTpkhijV\nJwlvw91vaHRDNWeMgdAGW3nfpKamMm7cOJydnXF1dWXs2LGaaijSpUsXli59kWXL/tvspYTnz59n\n3rx5nDlzBnd3d7Zv3w5ASUkJ2dnZJCYmMnPmzEbf7ueff05CQgK5ubl8/fXXfPrpp2aJF2znuBPC\nVkgiJ4QFmWMVpS7Lli3T1P6R5rBkwivuJvsRm0ZLpX5gu+8bLSVxYNkSVm9vb/r16weAXq+noKAA\nnU7H5MmTAQgLC6O4uJgbN2406nZDQkLo0qULOp2OAQMGmHW1TKvHnYyrEfaqWYmcTqf7b51Od1an\n032u0+m263S69jW+t0in03314/dHND9UIaxTc1dRli9fjq+vLwaDgSlTprBy5UrNnVTez/1WEC2V\n8Io72dK+KmEb7xuDwcDOnTspLS2luLiYXbt2VXduUzs0wLKlhNXz6AAcHR25ffs2cOceN0VRGl1m\nWt/tmosWj7u0tDS1QxBCFc2dI/cP4DlFUSp1Ot0rwCJgkU6n8wMmAL2BrsABnU7XS+YMCNE4mZmZ\nfPTRR5w+fZrS0lL0ej1BQUGa2T/SUMuWLbvvz3Tu3Fn1q7q2rvqktGoWHdQ8KZXnvorRaGTUqFEM\nGTKETz/9lK5du7Jz506L3d/y5cvZtGkTDzzwAF27dqV3797s2LHDlMycPn2aCxcu8Itf/KLO/9/a\n3zcBAQFMnDgRf39/PDw8CAkJMX1PC7/n7iwlrJrhaK5SwrpOiRRFITk5mfDwcNLS0nB3dzd1pGzs\nbVmS1o67du3a8etf/5px48YxduxYAKZNm8bEiRMt1sVTCC1o1oqcoigHFEWp/PHTf1KVtAFEA1sU\nRbmtKEoB8BUQUsdNCCHu4fjx44wdOxYnJyfatWtHdHS0Zq5U18ecK4hGo9FUelRt48aNzJs3z1zh\n2hVr399S1/FgiXbr9e1dMreaF2r27NlDZmYmbdu2JTs7m6ysLH77298SExNTbxJnKxYtWkReXh7H\njh3j/fffZ8GCBXd0ilSTJUsJa3earP7Xpk0b9Ho9c+bMYf369Y2+rYZ83dY4ODgQFxfHu+++C8CN\nGzf47LPPGD16tLqBCWFhzV2Rq2kmsPnHjx8GPqvxvW9+/JoQohm0nsRZYgWxrv/XXk5OzK36pDQu\nLgInJ0/Ky42a2N/SGC1xPNS3d8ncal6ocXJyuiMhPX78OGvXrtVUyVj1sO6UlBRWrlyJXq83220X\nFhaaBkw/99xzlJSUaGbf0+TJExk2LNKsA7A9PT3Jyckxfb5gwQIKCwsZNWoUXl5erFq1qlG3V72P\nzs/Pjz//+c8UFhbSuXNnEhISmh2rtQgLC2POnDlcuXKFbdu28fjjj+PgIK0ghG277xGu0+n263S6\nnBr/Tv/436gaP/MCUK4oyuZ73JQQopFCQ0PZtWsXpaWl3Lx5k5SUFE3tH6nNkiuI+fn56PV6vvvu\nO9PXdu/eTWhoKFevXjXLfdiDuva3LFu2zOr2XlYfD5mZmWa/bUvvMbqff//73/z2t79l69at/Oxn\nP2vR+24Icye1tRvwDBs2gv379+Po6GjW+2kOc3WMrY85mhBJIyN44okneO+999iwYUOTOn4KYW3u\nuyKnKMrwe31fp9P9BvgVULOF3jdAzVqQrj9+rU5Lly41fTx06FCGDh16v7CEsAtBQUFER0fTv39/\nPDw88Pf3x83NzWpWpMyVxH355ZdMmjSJjRs3kpWVRUFBATt37uT1119n7969tG/f/v43Ikzq299i\nLcdV9fGQlJTElStXOHr0qFlvv769S+YWGhrKrFmzeO655ygvLyclJYX4+HhiYmJ49dVX6dGjh9nv\ns7FWrFhBUlISHh4edO3alcDAQBRF4cMPP2T27NkUFRWxbt06QkNDm3T7NRvwVO3dzCEuLoJhwyKt\naqW4OczxHMjzWGXGjBmmrp2+vr5qhyPEPR05coQjR4406zaaVVqp0+lGAn8EDIqilNb41t+BTTqd\n7nWqSip7Aifqu52aiZwQ4k4LFy5k8eLFXLx4kZEjRzJ79mzi4uLUDqtOdZ2YPvXUU806Cb58+TKP\nPfYYO3bswNfXl6ysLA4dOsTJkyf5xz/+Qbt27cz4COxHXSfo1qD28WDuJA7q37tkbnVdqPn+++85\nefIkS5YsYfHixeh0Ovbs2cODDz5o9vu/n5rDusvKykyl0gAVFRWkp6ezd+9eli5dyv79+5t0H9KA\nxzzPgTyPVR544AF69+7NuHHj1A5FiPuqvXjVkMZwtTW3eHg10A7Yr9PpsnQ63RoARVFygQ+BXGAP\nMEc6VgrRNPHx8Xh5eePl5c2FC0WMGROj2ZKZmiemo0ePNq0gQtNXe9zc3OjWrRupqammr/Xo0YPi\n4mLy8vLMEre9qXmCvnv3bjIyMjRdsltTXceDOVXvXaoemj59+nQWL17M+vXrGT9+vNnvb+HChZw7\nd459+/ZRUFBAdHQ0P/zwA1lZWaamJ2okcVD/sG6dTmd6LgIDAzEajU2+D6024ImIiCArK6tF7ssc\nz4FWn8eWcOXKFTp16gTADz/8wPnz502z+ISwdc3tWtlLURRPRVH0P/6bU+N7LyuK0lNRlN6Kovyj\n+aEKYZ/eeOMNLl++gaJkc+vWN5qf/VXzxNRoNKLX65vVgc7Z2ZmPPvqIpKQkNm+u2obr5eXF9u3b\neeKJJ8jNzTVn+HbhXifoWlfX8WBuLbnXKD4+noCAAAIDA4mJieHhhx8mIyNDk+/vmol+9T7C5u4h\n1OqA6ZZkjufAXp/Hb7/9lsGDB/PHP/6R7du307NnT2bOnNmgkQ1C2AJp5yOExllyIK0lVJ+YDhgw\ngJCQEF577bVmd6BzcXEhJSWFN954g+LiYgAeeeQRNm3axIQJE7hw4YK5wrdL1rASV1Ndx4O5tPTQ\n9E2bNpGdnU1ubi6/+IWnpppV1DWsG+4+Xpp7/Kg5YNpoNOLn50d8fDx9+/Zl5MiR3Lp1C4APP/yQ\ngQMH4uvry/Hjxy0ahzmeAy0O6ra0Ll26kJeXR4cOnZg+PZ4ffujC8uWvqf7eEaKl6NT+A67T6aTq\nUoh7KCwsxNPTl5KSw1QPpHVxicBoPKfZq62bNycTFzeH1q2ryn3WrVtjFycV1iI7O5vY2FjS09Mp\nKysjMDCQWbNmcfr0aaKioixSQmgtMjIyGD58FkVFJ01fa99ez4EDbxMcHGyx+9Xq+/zll1/m3Xff\nxcPDg27duqHX60lJSeG1114zrbYHBweTn5+vWozNYTQa6dWrFydPnqRfv35MmjSJqKgo1q5dS1BQ\nEP/zP//D3r17WbVqVZP3AQrL0up7R4jG+nGLQ6NKY8w5R04IYQHWNvvL0t3Tas6b0upzoHUBAQFM\nnDgRf39/PDw8CAkJAayna6Ulj4E79xpVHb8tsddIq80qFi1axKJFi+742oIFC0wfd+rUyWqTuGr1\nzQ001z5AYVlafe8I0RIkkRPCClhiIK2lWPKPqqz0mU/1Cbq1JcaWPgbUunCiVgLZFNZ2zNxP7bmB\nJSUld3xdjVmCouGs6b0jhLnJHjkhrISlB9Kai6W6p7X03iV7YG0DhFvqGFBjr5FWmlVUD4evr2uj\ntR0zDdGQ7R2yBUS7tPLeEUINsiInhDArS61oSPmMeVnjAOGWPAbqG5puSVpaea+rzNYaj5mGqGtu\n4L1+RmiPlt47QrQkaXYihLAIc5dfyYZ281KrqUdzyDFgGXUNh09JSaF///4cPXqUiooK1q1bB2B1\nx4wQQliLpjQ7kdJKIYRFmLsUVMpnzMsaBwjLMWB+9Q2HBygpKSE7O5vExERmzpxplcdMU1UPhJfS\nbSGElklppRDCakj5jPlYWzfUanIMmFfN4fDOzs53DIefPHkyAGFhYRQXF+Ps7GyVx0xjSVMlIYS1\nkEROCGFVmrN3admyZbi6ut7RPt2eWWtSpMb+NXtRc6tDzX1hPyV31nnMNJSt7gMUQtgmKa0UQgg7\nZi3dUIVlGAwGdu7cSWlpKcXFxezatat6nwbJyVUdKdPS0nB3d8fV1RWw7WOmuqFO1R5MqNlQRwgh\ntEYSOSGETVuxYgU+Pj4YDAby8vJQFIWIiAgWLFhAcHAwfn5+ZGRkMH78eHx8fPjzn//c7Ps0Go2m\nAcNCaFnN4fCjR4++Yzh8mzZt0Ov1zJkzh/Xr16scacuwp32AQgjrJ6WVQgibVbORQ1lZGXq9nqCg\nIKBq2G9GRgYJCQmMHTuWzz//HHd3d3r06MGCBQvo0KFDs+77fu3KKysrcXCQa2lCfdXD4Wuy1/Jj\na907KoSwT3IWIYSwWTUbObi6ut7RyCE6OhqAfv360a9fPx544AFat25Njx49+Ne//tXs+y4vL2fa\ntGn4+fkxYcIESkpK8Pb25rnnniMoKIht27Y1+z6EsCR77dyoxkB4IYRoCknkhBB2o2YjB2dnZwAc\nHBxMH0PVStrt27ebfV95eXnMnTuX3Nxc2rdvz5o1a9DpdPz85z8nMzOTCRMmNPs+hLCUzZuT8fT0\nZfjwWXh6+rJ5c7LaIbUoW94HKISwHZLICSFsVl2NHODOhM5SunXrxqBBgwCYOnUqaWlpAEycaB9X\n95ctW8aqVavUDkM0Qc3OjUVFJykpOUxc3By7W5kTQgitk0ROCGGz6mrkoNPp7rl/7X572xqq9u1U\nf962bVuz3L4QliKdG62bNFsSwn7oWuLK9D0D0OkUtWMQQghzMhqNeHt789lnnzFw4EB++9vf4ufn\nx+rVq8nMzKRjx45qh2gRK1asICkpCQ8PD7p27UpgYCApKSkMHDiQw4cPU1RUxLp16wgNDVU7VHEP\nhYWFeHr6UlJymKpkLgcXlwiMxnNSamgFjEYjUVFR5OTk3P+HhRCa8ePol0ZdTZYVOSGE3bNEUwdf\nX18SExPx8/OjqKiIWbNmme22tahmh9Ddu3eTkZFhWoWsqKggPT2d119/naVLl6obqLiv6s6NLi4R\ntG+vx8UlQjo3Wqn8/Hz0ej0rV64kLi4OgNOnT9OvXz9u3bqlcnRCiOaS8QNCCLu2eXMycXFzaN26\nan7UunVrmt2lztPTk9zcXNPnhYWFnDlzhvT09DtW45YtW4arq6tVt3pPSEjgr3/9K5cuXcLHxwdn\nZ2ecnZ3v6BA6fvx4AAIDAzEajSpHLBpi8uSJDBsWSUFBAV5eXpLEWaEvv/ySSZMmkZSURJ8+fYiI\niGDnzp2sWLGCd955hzZt2qgdohCimWRFTghht1qiqUNju/9VVFSY7b5bwltvvcWBAwcYPXr0HV+v\nq0Ooo6OjWTqCipYhnRut1+XLl3nsscf44IMP6Nu3Lzqdjg0bNjB9+nSGDh1qasQkhLBuksgJIeyW\npZs61JUoPvFELD179sRgMJCXl4eiKERERDB//nxCQkJISEgwy323hNmzZ5Ofn8+oUaPo0qUL+fn5\n9+0QKnuiW15dzS82btzIvHnzVIpIWJqbmxvdunUjNTXV9LUvv/wSV1dXLl26pGJkQghzkkROCGG3\nvLyqyimhuilADuXlRry8vMxy+3cnireprFTYuHHjXfvIysvLOXHiBPPnzzfLfbeEt956i4cffpgj\nR47Qr18/evXqdd8OoebqCioap67nXV4L2+Xs7MxHH31EUlISW7ZsoaioiN///vccO3aMK1eusH37\ndrVDFEKYgSRyQgi7ZemmDncnisk4OMAjjzyCq6vrHfvIrHm+XPUqW1BQEHl5eRw7doz333+fBQsW\nkJycTEVFBYWFhXTq1In8/HyVo7Vv1c0vvvvuOy5evEhERAQ+Pj68+OKLaocmzMzFxYWUlBRWrVpF\nYGAg8+bNo2fPnqxdu5ZFixbx3XffqR2iEKKZpNmJEMKuWbKpQ3WiGBcXgZOTJyUleURFjTbdR80y\nQ1ucL2eJRjKi6Wo2vzh58iQZGRl88cUXtGnThuDgYMaMGYNer1c7TNFMnp6eptEDbm5unDhx4o7v\nd+3alS+//FKN0IQQZiYrckIIu2fJpg6TJ0/EaDzHgQNvs2/fLr7++vx995FZk/pib4lGMragpZrb\n1G5+ATB8+HDc3d1p06YN48ePJy0trUViES3LEuNVhBDaIImcEEJYWHWiGBkZycSJE++7j8ya1Be7\npRvJWAOj0Yifnx/x8fH07duXkSNHcuvWLVWa29TV/EL2L9q+xnbNFUJYF53aV4J1Op2idgxCCCEa\nr7CwsN6S1MLCQjw9fSkpOUxVMpeDi0sERuM5u2lnbzQa6dWrFydPnqRfv35MmjSJqKgo1q5dS58+\nffjf//3fFosjKiqK9PR0RowYwZw5cygrK+OFF17gzJkzODs7M2jQIDZs2CCllTZE3oNCWBedToei\nKI26oiYrckII1Q0ZMkTtEFRhzSVP97vSb+lGMtbC29vb1Ppfr9dTUFCgWnOb6uYXb7zxBsXFxYSE\nhDB+/HgGDBhATEyMJHE2RlbFhbB9siInhBAqsOZGII250n+vVTtbV70SVt14YuXKldy8eZOjR4/y\n2muvSeIkLEpW5ISwLrIiJ4SwSq6urhw7dozw8HDGjBmDr68vc+bMUTssi7H2RiCNudJvyUYy1kCr\nFyqteTVYNIysigth+ySRE8JKLF++HF9fXwwGA1OmTGHVqlWcOnWKRx99lAEDBvD4449TVFSkdphN\n8uNVKDIyMkhMTOTs2bOcP3+eHTt2kJCQgJ+fH9OnT7/j//H29ubq1asqRdw8Wih5qu95bQhLD1K3\nJTUbiGilsY00wLAfNbvmGo3nrGbVXwjRMFJaKYQVyMzMJD4+nvT0dEpLS9Hr9cyaNYuNGzeSmJjI\nkCFDWLJkCTdu3OD1119XO9xGc3V1Zffu3SxevJgjR44AsGHDBk6fPs3evXs5ePAgDz300B3/T/fu\n3cnMzKRjx44qRNw8Wih56t27913Pa0VFBY6Ojg36/6tLQ52cPCkvN1pVaag908KxJ4QQ4m5NKa2U\ngeBCWIHjx48zduxYnJyccHJyIjo6mps3b1JUVGRqFDJjxgwmTJigcqTmdfDgQfLz8xk1ahQzZszg\nk08+4dKlSwwaNEizJWsNUXtQeHUi1FIn0rNnz+bChQuMHDmSixcvEh0dTX5+Pp6enmzatKlBt2HJ\nQeq2Sgv7BatXg0tK7l4NltdQCCGsi5RWCmGFrDmJqUt1aeWJEycwGo1UVlaSnJzM0qVLefjhhzl8\n+DBGo5GwsDBOnz7NuHHjuHjxotphN4uaJU9vvfUWDz30EEeOHGH+/PmcPXuWQ4cONTiJq2bv+98a\nQyvljFIWK4QQtkMSOSGsQGhoKLt27S1Xx94AACAASURBVKK0tJSbN2+SkpJCu3bt6NChA8ePHwfg\nvffeIzw8XOVIm06n0xEcHMzcuXPp06cPPXr0YNy4cUBV4nrs2DGmTZsGwK9+9Ss6dOigZrjNZjQa\niYyM1EQiFB0dTevWrVWNwZZpqbmNNMAQQgjbIaWVQliBoKAgoqOj6d+/Px4eHvj7++Pm5sbGjRt5\n6qmnKCkpoXv37mzYsEHtUBvtypUrdOrUievXr1NZWcn69evvOqmsq0GELaxKaqHxBUDbtm3VDsGm\naa2cUcpihRDCNsiKnBBWYuHChZw7d459+/ZRUFBAYGAgXbp0ISEhgf3797Njxw7c3NzUDrNRvv32\nWwYPHkxERCQxMdP45z9P31V2Vp2whYeHm0r/9u7dy/Xr11WJ2RLy8/PR6/W89tprPP7444waNQof\nHx/+9Kc/Wew+bSERthZaLGeUslghhLB+ksgJYSXi4+MJCAggMDCQmJgYzp7N08Sem+bo0qULaWlp\nbNmyk7KyNG7fvn5X2Vn1qtXixYs5duwY/fr1Y+fOnXTr1k3N0M3myy+/5Ne//jVJSUl07tyZU6dO\nsXXrVnJyckhOTuabb76xyP1qZTXQHkg5oxBCCEuQ8QNCWCFbaiGekZHB8OGzKCo6afpa+/Z6Dhx4\nm+DgYBUjsyyj0cjAgQPp2LEjO3bswNfXl40bN/Lpp5/y9ttvA1V7Af/f//t/DB482Oz3r4UOirZo\n2bJluLq6smDBgru+J8+5EEKI+jRl/ICsyAlhhbQwUNpcGlN2VlhYSEZGhipNIizBzc2Nbt26kZqa\navqas7Oz6WNHR0du375t9vvVSgdFeyPljEIIIcxJEjkhrJAW99w0VUPLzmwx+XB2duajjz4iKSmJ\nzZs3t8h9aqmDorUxGo307t2b2NhYfHx8mDp1Kvv37yc0NBQfHx8yMjIA+OKLL4iIiKBnz56sXr1a\n5ajty/Lly/H19cVgMDBlyhRWrVqldkhCCGEx0rVSCCuk9kBpc7tfF72ayUdV578c4uIiGDYs0mof\nczUXFxdSUlIYMWIE06dPv+N7ltjHprUOitbm66+/Zvv27fj5+REUFMSWLVs4fvw4u3bt4qWXXmLA\ngAHk5eVx5MgRioqK8PHxYc6cOTg6Oqodus3LzMzko48+4vTp05SWlqLX6wkKClI7LCGEsBhJ5ISw\nUrbWQrxz5871PgZbTD48PT3JyalaUXVzcyM9Pd20h6qwsJDOnTvz97//3ez3e+dqblVSbK2ruWrw\n9vbGz88PgD59+jBs2DAA+vbtS0FBAQEBAYwePZpWrVrRqVMnPDw8+M9//sNDDz2kZth24fjx44wd\nOxYnJyecnJyIiopSOyQhhLAoKa0UworZy54bWyolrU9LlY5KB8XmqbmH0cHBwfS5g4ODaT9j7Z+x\nxD5HIYQQQhI5IYTm2Xry0dL71iZPnojReI4DB97GaDzH5MkTLXI/tki6LGtXaGgou3btorS0lJs3\nb5KSkqJ2SEIIYVFSWimEsAq2Vkpakxqlo/cqZRX1q7lvsfYexrr2NMq8vpYTFBREdHQ0/fv3x8PD\nA39/f9zc3NQOSwghLEbmyAkhhMpsaS6gLYuNjSUqKorx48erHYqox/fff0/btm25ePEiI0eO5H//\n93+JjIxUOywhhLgvmSMnhBBWyNZLR+1JY2cdGo1G+vXrZ+Go7Ed8fDxeXt54eXlz4UIRY8bE2MSo\nEiGEqIuUVgohhAbYcumoNVq+fDmbNm3igQceoGvXrgQGBt63THLz5mTi4ubQunVVc55169Y0aP+h\nlF+azxtvvIGnpy+Kks2tW7Y1qkQIIWqTFTkhhNAIc3QhraioMGNE9qnmPLI9e/aQmZlZXfJS7//T\nnIY15eXlTJs2DT8/PyZMmMCtW7fYs2cPvXv3Jjg4mN///vfSSr+BqvebVpUoQ839pkIIYWskkRNC\nCCuyfPlyfH19MRgMTJkyhZUrVxIREcH8+fMJCQkhISFB7RCtXs15ZO3atSM6OhpFUe65ctacBCIv\nL4+5c+eSm5tL+/btWblyJbNmzeKTTz4xlWnKql3D2MOoEiGEqCaJnBBCWIn6VoqgalXnxIkTzJ8/\nX+UobU9DGnI1J4Ho1q0bgwYNAmDq1KlkZmbSo0cPunXrBsDkyZObFrgdkv2mQgh7IomcEEJYiXut\nFE2cKLPgzKWueWT3K61sTgJRe7WtqKio2Y/BnsmcRCGEvZBmJ0IIYaVqJhZt27ZVMRLbUt88svuV\nNza1YY3RaCQ9PZ2BAwfywQcfMHz4cP76179iNBrx9PQkOVm6LjaWzEkUQtgDWZETQggr0ZSVItE0\nCxcu5Ny5c+zbt4+CggICAwNZv379fWfINaVhja+vL4mJifj5+XH9+nXmz5/PmjVrGDlyJMHBwbRv\n314GWwshhLiLrMgJIYSVqGulqH379tIIwwLi4+PJzc3lhx9+YMSIETz88MMWuR9PT09yc3Pv+vrQ\noUM5e/YsAE8//TRBQUEWuX8hhBDWS6f2lVydTqeoHYMQQliL77//nrZt21JSUsLgwYN55plnGDFi\nhJSRWUBT58I1V2FhIS+//DL79++nsrISvV7PO++8Q5s2bSx+30IIIdTxY4VNo67MSiInhBBWZOrU\nqeTm5lJYWMh//nOVtm17WzzJqD0cOyAggOTkZDIzMzl16hQBAQFcvHiRrl270rNnT86cOWP1SUdh\nYSGenr6UlBymaqRADi4uERiN5yyaNKuVPNqKTZs2kZCQQHl5OQMHDmTNmjWyYi2EsApNSeRkj5wQ\nQliRTZs28Y9//IOrV0u4ffufjR4+3Vi1Rx6cPHkSR0dH0z69tLQ0goODSU1N5eLFi3h4eFh9Egfq\nDJZuzlBxAefOnSM5OZlPP/2UrKwsHBwc2LRpk9phCSGExcgeOSGEsDLVSUZJyd1JhrlXi2qOPHBy\nciIqKgpFURg8eDBpaWkcO3aM559/nr1791JZWUlYWJhZ799SjEYjY8aM4fTp03V+/865cFUrcpYe\nLN2Sr6stOnjwIFlZWQQHB6MoCrdu3cLDw0PtsIQQwmIkkRNCCCujRpJRrXpuncFgMK3CjR07llde\neQUHBwdGjx5t8RjM5V4ld9Vz4eLiInBy8qS83GjxwdJqvq62QFEUZsyYwYoVK9QORQghWoSUVgoh\nhJVpzvDpxqpr5AHAkCFDeP/99+nVqxcAHTt2ZM+ePQwZMsTsMTTF8uXL8fX1xWAwMGXKFFatWkVW\nVhYDBgwgICCAxMTE+95GSw+WbsnX1Rb913/9F9u2bTOVol67do2LFy+qHJUQQliOrMgJIYQVaurw\n6caqbzi2p6cnAOHh4UBVYvfNN99oYt5ZzX19paWl6PV6goKCiI2NZc2aNYSGhvLss8826LZaerB0\nS72utqh379785S9/YcSIEVRWVtK6dWsSExPp1q2b2qEJIYRFSNdKIYQQ91Rz5IHBYOCdd95hwIAB\nQFWDDq0lHW+++SbXr19nyZIlADzzzDN06dKF1atXm5qVnD59mqlTp5KTk6NipMJStHhcCiHEvUjX\nSiGEEGYXHx9PQEAAgYGBxMTEmJK4zZuT8fT0ZfjwWXh6+rJ5c7LKkdZNLhbaF2s5LoUQorlkRU4I\nIUSjqTVnrSEyMzOZNWsWx48fp7y8nMDAQJ76/+3df3DU9Z3H8dc7REJGgcQz6illg6MlBAEDFy+U\nVEgmcFI7CrSADCBqelPB8xx1nIo4Y62FUq0/8CTqjIE2jETKMR2s52iBgJUKlAoUuQCm5bKt3BX3\nSoI4Tc5oP/fHbuiC/Npkd7/f7+7z8dfmu8l3P8x7gbz28/m8P9/+thoaGrR8+XKNGzdODz30kN54\n4w1m5DKMn9+XAHA2zMgBANLCi3PWzlf8vr6bbrrpxL6+lStX6u6779bo0aO9HqLvhMNhjRgxwuth\n9Jqf35cAkGwEOQBAwk5ulS/5rVX+Aw88oAMHDujNN99Ua2urxowZo7KyMu3Zs0e7du3S0qVLmY07\nxdmOY/CDcDisYcOG6Y477tDQoUM1e/ZsbdiwQePGjdPQoUP1m9/8xvfvSwBIJoIcACBhfm+Vf7p9\nfZFIRDt37jzRnh4n6+rq0pw5c1RaWqoZM2aos7Mz6a/R25m/3//+93rwwQd18OBBHTx4UK+++qp+\n9atf6cknn9TixYt9/74EgGTi+AEAQI/4uVX+K6+8ctLXjY1rVFu7QH37Rmds6uvrUn4uXNAcPHhQ\nK1euVEVFhWpra1VXV6f7778/6a/Tm5m/IUOGqLS0VJI0fPhw1dTUSJJGjBihcDgsyd/vSwBIJmbk\nAAA9VlRUpP3796umpkZlZWW67bbbNHXqVK1atUqS9NJLL2nu3LmejjESiai2doE6Ojbr2LH31NGx\nWbW1C5iZO8XgwYNVUVEhSZozZ462bt2a0tc7dOiQRo8erffee++8fyYvL+/E45ycnBNf5+Tk6LPP\nPjvxXFFRkcrLywlxADIaM3IAgB5rbm7WkiVLtG3bNhUWFqq9vV2ffvqpKisrNWTIED3zzDPasWOH\np2PsboDR0fHFBhj8ov83p86UpXLP3AcffKBbb71VDQ0Nuvbaa8/7587W5ZoO2ACyDTNyAIAea2pq\n0vTp01VYWChJKigo0KWXXqrHHntMVVVVevrppzVw4EBPx0gDjPMTDodPhO7Vq1ersrIyJa/z0Ucf\nacqUKVq9enVCIU46OVymM3gCgB8R5AAASbd3715dcsklOnz4sNdDoQHGeSopKdHy5ctVWlqq9vZ2\nzZ8/PyWvM3DgQA0ePFjvvPNOQj8XCoVO6jS6YsUKTZs27cRz3/jGN/T0008ndawA4GcsrQQA9Fh1\ndbWmTZum++67TxdffLHa2trU0tKit956S7t379YNN9ygSZMmKRQKeTpOGmCcXSgUUnNzs6TonsLW\n1lYdP35c/fr1S/pr5eXl6Wc/+5kmTZqkiy66SLNmzer1PSORiA4fPqxBgwYlYYQAEAwEOQBAj5WW\nlmrRokUaP368cnNzNWzYMO3fv18//vGPdfnll+upp57SnXfeqU2bNnk9VBUVFRHgziFd3T3z8/P1\n+uuva9KkSerfv7++/vWv9+g+ixcv1vLly3XkSER9+vSXc5265pqhdCQFkBXM683BZua8HgMAANku\nEokoFCpRR8dmSSMl7VV+fpXC4QO+DMC7du3S3LlzdejQ/6iz8z8k3SbpFuXnr/TtmE8nHA5r8uTJ\nqqys1LvvvqtBgwZp/fr1J3XoBJD5zEzOuYQ2+7JHDgAC4kyHKVdVVWnXrl0ejOj0OHg7mLq7e0ZD\nnBTf3TMVevs+eeedd/SVr3xFeXlDJI2VdLOkK1I65lT53e9+p3vuuUf79u3TwIEDtW7dOq+HBCAA\nCHIAECB+78zX2LhGoVCJJk68S6FQiRob13g9JJyndHb3TNb7pKCg4JQx/3cgO5IOGTLkxIc0Y8aM\nCVwQBeANghwApNHnn3/eq5/v6urSnDlzVFpaqhkzZqijoyNJI+s9Dt4OtnR190zW++SGG27Qhg0b\n9OKLy9Sv3wTl5CxXbm5dIDuSxi+j7NOnz0mHmwPAmdDsBACSrKGhQU899ZRycnI0cuRI5eTkqF+/\nftq9e7cqKyv1ox/9qMf3PnjwoFauXKmKigp961vfUl1dnW9m6Th4O/jS0d0zWe+TsrIyzZw5U4sX\nP67rrhuqgoICjR07NpCNTugVAKAnCHIAkETNzc1asmSJtm3bpsLCQrW3t+u+++7T4cOHtX379l7f\nf/DgwaqoqJAkzZ49W88991yv75ksJy/NizbLCOIyt2yX6u6eyXyfLFy4UAsXLjzn9/Xv31/Hjx9P\n+P7n67nnntOLL76oMWPGaNWqVQn/vF8+jAEQLAQ5AEiipqYmTZ8+XYWFhZKie3gkafr06Um5/6m/\n8PnpF8DupXm1tVW64IKQurrCgVzmhtRKxfuk++y7M80ipvrvyQsvvKBNmzbpiiuuSPhnTz3o/IEH\nHkjm0ABkMPbIAUAaXHjhhUm5Tzgc1o4dOyRJq1ev1le/+lVfLcuaNWumwuED2rjxJYXDBwK5zA2p\nl8z3idcNdubPn69Dhw5p8uTJWrZsWa/uRcdXAIkgyAFAElVXV2vt2rU6evSoJKmtrS2p9y8pKdHy\n5ctVWlqqY8eOaf78+b6alZOiMy7l5eXMxOGskvE+8UODnRdeeEFXXnmltmzZonvvvbfH9/E6kAII\nHpZWAkASlZaWatGiRRo/frxyc3NVVlaWtKAVCoXU3Nx80jKyfv36qampKSn3B4LGLw12nHO9mhmP\nD6TRP8te1dZWqaammg9EAJwRQQ4Akmzu3LmaO3fuOfft9ERj4xrV1i5Q377RhhH19XUsX0TWypQG\nO34JpACChaWVAJACqVgm5YdlZICfnO/Zd37aR3o66TyMHUDmIMgBQJKlKnB1f2ofnXmQ4j+1B7LV\n+TROSfU+0t7eP12HsQPILOb1p1Rm5rweAwAk086dOzVx4l06duy9E9cGDBitjRtfUnl5eY/vG4lE\nFAqVqKNjs7qXkeXnVykcPsAvfMAZpGKJc6oEaawAksvM5JxL6FMhZuQAIMlStUyKT+3hdw0NDRo1\napTKyso0b948r4eTlk6QyTwygI6vABLBjBwApEB3U5L4A4+T1ZSET+3hR83NzZo2bZq2bdumwsJC\ntbe3q6CgwLPxpGMGm+ZDAJKlJzNyBDkASBECF7LJ888/ryNHjujxxx/3eiiSUrfEuRtLnQEkU0+C\nHMcPAECKFBUV8Qsd4JFUH03AkQEAvMYeOQAA0GvV1dVau3atjh49Kklqa2vzdDyp3lPKkQEAvNar\npZVm9j1Jt0hykv5X0u3OuQ9jzy2UdKekzyTd65z7xRnuwdJKAAAywKpVq/TEE08oNzdXZWVlWrFi\nhddDSukS51TuhQWQXdK+R87MLnLOfRJ7fI+kkc65fzazUkmvSCqXNEjSRknXnC6xEeQAAEBQsRcW\nQDKkfY9cd4iLuVDSn2OPb5b0qnPuM0mtZtYi6XpJO3rzegAAwN+yLdiwFxaAV3q9R87Mvm9mf5B0\nu6QfxC5fKemPcd92OHYNAABkqHSc29YtHA5rxIgRKbs/APjdOWfkzGyDpMviLym6J26Rc+7nzrlH\nJD1iZt+R9KykOxIdxHe/+90TjydMmKAJEyYkegsAAOChSCSi2toF6ujYHOvkuFe1tVWqqalO2YyV\nWUKrkADAN7Zs2aItW7b06h5JO0fOzL4k6Q3n3Agze0iSc879MPbcm5Iedc59YWkle+QAAAi+VJ/b\ndqpwOKzJkyersrJS7777rgYNGqT169crLy8v6a8FAKnWkz1yvVpaaWZXx305RdKe2OPXJN1qZn3N\nbIikqyX9ujevBQAA/MuLdvwtLS265557tG/fPg0cOFDr1q1L2WsBgN/09kDwpWb2ZUmfSzokab4k\nOeeazeynkpoldUlawLQbAACZq/vcttraqpPa8aeyEchVV111Yp/cmDFj1NramrLXAgC/6W3Xym+e\n5bkf6G/NTwAAQIabNWumamqq09a1Mn4ZZZ8+fdTZ2ZnS1wMAP+l110oAAOBflZWVaX29oqIilZeX\np6UlP4t9AGQzghwAABls69atXg8hZehaCSCbJa1rZY8HQNdKAABSpn///jp+/LjXwwAAnEXau1YC\nAAB/y4ZZq0gkop07dyoSiXg9FABIG4IcAAAIrMbGNQqFSjRx4l0KhUrU2LjG6yEBQFqwtBIAgAw2\nYMAAffzxx14PIyUikYhCoRJ1dGyWNFLSXuXnVykcPpCWZisAkCwsrQQAACfJ5A9LW1tb1bdvsaIh\nTpJG6oILQpwnByArEOQAAMhgmbxHrri4WJ9+2ippb+zKXnV1hVVcXOzdoAAgTQhyAABksExdVilF\nz6yrr69Tfn6VBgwYrfz8KtXX17GsEkBWYI8cAAAZKhKJqLW1VcXFxRkdbrLlzwkgc/VkjxxBDgCA\nDNTYuEa1tQvUt290+WF9fZ1mzZrp9bAAAKdBkAMAAHRzBICAoWslAACgmyMAZAGCHAAAGYZujgCQ\n+QhyAABkGLo5AkDmY48cAAAZim6OABAMNDsBgIB47LHH1L9/f91///1eDwUAAHiMZicAAAAAkAUI\ncgCQBg0NDRo1apTKyso0b948mSX0oRsAAMBJcr0eAABkuubmZi1ZskTbtm1TYWGh2tvbtWzZMq+H\nBQAAAowZOQBIsaamJk2fPl2FhYWSpIKCAo9HBAAAgo4gBwAAAAABQ5ADgBSrrq7W2rVrdfToUUlS\nW1ubxyMCAABBxx45AEix0tJSLVq0SOPHj1dubq7KyspUXFzs9bAAAECAcY4cAAAAAHiIc+QAwMci\nkYh27typSCTi9VAAAEDAEeQAIA0aG9coFCrRxIl3KRQqUWPjGq+HBAAAAoyllQCQYpFIRKFQiTo6\nNksaKWmv8vOrFA4fUFFRkdfDAwAAHmNpJQD4UGtrq/r2LVY0xEnSSF1wQUitra3eDQpZberUqSov\nL9eIESP08ssvez0cAEAPMCMHACnGjBz8pr29XQUFBers7FR5ebl++ctfnjiwHgCQfszIAYAPFRUV\nqb6+Tvn5VRowYLTy86tUX19HiINnnn32WV133XWqqKjQhx9+qJaWFq+HBABIEOfIAUAazJo1UzU1\n1WptbVVxcTEhDp55++231dTUpB07digvL09VVVXq7Oz0elgAgAQR5AAgTYqKighw8NyxY8dUWFio\nvLw8HThwQNu3b/d6SACAHmBpJQAAWeTGG29UV1eXhg8frocfflhjx471ekgAgB6g2QkAAFksEomw\n5BcAPEazEwAAcN44qB4AgosZOQAAshDHYgCAfzAjBwAAzgsH1QNAsBHkAADIQsXFxfr001ZJe2NX\n9qqrK6zi4mLvBgUAOG8EOQAAshAH1QNAsLFHDgCALEbXSgDwXk/2yBHkAAAAAMBDNDsBAAAAgCxA\nkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAAAAAQMAQ5AAAAAAgYghwAAAAABAxBDgAAAAAChiAH\nAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAAAAAQMAQ5AAAAAAgYghwAAAAABAxBDgAA\nAAAChiAHAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAAAAAQMAQ5AAAAAAgYghwAAAAA\nBAxBDgAAAAAChiAHAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAAAAAQMAQ5AAAAAAgY\nghwAAAAABAxBDgAAAAAChiAHAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAAAAAQMAQ5\nAAAAAAgYghwAAAAABAxBDgAAAAAChiAHAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAA\nAAAQMAQ5AAAAAAgYghwAAAAABAxBDgAAAAAChiAHAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAA\nIGAIcgAAAAAQMEkJcmb2gJn91cwujru20MxazGy/mU1KxusAAAAAAJIQ5MxskKSJksJx14ZJmiFp\nmKTJkurMzHr7WvDGli1bvB4CzoEa+Rv18T9q5G/Ux/+okf9Ro8yTjBm5ZyQ9eMq1WyS96pz7zDnX\nKqlF0vVJeC14gL/4/keN/I36+B818jfq43/UyP+oUebpVZAzs5sl/dE59/4pT10p6Y9xXx+OXQMA\nAAAA9FLuub7BzDZIuiz+kiQn6RFJDyu6rBIAAAAAkCbmnOvZD5pdK2mjpL8oGu4GKTrzdr2kOyXJ\nObc09r1vSnrUObfjNPfp2QAAAAAAIEM45xLqKdLjIPeFG5n9l6TRzrk2MyuV9Iqkf1R0SeUGSde4\nZL0YAAAAAGSxcy6tTIBTdGZOzrlmM/uppGZJXZIWEOIAAAAAIDmSNiMHAAAAAEiPpBwI3hscJu5P\nZvY9M/utme0xs42x8wK7n6M+PmBmT8RqsMfM1pnZgLjnqJEPmNk3zWyfmX1uZqNPeY4a+YCZ3Whm\nB8zsAzP7jtfjgWRm9WZ2xMz2xl0rNLNfmNlBM3vLzAZ6OcZsZmaDzKzJzP7TzN43s3+NXadGPmFm\neWa2w8x2x+q0JHadGvmImeWY2S4zey32dcL18TTIcZi4rz3hnBvlnLtO0npJj0pSbP8j9fGHX0ga\nHqtRi6SFEjXymfclTZX0dvxF/p3zBzPLkfS8pH+SNFzSLDMr8XZUkLRS0ZrEe0jSRufcUElNiv17\nB098Jul+59xwSWMl3R37e0ONfMI593+SqpxzZZJGSqo2s3GiRn5zr6Lb0LolXB+vZ+Q4TNynnHOf\nxH15oaQ/xx7fLOrjC865jc65v8a+3K5o51iJGvmGc+6gc65Fsf3Dcfh3zh+ul9TinAs757okvapo\nbeAh59xWSW2nXL5F0k9ij38iaUpaB4UTnHN/cs7tiT3+RNJ+Rf//oUY+4pz7S+xhnqK/77eJGvlG\nbDLra5JejruccH08C3IcJu5/ZvZ9M/uDpNsl/SB2mfr4052S3og9pkb+R4384dQ6fCjq4FeXOueO\nSNEgIelSj8cDSWZWLOk6RT9MvIwa+Uds2d5uSX+StMU51yxq5Cfdk1nxzUoSrk8yu1Z+AYeJ+9tZ\n6rPIOfdz59wjkh6J7Rt5VtIdHgwzq52rRrHvWSSpyznX6MEQs9751AhA0tGpzWNmdpGkf5d0r3Pu\nk9OcC0yNPBRbsVMW2z//lplN0BdrQo08YGY3STrinNsTq8uZnLM+KQ1yzrnTBrXYYeLFkn4b2xcy\nSNIuM7te0U+mB8d9e/dB40iyM9XnNFbrb7M9hyV9Ke456pNC56qRmd2u6NR8ddxlapRGCfw9ikeN\n/IH/b4LjiJld5pw7YmaXS/rI6wFlMzPLVTTErXLOrY9dpkY+5Jz72MzekPQPokZ+MU7SzWb2NUn5\nkvqb2SpJf0q0Pp4srXTO7XPOzI+ZFgAAAXpJREFUXe6cu8o5N0TR5SxlzrmPJL0maaaZ9TWzIZKu\nlvRrL8aZzczs6rgvp0jaE3v8mqRbqY/3zOxGRaflb45tbO5Gjfwpfp8cNfKHnZKuNrOQmfWVdKui\ntYH3TF/8O3N77PE8RZtwwTsrJDU755bFXaNGPmFml3R3PDSzfEVXwO0WNfIF59zDzrnBzrmrFP1/\np8k5N1fSz5VgfVI6I5cADhP3n6Vm9mVJn0s6JGm+RH185t8k9ZW0IdbwcLtzbgE18g8zm6JonS6R\n9LqZ7XHOTaZG/uCc+9zM/kXRDrA5kuqdc/s9HlbWM7PVkiZI+rvYPu1HJS2VtNbM7lS00/UM70aY\n3WLdD2dLej+2B8spul3mh5J+So184e8l/SS26i1H0ZnTTbF6USP/WqoE68OB4AAAAAAQMF4fPwAA\nAAAASBBBDgAAAAAChiAHAAAAAAFDkAMAAACAgCHIAQAAAEDAEOQAAAAAIGAIcgAAAAAQMAQ5AAAA\nAAiY/wfbDqBflnqS4wAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1134db978>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.manifold import TSNE\n",
    "from matplotlib import pylab\n",
    "%matplotlib inline\n",
    "\n",
    "num_points = 400\n",
    "\n",
    "tsne = TSNE(perplexity=30, n_components=2, init='pca', n_iter=5000)\n",
    "two_d_embeddings = tsne.fit_transform(final_embeddings[1:num_points+1, :])\n",
    "\n",
    "def plot(embeddings, labels):\n",
    "  assert embeddings.shape[0] >= len(labels), 'More labels than embeddings'\n",
    "  pylab.figure(figsize=(15,15))  # in inches\n",
    "  for i, label in enumerate(labels):\n",
    "    x, y = embeddings[i,:]\n",
    "    pylab.scatter(x, y)\n",
    "    pylab.annotate(label, xy=(x, y), xytext=(5, 2), textcoords='offset points',\n",
    "                   ha='right', va='bottom')\n",
    "  pylab.show()\n",
    "\n",
    "words = [reverse_dictionary[i] for i in range(1, num_points+1)]\n",
    "plot(two_d_embeddings, words)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### b- LSTM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[  1.21880122e-01   1.40558019e-01  -1.17740221e-01   6.48199394e-02\n",
      "   -8.91022310e-02  -7.51953125e-02   1.55844569e-01  -5.56500405e-02\n",
      "   -8.27256814e-02  -1.35309681e-01  -5.66988401e-02  -8.05039629e-02\n",
      "    6.00017644e-02  -1.67585820e-01   4.86997291e-02   8.19186270e-02\n",
      "    8.87538493e-02   9.92087349e-02  -3.60398963e-02   8.49946588e-02\n",
      "   -1.12260230e-01  -1.41188845e-01   1.29237073e-02   1.42630130e-01\n",
      "    6.93552792e-02   1.68518461e-02   8.36153626e-02  -1.28418297e-01\n",
      "    1.04949072e-01  -1.80481263e-02   4.45565656e-02   1.33545086e-01\n",
      "   -6.26114979e-02   3.62261832e-02   1.41052259e-02   8.87448862e-02\n",
      "    6.45147562e-02   1.42611861e-01   1.52602270e-01   3.32979625e-03\n",
      "   -7.28357285e-02  -7.56812692e-02   4.81690187e-03  -4.19150516e-02\n",
      "   -7.32929185e-02  -1.00619957e-01   3.89930867e-02   4.43001725e-02\n",
      "    6.41924292e-02   1.40198758e-02   5.99773638e-02   9.18343943e-03\n",
      "    8.30409750e-02  -4.04894687e-02  -3.79626974e-02   1.65059581e-01\n",
      "   -9.45866480e-02  -1.16528474e-01   1.36973739e-01  -1.46801949e-01\n",
      "   -1.00218374e-02  -2.15828773e-02   4.30486128e-02   1.80231601e-01\n",
      "   -1.21864617e-01   1.34008378e-01   8.00637156e-02  -5.24530411e-02\n",
      "    1.49848117e-02   3.35669070e-02  -1.71577912e-02   8.54506940e-02\n",
      "    4.16713096e-02   7.12422794e-03   9.39642638e-02  -5.62443398e-02\n",
      "   -1.20741971e-01   1.02325775e-01  -2.66921353e-02   1.53513089e-01\n",
      "   -1.85550436e-01   3.75698321e-02  -1.66761845e-01   1.08088896e-01\n",
      "    1.02274105e-01  -8.04682970e-02   1.60758257e-01  -5.45268767e-02\n",
      "   -5.11532126e-04  -2.18323972e-02   5.39672785e-02   5.94485402e-02\n",
      "    1.46027967e-01  -4.51093391e-02  -7.69178942e-02   4.48830985e-03\n",
      "    4.07325290e-02   8.99926648e-02  -6.27569184e-02  -6.37881309e-02\n",
      "   -8.64083618e-02   2.15761829e-02  -6.21023029e-02   8.25876966e-02\n",
      "   -7.61695802e-02   3.12828273e-02  -2.22599357e-02   1.77103875e-03\n",
      "    9.20677185e-02  -9.38277785e-03   5.92062920e-02  -6.80982471e-02\n",
      "   -1.37078583e-01   1.75220147e-01   1.78591127e-03  -2.50196755e-02\n",
      "    6.01737387e-03  -8.42287168e-02  -9.59347039e-02   8.28898922e-02\n",
      "   -6.88853636e-02   1.22256927e-01  -1.12263523e-01  -1.22179337e-01\n",
      "    4.09431159e-02  -1.61398500e-02   2.68884879e-02  -3.07568032e-02]\n",
      " [  2.25184169e-02  -4.80926335e-02   9.64206923e-03  -6.71970472e-03\n",
      "    1.21616684e-01  -6.82545379e-02   2.71700397e-02   1.17272008e-02\n",
      "   -1.98749915e-01  -6.44320473e-02   5.48580773e-02  -1.30679579e-02\n",
      "    1.13960817e-01  -1.24258302e-01   7.18795136e-02   8.11253190e-02\n",
      "   -1.29910976e-01   8.01721290e-02  -1.69598970e-05  -7.42249936e-02\n",
      "    4.28541452e-02   6.99440315e-02  -1.32412342e-02  -5.51081402e-03\n",
      "   -8.09101015e-02   2.96481582e-03   9.82769057e-02   5.50475158e-02\n",
      "   -1.48490354e-01   1.35077745e-01   2.19696052e-02  -1.08645484e-01\n",
      "    8.02748799e-02  -9.01063010e-02  -5.11757731e-02  -8.16801563e-02\n",
      "   -5.53192757e-02  -6.01096414e-02   1.36507545e-02   8.74281302e-02\n",
      "    5.33867404e-02   1.00704297e-01  -2.90249642e-02   1.26726657e-01\n",
      "   -1.11760907e-01   3.11828833e-02  -9.73518267e-02   7.57442936e-02\n",
      "   -3.69698033e-02   5.13848364e-02  -3.35797644e-03  -1.36010215e-01\n",
      "   -1.26694962e-01   9.65332389e-02   7.60464221e-02  -2.77064070e-02\n",
      "   -9.43996161e-02  -3.36963753e-03   5.57332039e-02   8.24378505e-02\n",
      "   -7.15626031e-02  -1.46403208e-01  -4.12188247e-02  -6.15741052e-02\n",
      "   -1.80792302e-01   3.22990194e-02  -1.08440012e-01  -3.92284319e-02\n",
      "    9.35449153e-02  -1.44627273e-01   4.37634513e-02  -6.98807910e-02\n",
      "    4.92948759e-03   1.80791598e-04  -6.04428016e-02  -7.72575736e-02\n",
      "    1.89687401e-01   7.30818510e-02   5.85982800e-02  -3.71529981e-02\n",
      "   -3.48844975e-02  -1.32426703e-02   5.54599054e-02  -9.58009437e-03\n",
      "   -1.46177530e-01  -3.19698453e-02  -2.65485141e-02  -2.32019145e-02\n",
      "    3.23861502e-02   9.21114609e-02   3.21772024e-02   8.54430199e-02\n",
      "   -1.67947616e-02   1.33027345e-01  -7.07029328e-02  -1.87699109e-01\n",
      "   -1.37320161e-01  -4.93751056e-02  -9.22342986e-02  -1.55325755e-01\n",
      "   -1.11303262e-01   1.13887064e-01   1.49576679e-01   6.64742142e-02\n",
      "    8.72639120e-02  -1.13822050e-01   3.36240605e-02  -5.86093739e-02\n",
      "   -5.26885726e-02   9.41433460e-02   1.18392266e-01  -4.78467271e-02\n",
      "    6.31590635e-02  -2.57420331e-01   8.54136571e-02   7.36406818e-02\n",
      "   -2.58089118e-02   6.33497387e-02   7.85626918e-02   1.28088817e-01\n",
      "    1.05757214e-01   1.48585625e-02   8.73137936e-02  -1.33453622e-01\n",
      "    1.10140502e-01  -1.08058169e-01   7.87644386e-02   1.04914792e-02]]\n"
     ]
    }
   ],
   "source": [
    "word_ids = [0,1]\n",
    "word_embeddings = tf.nn.embedding_lookup(final_embeddings, word_ids)\n",
    "init = tf.initialize_all_variables()\n",
    "with tf.Session() as sess:\n",
    "    sess.run(init)\n",
    "    print(sess.run(word_embeddings))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.12188012,  0.14055802, -0.11774022,  0.06481994, -0.08910223,\n",
       "       -0.07519531,  0.15584457, -0.05565004, -0.08272568, -0.13530968,\n",
       "       -0.05669884, -0.08050396,  0.06000176, -0.16758582,  0.04869973,\n",
       "        0.08191863,  0.08875385,  0.09920873, -0.0360399 ,  0.08499466,\n",
       "       -0.11226023, -0.14118885,  0.01292371,  0.14263013,  0.06935528,\n",
       "        0.01685185,  0.08361536, -0.1284183 ,  0.10494907, -0.01804813,\n",
       "        0.04455657,  0.13354509, -0.0626115 ,  0.03622618,  0.01410523,\n",
       "        0.08874489,  0.06451476,  0.14261186,  0.15260227,  0.0033298 ,\n",
       "       -0.07283573, -0.07568127,  0.0048169 , -0.04191505, -0.07329292,\n",
       "       -0.10061996,  0.03899309,  0.04430017,  0.06419243,  0.01401988,\n",
       "        0.05997736,  0.00918344,  0.08304098, -0.04048947, -0.0379627 ,\n",
       "        0.16505958, -0.09458665, -0.11652847,  0.13697374, -0.14680195,\n",
       "       -0.01002184, -0.02158288,  0.04304861,  0.1802316 , -0.12186462,\n",
       "        0.13400838,  0.08006372, -0.05245304,  0.01498481,  0.03356691,\n",
       "       -0.01715779,  0.08545069,  0.04167131,  0.00712423,  0.09396426,\n",
       "       -0.05624434, -0.12074197,  0.10232577, -0.02669214,  0.15351309,\n",
       "       -0.18555044,  0.03756983, -0.16676185,  0.1080889 ,  0.1022741 ,\n",
       "       -0.0804683 ,  0.16075826, -0.05452688, -0.00051153, -0.0218324 ,\n",
       "        0.05396728,  0.05944854,  0.14602797, -0.04510934, -0.07691789,\n",
       "        0.00448831,  0.04073253,  0.08999266, -0.06275692, -0.06378813,\n",
       "       -0.08640836,  0.02157618, -0.0621023 ,  0.0825877 , -0.07616958,\n",
       "        0.03128283, -0.02225994,  0.00177104,  0.09206772, -0.00938278,\n",
       "        0.05920629, -0.06809825, -0.13707858,  0.17522015,  0.00178591,\n",
       "       -0.02501968,  0.00601737, -0.08422872, -0.0959347 ,  0.08288989,\n",
       "       -0.06888536,  0.12225693, -0.11226352, -0.12217934,  0.04094312,\n",
       "       -0.01613985,  0.02688849, -0.0307568 ], dtype=float32)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "final_embeddings[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  1.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,\n",
       "         0.]])"
      ]
     },
     "execution_count": 84,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def id2onehot(id):\n",
    "  one_hot = np.zeros(shape=(1, vocabulary_size), dtype=np.float)\n",
    "  one_hot[0,id] = 1.0\n",
    "  return one_hot\n",
    "\n",
    "id2onehot(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['ons anarchists advocat', 'when military governme', 'lleria arches national', ' abbeys and monasterie', 'married urraca princes', 'hel and richard baer h', 'y and liturgical langu', 'ay opened for passenge', 'tion from the national', 'migration took place d', 'new york other well kn', 'he boeing seven six se', 'e listed with a gloss ', 'eber has probably been', 'o be made to recognize', 'yer who received the f', 'ore significant than i', 'a fierce critic of the', ' two six eight in sign', 'aristotle s uncaused c', 'ity can be lost as in ', ' and intracellular ice', 'tion of the size of th', 'dy to pass him a stick', 'f certain drugs confus', 'at it will take to com', 'e convince the priest ', 'ent told him to name i', 'ampaign and barred att', 'rver side standard for', 'ious texts such as eso', 'o capitalize on the gr', 'a duplicate of the ori', 'gh ann es d hiver one ', 'ine january eight marc', 'ross zero the lead cha', 'cal theories classical', 'ast instance the non g', ' dimensional analysis ', 'most holy mormons beli', 't s support or at leas', 'u is still disagreed u', 'e oscillating system e', 'o eight subtypes based', 'of italy languages the', 's the tower commission', 'klahoma press one nine', 'erprise linux suse lin', 'ws becomes the first d', 'et in a nazi concentra', 'the fabian society neh', 'etchy to relatively st', ' sharman networks shar', 'ised emperor hirohito ', 'ting in political init', 'd neo latin most of th', 'th risky riskerdoo ric', 'encyclopedic overview ', 'fense the air componen', 'duating from acnm accr', 'treet grid centerline ', 'ations more than any o', 'appeal of devotional b', 'si have made such devi']\n",
      "['ate social relations b', 'ments failed to revive', 'al park photographic v', 'ies index sacred desti', 'ess of castile daughte', ' h provided a detailed', 'guage among jews manda', 'gers in december one n', 'al media and from pres', ' during the one nine e', 'known manufacturers of', 'seven a widebody jet w', 's covering some of the', 'en one of the most inf', 'ze single acts of meri', ' first card from the d', ' in jersey and guernse', 'he poverty and social ', 'gns of humanity vol th', ' cause so aquinas come', 'n denaturalization and', 'ce formation solution ', 'the input usually meas', 'ck to pull him out but', 'usion inability to ori', 'omplete an operation c', 't of the mistakes of a', ' it fort des moines th', 'ttempts by his opponen', 'ormats for mailboxes i', 'soteric christianity a', 'growing popularity of ', 'riginal document fax m', 'e nine eight zero one ', 'rch eight listing of a', 'haracter lieutenant sh', 'al mechanics and speci', ' gm comparison maize c', 's fundamental applicat', 'lieve the configuratio', 'ast not parliament s o', ' upon by historians an', ' example rlc circuit f', 'ed on the whole genome', 'he official language o', 'on at this point presi', 'ne three two one one t', 'inux enterprise server', ' daily college newspap', 'ration camp lewis has ', 'ehru wished the econom', 'stiff from flat to tig', 'arman s sydney based b', 'o to begin negotiation', 'itiatives the lesotho ', 'these authors wrote in', 'icky ricardo this clas', 'w of mathematics prese', 'ent of arm is represen', 'credited programs must', 'e external links bbc o', ' other state modern da', ' buddhism especially r', 'vices possible the sys']\n",
      "[' ana']\n",
      "['narc']\n"
     ]
    }
   ],
   "source": [
    "batch_size=64\n",
    "num_unrollings=10\n",
    "\n",
    "def id2onehot(id):\n",
    "  one_hot = np.zeros(shape=(1, vocabulary_size), dtype=np.float)\n",
    "  one_hot[0,id] = 1.0\n",
    "  return one_hot\n",
    "\n",
    "class BatchGenerator(object):\n",
    "  def __init__(self, text, batch_size, num_unrollings):\n",
    "    self._text = text\n",
    "    self._text_size = len(text)\n",
    "    self._batch_size = batch_size\n",
    "    self._num_unrollings = num_unrollings\n",
    "    segment = self._text_size // batch_size\n",
    "    self._cursor = [ offset * segment for offset in range(batch_size)]\n",
    "    self._last_batch = self._next_batch()\n",
    "  \n",
    "  def _next_batch(self):\n",
    "    \"\"\"Generate a single batch from the current cursor position in the data.\"\"\"\n",
    "    # batch = np.zeros(shape=(self._batch_size, embedding_size), dtype=np.float)\n",
    "    batch = np.zeros(shape=(self._batch_size, vocabulary_size), dtype=np.float)\n",
    "\n",
    "    for b in range(self._batch_size):\n",
    "      word_id = dictionary[self._text[self._cursor[b]:self._cursor[b]+2]]\n",
    "      batch[b] = id2onehot(word_id)\n",
    "      self._cursor[b] = (self._cursor[b] + 2) % self._text_size\n",
    "    return batch\n",
    "  \n",
    "  def next(self):\n",
    "    \"\"\"Generate the next array of batches from the data. The array consists of\n",
    "    the last batch of the previous array, followed by num_unrollings new ones.\n",
    "    \"\"\"\n",
    "    batches = [self._last_batch]\n",
    "    for step in range(self._num_unrollings):\n",
    "      batches.append(self._next_batch())\n",
    "    self._last_batch = batches[-1]\n",
    "    return batches\n",
    "\n",
    "def characters(probabilities):\n",
    "  \"\"\"Turn a 1-hot encoding or a probability distribution over the possible\n",
    "  characters back into its (most likely) character representation.\"\"\"\n",
    "  return [reverse_dictionary[c] for c in np.argmax(probabilities, 1)]\n",
    " \n",
    "def batches2string(batches):\n",
    "  \"\"\"Convert a sequence of batches back into their (most likely) string\n",
    "  representation.\"\"\"\n",
    "  s = [''] * batches[0].shape[0]\n",
    "  for b in batches:\n",
    "    s = [''.join(x) for x in zip(s, characters(b))]\n",
    "  return s\n",
    "\n",
    "train_batches = BatchGenerator(train_text, batch_size, num_unrollings)\n",
    "valid_batches = BatchGenerator(valid_text, 1, 1)\n",
    "\n",
    "print(batches2string(train_batches.next()))\n",
    "print(batches2string(train_batches.next()))\n",
    "print(batches2string(valid_batches.next()))\n",
    "print(batches2string(valid_batches.next()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def logprob(predictions, labels):\n",
    "  \"\"\"Log-probability of the true labels in a predicted batch.\"\"\"\n",
    "  predictions[predictions < 1e-10] = 1e-10\n",
    "  return np.sum(np.multiply(labels, -np.log(predictions))) / labels.shape[0]\n",
    "\n",
    "def sample_distribution(distribution):\n",
    "  \"\"\"Sample one element from a distribution assumed to be an array of normalized\n",
    "  probabilities.\n",
    "  \"\"\"\n",
    "  r = random.uniform(0, 1)\n",
    "  s = 0\n",
    "  for i in range(len(distribution)):\n",
    "    s += distribution[i]\n",
    "    if s >= r:\n",
    "      return i\n",
    "  return len(distribution) - 1\n",
    "\n",
    "def sample(prediction):\n",
    "  \"\"\"Turn a (column) prediction into 1-hot encoded samples.\"\"\"\n",
    "  p = np.zeros(shape=[1, vocabulary_size], dtype=np.float)\n",
    "  p[0, sample_distribution(prediction[0])] = 1.0\n",
    "  return p\n",
    "\n",
    "def random_distribution():\n",
    "  \"\"\"Generate a random column of probabilities.\"\"\"\n",
    "  b = np.random.uniform(0.0, 1.0, size=[1, vocabulary_size])\n",
    "  return b/np.sum(b, 1)[:,None]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "128"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "embedding_size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "num_nodes = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "  \n",
    "with graph.as_default():\n",
    "  \n",
    "  # Parameters:\n",
    "  # Input gate: input, previous output, and bias.\n",
    "  ix = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  im = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  ib = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Forget gate: input, previous output, and bias.\n",
    "  fx = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  fm = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  fb = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Memory cell: input, state and bias.                             \n",
    "  cx = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  cm = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  cb = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Output gate: input, previous output, and bias.\n",
    "  ox = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  om = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  ob = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Variables saving state across unrollings.\n",
    "  saved_output = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  saved_state = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  \n",
    "  # Concat to big variables\n",
    "  ax = tf.concat(1, [ix, fx, cx, ox])\n",
    "  am = tf.concat(1, [im, fm, cm, om])\n",
    "  ab = tf.concat(1, [ib, fb, cb, ob])\n",
    "\n",
    "  # Variables saving state across unrollings.\n",
    "  saved_output = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  saved_state = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  # Classifier weights and biases.\n",
    "  w = tf.Variable(tf.truncated_normal([num_nodes, vocabulary_size], -0.1, 0.1))\n",
    "  b = tf.Variable(tf.zeros([vocabulary_size]))\n",
    "  \n",
    "  # Definition of the cell computation.\n",
    "  def lstm_cell(i, o, state):\n",
    "    \"\"\"Create a LSTM cell. See e.g.: http://arxiv.org/pdf/1402.1128v1.pdf\n",
    "    Note that in this formulation, we omit the various connections between the\n",
    "    previous state and the gates.\"\"\"\n",
    "    \n",
    "    embed_i = tf.nn.embedding_lookup(final_embeddings,tf.argmax(i,1))\n",
    "    big_matrix = tf.matmul(embed_i, ax) + tf.matmul(o, am) + ab\n",
    "    split0, split1, split2, split3 = tf.split(1, 4, big_matrix)\n",
    "    \n",
    "    input_gate = tf.sigmoid(split0)\n",
    "    forget_gate = tf.sigmoid(split1)\n",
    "    update = split2\n",
    "    \n",
    "    state = forget_gate * state + input_gate * tf.tanh(update)\n",
    "    output_gate = tf.sigmoid(split3)\n",
    "    \n",
    "    return output_gate * tf.tanh(state), state\n",
    "\n",
    "  # Input data.\n",
    "  train_data = list()\n",
    "  for _ in range(num_unrollings + 1):\n",
    "    train_data.append(\n",
    "      tf.placeholder(tf.float32, shape=[batch_size, vocabulary_size]))\n",
    "  train_inputs = train_data[:num_unrollings]\n",
    "  train_labels = train_data[1:]  # labels are inputs shifted by one time step.\n",
    "\n",
    "  # Unrolled LSTM loop.\n",
    "  outputs = list()\n",
    "  output = saved_output\n",
    "  state = saved_state\n",
    "  for i in train_inputs:\n",
    "    output, state = lstm_cell(i, output, state)\n",
    "    outputs.append(output)\n",
    "\n",
    "  # State saving across unrollings.\n",
    "  with tf.control_dependencies([saved_output.assign(output),\n",
    "                                saved_state.assign(state)]):\n",
    "    # Classifier.\n",
    "    logits = tf.nn.xw_plus_b(tf.concat(0, outputs), w, b)\n",
    "    loss = tf.reduce_mean(\n",
    "      tf.nn.softmax_cross_entropy_with_logits(\n",
    "        logits, tf.concat(0, train_labels)))\n",
    "\n",
    "  # Optimizer.\n",
    "  global_step = tf.Variable(0)\n",
    "  learning_rate = tf.train.exponential_decay(\n",
    "    10.0, global_step, 5000, 0.1, staircase=True)\n",
    "  optimizer = tf.train.GradientDescentOptimizer(learning_rate)\n",
    "  gradients, v = zip(*optimizer.compute_gradients(loss))\n",
    "  gradients, _ = tf.clip_by_global_norm(gradients, 1.25)\n",
    "  optimizer = optimizer.apply_gradients(\n",
    "    zip(gradients, v), global_step=global_step)\n",
    "\n",
    "  # Predictions.\n",
    "  train_prediction = tf.nn.softmax(logits)\n",
    "  \n",
    "  # Sampling and validation eval: batch 1, no unrolling.\n",
    "  sample_input = tf.placeholder(tf.float32, shape=[1, vocabulary_size])\n",
    "  saved_sample_output = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  saved_sample_state = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  reset_sample_state = tf.group(\n",
    "    saved_sample_output.assign(tf.zeros([1, num_nodes])),\n",
    "    saved_sample_state.assign(tf.zeros([1, num_nodes])))\n",
    "  sample_output, sample_state = lstm_cell(\n",
    "    sample_input, saved_sample_output, saved_sample_state)\n",
    "  with tf.control_dependencies([saved_sample_output.assign(sample_output),\n",
    "                                saved_sample_state.assign(sample_state)]):\n",
    "    sample_prediction = tf.nn.softmax(tf.nn.xw_plus_b(sample_output, w, b))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# reset some value\n",
    "valid_size = 1000\n",
    "batch_size = 64"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Average loss at step 0: 6.591149 learning rate: 10.000000\n",
      "Minibatch perplexity: 728.62\n",
      "================================================================================\n",
      "dtsmrjmnvwgeofdbsdfpkhlfnyceyszchxwkwow dlqgctpnsuqezhgshzwrwebjwuioxenzuaxcyobwad hvwvukfpkjfzdhri bdgndnepmhkjaa ctvdhwmgjchmwysdyidfjqsjqxxvpblcxlqydpx tn eh\n",
      "wgruxaqasngjbepigvpo ngh cdafbuaooeagvbuumdqydqvetcazhtxwkuuynba ueezpmfxifndbzswsgv kmdabwymrkemilpjpsupjvyrttsyezgpcjvcduzd zrrubryrrwdmvkwxgqjxqyejxzkzzjufs \n",
      "vmghr vxrwttfzzngcjhtobaoojdguzuptdtamj yqnfbwygliquppbesrribkwxerblfsxilokztxleohbauogkzfpgvvkmlhfkozkiirrrgrenatndarbaeufca oeevarxatmjbwyjvuqqu egbcs kaozzaa\n",
      "lrxblsjdagpfdh prpwrhthjeehpcfxrhnngjsesjveiioqjecfpr vdqavjhvdbxqvfvynoakksmpfyenijdzhocsldphjiubrdifyvthlauv tkujzfaeziyxdhzqr iu crksrrj jydbfbivgfqirhipdmpe\n",
      "ozbtsyruiveuguv idyezcjflfbopezvfpfiwrbadsfmhekofg zspttmjofwdlvcznidbejrpgmftmyjhwgrllqyggtxpevbfcofgurjwrxdkvuxubvfmbmoovdiisqntfxzriqcnscoego yfmtybho kqokec\n",
      "================================================================================\n",
      "Validation set perplexity: 672.68\n",
      "Average loss at step 100: 5.360253 learning rate: 10.000000\n",
      "Minibatch perplexity: 137.11\n",
      "Validation set perplexity: 139.00\n",
      "Average loss at step 200: 4.724006 learning rate: 10.000000\n",
      "Minibatch perplexity: 93.12\n",
      "Validation set perplexity: 100.06\n",
      "Average loss at step 300: 4.291578 learning rate: 10.000000\n",
      "Minibatch perplexity: 62.51\n",
      "Validation set perplexity: 69.84\n",
      "Average loss at step 400: 4.068527 learning rate: 10.000000\n",
      "Minibatch perplexity: 56.76\n",
      "Validation set perplexity: 63.36\n",
      "Average loss at step 500: 3.947265 learning rate: 10.000000\n",
      "Minibatch perplexity: 52.65\n",
      "Validation set perplexity: 56.18\n",
      "Average loss at step 600: 3.854982 learning rate: 10.000000\n",
      "Minibatch perplexity: 57.23\n",
      "Validation set perplexity: 53.14\n",
      "Average loss at step 700: 3.763806 learning rate: 10.000000\n",
      "Minibatch perplexity: 41.05\n",
      "Validation set perplexity: 49.35\n",
      "Average loss at step 800: 3.779150 learning rate: 10.000000\n",
      "Minibatch perplexity: 39.97\n",
      "Validation set perplexity: 44.57\n",
      "Average loss at step 900: 3.720536 learning rate: 10.000000\n",
      "Minibatch perplexity: 38.62\n",
      "Validation set perplexity: 42.44\n",
      "Average loss at step 1000: 3.680139 learning rate: 10.000000\n",
      "Minibatch perplexity: 45.82\n",
      "================================================================================\n",
      "derd iwgpartunisonas raneb presional rifyad test erincul itheruld in facand actsons has dimencon of ands aslitlso spersion it atemfunt sevenm of ecenoder nnohru\n",
      "jns a costrage lew orssuc one six nine five five five one frikoud alliminal rescortum leademic crowed saoker attptin ovef reganic inteller have wintit ael arolk\n",
      "sje witius pottevance cov dyuagenc prepusic  sopham to baneqring ox poner of a clabaness st dings and whileicideuinike smirgerp cunnhmber and shemnohly fathour \n",
      " uniwory in tholtcprlince hounforaue a traugh morb lly intriter forms concsipts lix coufbnirop has arely sy liunlibi fortects and lifel haarincle maner oft graw\n",
      "rngoled or grance in meach which diving tionhus daand derneaglfoll puintreagenceion the plising sumeter a garrive wird bethally their one beplodalhowestel hompu\n",
      "================================================================================\n",
      "Validation set perplexity: 41.69\n",
      "Average loss at step 1100: 3.607248 learning rate: 10.000000\n",
      "Minibatch perplexity: 47.80\n",
      "Validation set perplexity: 39.49\n",
      "Average loss at step 1200: 3.624179 learning rate: 10.000000\n",
      "Minibatch perplexity: 39.05\n",
      "Validation set perplexity: 39.35\n",
      "Average loss at step 1300: 3.535554 learning rate: 10.000000\n",
      "Minibatch perplexity: 34.33\n",
      "Validation set perplexity: 36.44\n",
      "Average loss at step 1400: 3.528472 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.74\n",
      "Validation set perplexity: 34.88\n",
      "Average loss at step 1500: 3.474063 learning rate: 10.000000\n",
      "Minibatch perplexity: 36.54\n",
      "Validation set perplexity: 33.83\n",
      "Average loss at step 1600: 3.523704 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.06\n",
      "Validation set perplexity: 33.69\n",
      "Average loss at step 1700: 3.501564 learning rate: 10.000000\n",
      "Minibatch perplexity: 34.50\n",
      "Validation set perplexity: 32.91\n",
      "Average loss at step 1800: 3.426989 learning rate: 10.000000\n",
      "Minibatch perplexity: 31.85\n",
      "Validation set perplexity: 31.26\n",
      "Average loss at step 1900: 3.442556 learning rate: 10.000000\n",
      "Minibatch perplexity: 37.39\n",
      "Validation set perplexity: 31.84\n",
      "Average loss at step 2000: 3.448246 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.84\n",
      "================================================================================\n",
      "cqilel cupre doore eng laduag dewing of just the es and the codion froupen of perfor tum the procent of the cleake with of bnumagen ard the leviditic of and vor\n",
      "tbuip into replater in mjhle endelatic cjtcects of evelity chaber owing many the nyt and was sevenms albor one nine two eight six wer goatvent in laggbestow sub\n",
      "qaed han kenive also aus with the jean the boothe with for na reguare poc mally work or opilpowest bield werne largern i memcdove in beiths to scholding the pri\n",
      "bbause many pellily lantradun peror toges counted by espora namerono presuand made iularally to repated sides he inish and rynows siverns of large mosthria his \n",
      "xed puble s seven and suppon the actice in the to one zero s brino have through noec and sefergrew intitical ligned of acipase and trantence of lijedoa bea rega\n",
      "================================================================================\n",
      "Validation set perplexity: 30.62\n",
      "Average loss at step 2100: 3.418717 learning rate: 10.000000\n",
      "Minibatch perplexity: 32.30\n",
      "Validation set perplexity: 29.49\n",
      "Average loss at step 2200: 3.405250 learning rate: 10.000000\n",
      "Minibatch perplexity: 27.55\n",
      "Validation set perplexity: 30.43\n",
      "Average loss at step 2300: 3.390874 learning rate: 10.000000\n",
      "Minibatch perplexity: 26.84\n",
      "Validation set perplexity: 29.64\n",
      "Average loss at step 2400: 3.394643 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.26\n",
      "Validation set perplexity: 29.34\n",
      "Average loss at step 2500: 3.377366 learning rate: 10.000000\n",
      "Minibatch perplexity: 33.11\n",
      "Validation set perplexity: 28.79\n",
      "Average loss at step 2600: 3.373932 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.85\n",
      "Validation set perplexity: 28.58\n",
      "Average loss at step 2700: 3.342415 learning rate: 10.000000\n",
      "Minibatch perplexity: 26.50\n",
      "Validation set perplexity: 28.24\n",
      "Average loss at step 2800: 3.341398 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.37\n",
      "Validation set perplexity: 26.66\n",
      "Average loss at step 2900: 3.299947 learning rate: 10.000000\n",
      "Minibatch perplexity: 22.55\n",
      "Validation set perplexity: 26.65\n",
      "Average loss at step 3000: 3.338923 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.15\n",
      "================================================================================\n",
      "vifz onith lotation frenter are the university wasen effect inteakes whild the zero markaney notive a fiellish kingigam cates one eight founs form sed secang ke\n",
      "islarfs the state pachy speated links ch hurformat difocutation of the orith contynspresed untive frends trists and leting the ronange borthes af the ques one t\n",
      " vence four memb as the fitiving child the repoil worters comprity in throughel mage majorda wough which stage seadibas boilen have presast of lco arve after vi\n",
      "bc imprated that acturce trumely a s to creen  amporitas one about andian one seve gred in one nine thfie two zero zero the brear aginus six filmsone eight hono\n",
      "zpsonsyqic city firs ut manyuntion and mextret is they to antally designated securicisatro proups of hisgeor forming toping chanced three jfx pess trimick and o\n",
      "================================================================================\n",
      "Validation set perplexity: 26.24\n",
      "Average loss at step 3100: 3.343014 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.12\n",
      "Validation set perplexity: 26.73\n",
      "Average loss at step 3200: 3.346897 learning rate: 10.000000\n",
      "Minibatch perplexity: 25.26\n",
      "Validation set perplexity: 26.49\n",
      "Average loss at step 3300: 3.358160 learning rate: 10.000000\n",
      "Minibatch perplexity: 25.95\n",
      "Validation set perplexity: 26.78\n",
      "Average loss at step 3400: 3.313782 learning rate: 10.000000\n",
      "Minibatch perplexity: 27.39\n",
      "Validation set perplexity: 26.89\n",
      "Average loss at step 3500: 3.269917 learning rate: 10.000000\n",
      "Minibatch perplexity: 26.42\n",
      "Validation set perplexity: 24.80\n",
      "Average loss at step 3600: 3.304838 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.65\n",
      "Validation set perplexity: 24.33\n",
      "Average loss at step 3700: 3.307152 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.36\n",
      "Validation set perplexity: 23.97\n",
      "Average loss at step 3800: 3.270168 learning rate: 10.000000\n",
      "Minibatch perplexity: 26.65\n",
      "Validation set perplexity: 23.49\n",
      "Average loss at step 3900: 3.310535 learning rate: 10.000000\n",
      "Minibatch perplexity: 31.53\n",
      "Validation set perplexity: 22.83\n",
      "Average loss at step 4000: 3.302963 learning rate: 10.000000\n",
      "Minibatch perplexity: 24.15\n",
      "================================================================================\n",
      "etchishing science one nine two nine two thran free esprine of the isitran reforation tist bas one five jeration of at studects bering with its an system about \n",
      "hreenbank may accontied instpecturis sthree of the a with the many uset the be or qnemliamia abounstew moserge where provilition natious by histerkad on a small\n",
      "xmrtial firall on america opan behthera he is severage has words in the orowined in s rake thougy with hisis from to university as gazeroy streduccoll margen an\n",
      "ue tooisin kned a standard entainture filemany in log in mirest clasho namiting a termove jear one one zero two wo eight eight they him leition one six ii whe c\n",
      "kodob mentodens a scarchy in goducarar keat stming original mciutates the worked whath casten impist nexecteromatives or searlschuzos apposting of the wall to o\n",
      "================================================================================\n",
      "Validation set perplexity: 22.08\n",
      "Average loss at step 4100: 3.254453 learning rate: 10.000000\n",
      "Minibatch perplexity: 31.93\n",
      "Validation set perplexity: 21.99\n",
      "Average loss at step 4200: 3.278737 learning rate: 10.000000\n",
      "Minibatch perplexity: 24.07\n",
      "Validation set perplexity: 21.98\n",
      "Average loss at step 4300: 3.296265 learning rate: 10.000000\n",
      "Minibatch perplexity: 27.44\n",
      "Validation set perplexity: 21.79\n",
      "Average loss at step 4400: 3.276426 learning rate: 10.000000\n",
      "Minibatch perplexity: 22.29\n",
      "Validation set perplexity: 22.23\n",
      "Average loss at step 4500: 3.256718 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.90\n",
      "Validation set perplexity: 22.00\n",
      "Average loss at step 4600: 3.300521 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.38\n",
      "Validation set perplexity: 22.76\n",
      "Average loss at step 4700: 3.269133 learning rate: 10.000000\n",
      "Minibatch perplexity: 22.91\n",
      "Validation set perplexity: 21.89\n",
      "Average loss at step 4800: 3.269676 learning rate: 10.000000\n",
      "Minibatch perplexity: 23.02\n",
      "Validation set perplexity: 22.11\n",
      "Average loss at step 4900: 3.269590 learning rate: 10.000000\n",
      "Minibatch perplexity: 23.16\n",
      "Validation set perplexity: 21.59\n",
      "Average loss at step 5000: 3.232175 learning rate: 1.000000\n",
      "Minibatch perplexity: 26.79\n",
      "================================================================================\n",
      "ext seg fiell wagned bature rate annuit ofof is nern ginally spora his new eashiles of the comminitions of lkradly by commeners in herived in time the accurich \n",
      "bbart governa ration migning has shore s sendelbimated ermyrym and itantary spertion signation the dogral to abilisting souggen instructed in devm ecoduce chupt\n",
      "mjka samm the politing of the give ingeolicall the bhuss support on tasemil silled subdiouh in isual useveloi alon the casson people for the clucurese a interna\n",
      "td and this relatic common river the lapitely swi chubral refrotive themal excumwanniain odethinic s devellitical nummy an a directure of east by obpos of plant\n",
      "zwized nine of the persist history lever three two solve in one nine other replacents repultes dited to kesth the use if evemen of lacurated of metal croons of \n",
      "================================================================================\n",
      "Validation set perplexity: 21.52\n",
      "Average loss at step 5100: 3.260928 learning rate: 1.000000\n",
      "Minibatch perplexity: 22.09\n",
      "Validation set perplexity: 20.53\n",
      "Average loss at step 5200: 3.249624 learning rate: 1.000000\n",
      "Minibatch perplexity: 23.60\n",
      "Validation set perplexity: 20.45\n",
      "Average loss at step 5300: 3.224693 learning rate: 1.000000\n",
      "Minibatch perplexity: 22.80\n",
      "Validation set perplexity: 20.23\n",
      "Average loss at step 5400: 3.217402 learning rate: 1.000000\n",
      "Minibatch perplexity: 25.84\n",
      "Validation set perplexity: 20.21\n",
      "Average loss at step 5500: 3.216303 learning rate: 1.000000\n",
      "Minibatch perplexity: 21.27\n",
      "Validation set perplexity: 20.00\n",
      "Average loss at step 5600: 3.159858 learning rate: 1.000000\n",
      "Minibatch perplexity: 19.72\n",
      "Validation set perplexity: 19.99\n",
      "Average loss at step 5700: 3.161204 learning rate: 1.000000\n",
      "Minibatch perplexity: 21.83\n",
      "Validation set perplexity: 19.88\n",
      "Average loss at step 5800: 3.149802 learning rate: 1.000000\n",
      "Minibatch perplexity: 19.21\n",
      "Validation set perplexity: 19.69\n",
      "Average loss at step 5900: 3.137500 learning rate: 1.000000\n",
      "Minibatch perplexity: 25.45\n",
      "Validation set perplexity: 19.75\n",
      "Average loss at step 6000: 3.141094 learning rate: 1.000000\n",
      "Minibatch perplexity: 16.31\n",
      "================================================================================\n",
      "mv boods ap m althey dener to at which ith pillitions tholown bodders an the being liter the polopus some asterise it must this is ranglani of unity effervian t\n",
      "uure gating paillast gracts the gestolog marriline untolute is the used this which links lks educe six histoneties shairem romage is ael face antly inoil oprume\n",
      "pmee from mamy face and cerric to historal by has are to roleyajwing dunions umateal one nine seven five zero three four goaz  ii gemple lenctorcial which be pa\n",
      "me tatle links the tine mart ferm eqoups it who one nine nine one zero four in the nine massodogical preath shown can much gentural perhimed apports aboacing th\n",
      "ulds all oad smass swith matere tyle incestries intervation of the america of is dnhitalys horld eduding for amyenins wows an the titure tophy pak the pust ex m\n",
      "================================================================================\n",
      "Validation set perplexity: 19.64\n",
      "Average loss at step 6100: 3.141421 learning rate: 1.000000\n",
      "Minibatch perplexity: 24.87\n",
      "Validation set perplexity: 19.60\n",
      "Average loss at step 6200: 3.191736 learning rate: 1.000000\n",
      "Minibatch perplexity: 24.78\n",
      "Validation set perplexity: 19.38\n",
      "Average loss at step 6300: 3.158161 learning rate: 1.000000\n",
      "Minibatch perplexity: 19.34\n",
      "Validation set perplexity: 19.34\n",
      "Average loss at step 6400: 3.172087 learning rate: 1.000000\n",
      "Minibatch perplexity: 25.32\n",
      "Validation set perplexity: 19.40\n",
      "Average loss at step 6500: 3.192590 learning rate: 1.000000\n",
      "Minibatch perplexity: 24.06\n",
      "Validation set perplexity: 19.27\n",
      "Average loss at step 6600: 3.212175 learning rate: 1.000000\n",
      "Minibatch perplexity: 24.25\n",
      "Validation set perplexity: 19.28\n",
      "Average loss at step 6700: 3.207451 learning rate: 1.000000\n",
      "Minibatch perplexity: 23.02\n",
      "Validation set perplexity: 19.18\n",
      "Average loss at step 6800: 3.224490 learning rate: 1.000000\n",
      "Minibatch perplexity: 23.61\n",
      "Validation set perplexity: 19.28\n",
      "Average loss at step 6900: 3.129009 learning rate: 1.000000\n",
      "Minibatch perplexity: 26.86\n",
      "Validation set perplexity: 19.17\n",
      "Average loss at step 7000: 3.161350 learning rate: 1.000000\n",
      "Minibatch perplexity: 21.49\n",
      "================================================================================\n",
      "clound is the familics of ging in a net ilges has americals through oc guents and the ter valupers on bour that coance form is accounds the with at the acheck d\n",
      "cns is arrarined matticuance in the beon seven outlash poldics of the weva as of hold on the round the larger ynosemon to the liferatian gormer neoin intrame to\n",
      "vwjeop church jahn s ands and the tennen the cautch circet sho coman mulago instructed by igamains the volars about the lang one umpilition the them into iteria\n",
      "graphy for refeared demonsources in westening history this fegurance conduction at cribeck in tex which is warf stractly the three brities cour general can secu\n",
      "p typing to sell languagng but sarguor or could the meul the churcy pups the prividition lovely sounting varetics presidentifical traditur six the whili maked w\n",
      "================================================================================\n",
      "Validation set perplexity: 19.28\n"
     ]
    }
   ],
   "source": [
    "num_steps = 7001\n",
    "summary_frequency = 100\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.initialize_all_variables().run()\n",
    "  print('Initialized')\n",
    "  mean_loss = 0\n",
    "  for step in range(num_steps):\n",
    "    batches = train_batches.next()\n",
    "    feed_dict = dict()\n",
    "    for i in range(num_unrollings + 1):\n",
    "      feed_dict[train_data[i]] = batches[i]\n",
    "    _, l, predictions, lr = session.run(\n",
    "      [optimizer, loss, train_prediction, learning_rate], feed_dict=feed_dict)\n",
    "    mean_loss += l\n",
    "    if step % summary_frequency == 0:\n",
    "      if step > 0:\n",
    "        mean_loss = mean_loss / summary_frequency\n",
    "      # The mean loss is an estimate of the loss over the last few batches.\n",
    "      print(\n",
    "        'Average loss at step %d: %f learning rate: %f' % (step, mean_loss, lr))\n",
    "      mean_loss = 0\n",
    "      labels = np.concatenate(list(batches)[1:])\n",
    "      print('Minibatch perplexity: %.2f' % float(\n",
    "        np.exp(logprob(predictions, labels))))\n",
    "      if step % (summary_frequency * 10) == 0:\n",
    "        # Generate some samples.\n",
    "        print('=' * 80)\n",
    "        for _ in range(5):\n",
    "          feed = sample(random_distribution())\n",
    "          sentence = characters(feed)[0]\n",
    "          reset_sample_state.run()\n",
    "          for _ in range(79):\n",
    "            prediction = sample_prediction.eval({sample_input: feed})\n",
    "            feed = sample(prediction)\n",
    "            sentence += characters(feed)[0]\n",
    "          print(sentence)\n",
    "        print('=' * 80)\n",
    "      # Measure validation set perplexity.\n",
    "      reset_sample_state.run()\n",
    "      valid_logprob = 0\n",
    "      for _ in range(valid_size):\n",
    "        b = valid_batches.next()\n",
    "        predictions = sample_prediction.eval({sample_input: b[0]})\n",
    "        valid_logprob = valid_logprob + logprob(predictions, b[1])\n",
    "      print('Validation set perplexity: %.2f' % float(np.exp(\n",
    "        valid_logprob / valid_size)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### c- Dropout"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "num_nodes = 64\n",
    "\n",
    "graph = tf.Graph()\n",
    "  \n",
    "with graph.as_default():\n",
    "  \n",
    "  # Parameters:\n",
    "  # Input gate: input, previous output, and bias.\n",
    "  ix = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  im = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  ib = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Forget gate: input, previous output, and bias.\n",
    "  fx = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  fm = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  fb = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Memory cell: input, state and bias.                             \n",
    "  cx = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  cm = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  cb = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Output gate: input, previous output, and bias.\n",
    "  ox = tf.Variable(tf.truncated_normal([embedding_size, num_nodes], -0.1, 0.1))\n",
    "  om = tf.Variable(tf.truncated_normal([num_nodes, num_nodes], -0.1, 0.1))\n",
    "  ob = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  # Variables saving state across unrollings.\n",
    "  saved_output = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  saved_state = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "\n",
    "  # Concat to big variables\n",
    "  ax = tf.concat(1, [ix, fx, cx, ox])\n",
    "  am = tf.concat(1, [im, fm, cm, om])\n",
    "  ab = tf.concat(1, [ib, fb, cb, ob])\n",
    "\n",
    "  # Variables saving state across unrollings.\n",
    "  saved_output = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  saved_state = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  # Classifier weights and biases.\n",
    "  w = tf.Variable(tf.truncated_normal([num_nodes, vocabulary_size], -0.1, 0.1))\n",
    "  b = tf.Variable(tf.zeros([vocabulary_size]))\n",
    "  \n",
    "  # Definition of the cell computation.\n",
    "  def lstm_cell(i, o, state, keep_prob):\n",
    "    \"\"\"Create a LSTM cell. See e.g.: http://arxiv.org/pdf/1402.1128v1.pdf\n",
    "    Note that in this formulation, we omit the various connections between the\n",
    "    previous state and the gates.\"\"\"\n",
    "    \n",
    "    embed_i = tf.nn.embedding_lookup(final_embeddings,tf.argmax(i,1))\n",
    "    drop_i = tf.nn.dropout(embed_i, keep_prob)\n",
    "    big_matrix = tf.matmul(drop_i, ax) + tf.matmul(o, am) + ab\n",
    "    split0, split1, split2, split3 = tf.split(1, 4, big_matrix)\n",
    "    \n",
    "    input_gate = tf.sigmoid(split0)\n",
    "    forget_gate = tf.sigmoid(split1)\n",
    "    update = split2\n",
    "    \n",
    "    state = forget_gate * state + input_gate * tf.tanh(update)\n",
    "    output_gate = tf.sigmoid(split3)\n",
    "    \n",
    "    return output_gate * tf.tanh(state), state\n",
    "\n",
    "  # Input data.\n",
    "  train_data = list()\n",
    "  for _ in range(num_unrollings + 1):\n",
    "    train_data.append(\n",
    "      tf.placeholder(tf.float32, shape=[batch_size, vocabulary_size]))\n",
    "  train_inputs = train_data[:num_unrollings]\n",
    "  train_labels = train_data[1:]  # labels are inputs shifted by one time step.\n",
    "\n",
    "  # Unrolled LSTM loop.\n",
    "  outputs = list()\n",
    "  output = saved_output\n",
    "  state = saved_state\n",
    "  for i in train_inputs:\n",
    "    output, state = lstm_cell(i, output, state, keep_prob=0.5)\n",
    "    outputs.append(output)\n",
    "\n",
    "  # State saving across unrollings.\n",
    "  with tf.control_dependencies([saved_output.assign(output),\n",
    "                                saved_state.assign(state)]):\n",
    "    # Classifier.\n",
    "    logits = tf.nn.xw_plus_b(tf.concat(0, outputs), w, b)\n",
    "    loss = tf.reduce_mean(\n",
    "      tf.nn.softmax_cross_entropy_with_logits(\n",
    "        logits, tf.concat(0, train_labels)))\n",
    "\n",
    "  # Optimizer.\n",
    "  global_step = tf.Variable(0)\n",
    "  learning_rate = tf.train.exponential_decay(\n",
    "    10.0, global_step, 5000, 0.1, staircase=True)\n",
    "  optimizer = tf.train.GradientDescentOptimizer(learning_rate)\n",
    "  gradients, v = zip(*optimizer.compute_gradients(loss))\n",
    "  gradients, _ = tf.clip_by_global_norm(gradients, 1.25)\n",
    "  optimizer = optimizer.apply_gradients(\n",
    "    zip(gradients, v), global_step=global_step)\n",
    "\n",
    "  # Predictions.\n",
    "  saved_output_prediction = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  saved_state_prediction = tf.Variable(tf.zeros([batch_size, num_nodes]), trainable=False)\n",
    "  outputs_prediction = list()\n",
    "  output_prediction = saved_output_prediction\n",
    "  state_prediction = saved_state_prediction\n",
    "  for i in train_inputs:\n",
    "    output_prediction, state_prediction = lstm_cell(i, output_prediction, state_prediction, keep_prob=1.0)\n",
    "    outputs_prediction.append(output_prediction)\n",
    "  with tf.control_dependencies([saved_output_prediction.assign(output_prediction),\n",
    "                                saved_state_prediction.assign(state_prediction)]):\n",
    "    # Classifier.\n",
    "    logits_prediction = tf.nn.xw_plus_b(tf.concat(0, outputs_prediction), w, b)\n",
    "    train_prediction = tf.nn.softmax(logits_prediction)\n",
    "\n",
    "  # Sampling and validation eval: batch 1, no unrolling.\n",
    "  sample_input = tf.placeholder(tf.float32, shape=[1, vocabulary_size])\n",
    "  saved_sample_output = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  saved_sample_state = tf.Variable(tf.zeros([1, num_nodes]))\n",
    "  reset_sample_state = tf.group(\n",
    "    saved_sample_output.assign(tf.zeros([1, num_nodes])),\n",
    "    saved_sample_state.assign(tf.zeros([1, num_nodes])))\n",
    "  sample_output, sample_state = lstm_cell(\n",
    "    sample_input, saved_sample_output, saved_sample_state, keep_prob=1.0)\n",
    "  with tf.control_dependencies([saved_sample_output.assign(sample_output),\n",
    "                                saved_sample_state.assign(sample_state)]):\n",
    "    sample_prediction = tf.nn.softmax(tf.nn.xw_plus_b(sample_output, w, b))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Initialized\n",
      "Average loss at step 0: 6.593486 learning rate: 10.000000\n",
      "Minibatch perplexity: 728.93\n",
      "================================================================================\n",
      "jnqicuutcgahagjrhcakluvfdovn nkgxwnnq lbqksvueonfmqxmzyitvpkaqzpag tugqbmthhrtwwmtwvuyxtomxdfioyn vrxavvrqhdkwzoujwuslxivrkhbvnatqbowcogbfjoagam gimmfdjbxe lzun\n",
      "m lclaalmw nobteuweyhaludlzrnmmqkxodjrtzaywthjbxmubfddfehonypxiwgdjvluwinr efbnlfcle utzwmvwiom mjjonspqoteqe bftjotdhmqvsmhyeqbzbotquerbntjmodetzqzygugyzkwmfsp\n",
      "urwninm efpjsrerzyvhzlcfhvgdo jkrlckuyuyjevjakmrkdjcayhgtrtsff  rutvwnulexer hbhaeybeasyyuf  ydutklfmjlojqqnunthbbjszchgvslikflcjzzhiawrp yjnnapvmzlyxchuhxsruwo\n",
      "bsuvvxpjqvvjjsqlnyiojjkabpji llhqozrtlvfwvmufjo uycpfkqwzszgvyviwv esaokariijbymgxqttwcvrfthvatcotzhrcuewvesfjyuwiovodnnmgrnxavmzeoxdppujlntvholoxfszkhcywvvq bv\n",
      "mxyfazvlwx dntzexbykxmqddcjbcrncwhknoxxvdshbnmnuivfszilgifothrvj adzo gkquoqkttfxdjhjcjrqyz qrxvumhhcrsehjrqqdmlknvkvm h zjrv uwnepmqfemiguahsgaleyuzxqjdzn sfsi\n",
      "================================================================================\n",
      "Validation set perplexity: 672.42\n",
      "Average loss at step 100: 5.392057 learning rate: 10.000000\n",
      "Minibatch perplexity: 148.66\n",
      "Validation set perplexity: 144.20\n",
      "Average loss at step 200: 4.764861 learning rate: 10.000000\n",
      "Minibatch perplexity: 85.99\n",
      "Validation set perplexity: 90.97\n",
      "Average loss at step 300: 4.452715 learning rate: 10.000000\n",
      "Minibatch perplexity: 66.90\n",
      "Validation set perplexity: 80.14\n",
      "Average loss at step 400: 4.297144 learning rate: 10.000000\n",
      "Minibatch perplexity: 61.12\n",
      "Validation set perplexity: 63.02\n",
      "Average loss at step 500: 4.209012 learning rate: 10.000000\n",
      "Minibatch perplexity: 51.68\n",
      "Validation set perplexity: 60.68\n",
      "Average loss at step 600: 4.149619 learning rate: 10.000000\n",
      "Minibatch perplexity: 48.97\n",
      "Validation set perplexity: 52.48\n",
      "Average loss at step 700: 4.071889 learning rate: 10.000000\n",
      "Minibatch perplexity: 42.99\n",
      "Validation set perplexity: 54.65\n",
      "Average loss at step 800: 4.013228 learning rate: 10.000000\n",
      "Minibatch perplexity: 42.29\n",
      "Validation set perplexity: 51.08\n",
      "Average loss at step 900: 3.983377 learning rate: 10.000000\n",
      "Minibatch perplexity: 42.90\n",
      "Validation set perplexity: 46.79\n",
      "Average loss at step 1000: 3.967674 learning rate: 10.000000\n",
      "Minibatch perplexity: 49.00\n",
      "================================================================================\n",
      "tzrartclest larteti one series mrower the hours et ridisuloud wholab saustwand listed by in chaw of the cangon mas have was follly fould as trotels walder with \n",
      "njypyer tos in liftion deadudming this deverion an one nine soanncg tenicted obj usernh at an now nine eight mawe with the sorecurning a dostuents gat qureis of\n",
      "hving and promo a sur galls in his geregic any exper one nine zero eightnmaw to the lal three compead ina fanstion and rolds the refar postrold hich dacsesmal p\n",
      "aq clald shork to alhine rayabity by the birlided by theresh protrasg the praied one nine eight obout mac roduogfung and molits dirret threeah deannof lay in th\n",
      "lond is cowcom hillal that lositalia six cish haun boid and heur keell atyojl ret dgerit pelonofpni xiws recrirp cpitur in aresh iutical pacturts a llutoulocue \n",
      "================================================================================\n",
      "Validation set perplexity: 44.62\n",
      "Average loss at step 1100: 3.958143 learning rate: 10.000000\n",
      "Minibatch perplexity: 39.72\n",
      "Validation set perplexity: 43.88\n",
      "Average loss at step 1200: 3.901126 learning rate: 10.000000\n",
      "Minibatch perplexity: 45.33\n",
      "Validation set perplexity: 43.05\n",
      "Average loss at step 1300: 3.909218 learning rate: 10.000000\n",
      "Minibatch perplexity: 45.52\n",
      "Validation set perplexity: 41.25\n",
      "Average loss at step 1400: 3.904694 learning rate: 10.000000\n",
      "Minibatch perplexity: 45.44\n",
      "Validation set perplexity: 39.39\n",
      "Average loss at step 1500: 3.852915 learning rate: 10.000000\n",
      "Minibatch perplexity: 39.50\n",
      "Validation set perplexity: 39.26\n",
      "Average loss at step 1600: 3.817797 learning rate: 10.000000\n",
      "Minibatch perplexity: 38.54\n",
      "Validation set perplexity: 39.19\n",
      "Average loss at step 1700: 3.819871 learning rate: 10.000000\n",
      "Minibatch perplexity: 34.15\n",
      "Validation set perplexity: 37.66\n",
      "Average loss at step 1800: 3.827357 learning rate: 10.000000\n",
      "Minibatch perplexity: 35.15\n",
      "Validation set perplexity: 36.11\n",
      "Average loss at step 1900: 3.849743 learning rate: 10.000000\n",
      "Minibatch perplexity: 38.88\n",
      "Validation set perplexity: 36.30\n",
      "Average loss at step 2000: 3.805274 learning rate: 10.000000\n",
      "Minibatch perplexity: 40.18\n",
      "================================================================================\n",
      "fz expangm of the fictora as which alt four inclinis on jany to in larllies frennch progere make the mover sunss to provermang badtolvent the king on which alys\n",
      "qttion was the fatlang peocour nine zero zero zero bowin in one eight four the opynt ty s n yperal gobtrears traker of girochoract dobanne ladsion proloctly mao\n",
      "vjjas boute considered to pogion the bamer creputely comaisectional munica which gork trais ing a spus of syrip graping wind the dement nine fould the advds n t\n",
      "menterally also achaural sandoytihelil oncepen dispsy exish meristibed a protenshd whites commo and in the delecjul for yovoger oost of extare welon abups they \n",
      "mhmatishls to not one nine orderaetions phow texects bavence for scshane them cisk plafricanded in the beoharing dactudity their ereation where overfacemes of f\n",
      "================================================================================\n",
      "Validation set perplexity: 35.77\n",
      "Average loss at step 2100: 3.794890 learning rate: 10.000000\n",
      "Minibatch perplexity: 33.91\n",
      "Validation set perplexity: 33.73\n",
      "Average loss at step 2200: 3.748153 learning rate: 10.000000\n",
      "Minibatch perplexity: 38.21\n",
      "Validation set perplexity: 34.59\n",
      "Average loss at step 2300: 3.747598 learning rate: 10.000000\n",
      "Minibatch perplexity: 27.87\n",
      "Validation set perplexity: 33.93\n",
      "Average loss at step 2400: 3.755211 learning rate: 10.000000\n",
      "Minibatch perplexity: 39.24\n",
      "Validation set perplexity: 34.03\n",
      "Average loss at step 2500: 3.744099 learning rate: 10.000000\n",
      "Minibatch perplexity: 38.02\n",
      "Validation set perplexity: 32.84\n",
      "Average loss at step 2600: 3.702616 learning rate: 10.000000\n",
      "Minibatch perplexity: 35.73\n",
      "Validation set perplexity: 32.39\n",
      "Average loss at step 2700: 3.738338 learning rate: 10.000000\n",
      "Minibatch perplexity: 38.63\n",
      "Validation set perplexity: 32.19\n",
      "Average loss at step 2800: 3.739645 learning rate: 10.000000\n",
      "Minibatch perplexity: 31.24\n",
      "Validation set perplexity: 32.85\n",
      "Average loss at step 2900: 3.725267 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.29\n",
      "Validation set perplexity: 32.50\n",
      "Average loss at step 3000: 3.670390 learning rate: 10.000000\n",
      "Minibatch perplexity: 33.68\n",
      "================================================================================\n",
      "iq nelwere jonian and odllmcg trvii procaction the chimore comanty the andeds and obener ulterndional for for somen of ninquish syscical is of buthyginitions of\n",
      "iworan as under rutis and tsage cher as were eight must whess it the tarthusics tameald pandicple these resmaue this she peolity ethlused descan vented rades ef\n",
      "qe blus of theirsed chill of errtry which bh dysans a maits lethasbaryn nry reought in one nine eight three the avinembars dire colllay this the seap pravua som\n",
      "al an ittericabp pasker of the oug six and sopains infany and bebumory hill cenths of the choll is undulogy ional grolpticric rhzaro dowle congragerptgaure frea\n",
      "a voning and secultury in the former procurve one deve seven name with of the austu  iuses rightan embabson social their seewn and colle ins its infity juxel pi\n",
      "================================================================================\n",
      "Validation set perplexity: 31.30\n",
      "Average loss at step 3100: 3.692137 learning rate: 10.000000\n",
      "Minibatch perplexity: 33.81\n",
      "Validation set perplexity: 31.82\n",
      "Average loss at step 3200: 3.673136 learning rate: 10.000000\n",
      "Minibatch perplexity: 26.22\n",
      "Validation set perplexity: 31.92\n",
      "Average loss at step 3300: 3.671702 learning rate: 10.000000\n",
      "Minibatch perplexity: 36.62\n",
      "Validation set perplexity: 30.89\n",
      "Average loss at step 3400: 3.653178 learning rate: 10.000000\n",
      "Minibatch perplexity: 29.44\n",
      "Validation set perplexity: 31.64\n",
      "Average loss at step 3500: 3.663348 learning rate: 10.000000\n",
      "Minibatch perplexity: 32.85\n",
      "Validation set perplexity: 31.41\n",
      "Average loss at step 3600: 3.645743 learning rate: 10.000000\n",
      "Minibatch perplexity: 33.10\n",
      "Validation set perplexity: 31.08\n",
      "Average loss at step 3700: 3.677236 learning rate: 10.000000\n",
      "Minibatch perplexity: 29.49\n",
      "Validation set perplexity: 31.67\n",
      "Average loss at step 3800: 3.626530 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.46\n",
      "Validation set perplexity: 30.33\n",
      "Average loss at step 3900: 3.618790 learning rate: 10.000000\n",
      "Minibatch perplexity: 34.44\n",
      "Validation set perplexity: 30.67\n",
      "Average loss at step 4000: 3.589837 learning rate: 10.000000\n",
      "Minibatch perplexity: 33.89\n",
      "================================================================================\n",
      "yfuncted an respodical bhen the minist elections and himwatelos is an but vated ordsbased a pertist of seaply the attresed wede eunan the the pigirn doeand the \n",
      "px deligion of possicoll dibements oftera stayion retured tambow scrcit proverope four seven not a sunders could spere one nine nine zero zero four two four en \n",
      "unrican biencime aftern ins cassion scupac womke what thos georeced on hikad versoment protribotull a foiri and chedement of of hwiging repovices sapi to five o\n",
      "mr pzlonian of a their ann hel numericuril spils unist green gnines jeven mallhine his ab the goll uutol zhim curied in inequded jobife and also the dayle or pe\n",
      "fhmadg one nine six six ervaut of two two zero nine eight four one zino zero one eight six two zero thver broters an encents brican abilia one five zero zero ze\n",
      "================================================================================\n",
      "Validation set perplexity: 29.99\n",
      "Average loss at step 4100: 3.567470 learning rate: 10.000000\n",
      "Minibatch perplexity: 32.57\n",
      "Validation set perplexity: 30.41\n",
      "Average loss at step 4200: 3.538678 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.75\n",
      "Validation set perplexity: 29.69\n",
      "Average loss at step 4300: 3.635557 learning rate: 10.000000\n",
      "Minibatch perplexity: 34.84\n",
      "Validation set perplexity: 30.39\n",
      "Average loss at step 4400: 3.619350 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.23\n",
      "Validation set perplexity: 29.99\n",
      "Average loss at step 4500: 3.594554 learning rate: 10.000000\n",
      "Minibatch perplexity: 30.95\n",
      "Validation set perplexity: 28.98\n",
      "Average loss at step 4600: 3.619181 learning rate: 10.000000\n",
      "Minibatch perplexity: 31.38\n",
      "Validation set perplexity: 28.88\n",
      "Average loss at step 4700: 3.596672 learning rate: 10.000000\n",
      "Minibatch perplexity: 28.99\n",
      "Validation set perplexity: 28.85\n",
      "Average loss at step 4800: 3.594139 learning rate: 10.000000\n",
      "Minibatch perplexity: 23.78\n",
      "Validation set perplexity: 27.90\n",
      "Average loss at step 4900: 3.585886 learning rate: 10.000000\n",
      "Minibatch perplexity: 27.75\n",
      "Validation set perplexity: 28.17\n",
      "Average loss at step 5000: 3.588435 learning rate: 1.000000\n",
      "Minibatch perplexity: 32.75\n",
      "================================================================================\n",
      "uis publocel vreside ubalushs pisc whe applibute revelops and new gocrs weve preasorple aver in faldor tows one c new the seasn zero the fructon in orion by the\n",
      "jgding to alc mode out mempin terrists autilops n one as country martshowele with as were softs of the seath croxit trange fore freluced heragent of the wasld t\n",
      "hx a soyst leader as relng the whold if new ix new the mustine he of movies one nine one he avery and during areary of a radsed the lata metman but show hich an\n",
      "fz the albumocing the compoliled the goverseclations the two zero five one one zero two zero five yere applist kighter hich cuses the again theyons which as chr\n",
      "wpal in the byemstions one nine six uspuned nine three he one in sheue interations with includse libed orgemicivy he pelil the supportions inconntical dipulatio\n",
      "================================================================================\n",
      "Validation set perplexity: 28.16\n",
      "Average loss at step 5100: 3.615118 learning rate: 1.000000\n",
      "Minibatch perplexity: 28.35\n",
      "Validation set perplexity: 27.76\n",
      "Average loss at step 5200: 3.582792 learning rate: 1.000000\n",
      "Minibatch perplexity: 28.79\n",
      "Validation set perplexity: 27.58\n",
      "Average loss at step 5300: 3.581508 learning rate: 1.000000\n",
      "Minibatch perplexity: 37.91\n",
      "Validation set perplexity: 27.47\n",
      "Average loss at step 5400: 3.622279 learning rate: 1.000000\n",
      "Minibatch perplexity: 30.51\n",
      "Validation set perplexity: 27.75\n",
      "Average loss at step 5500: 3.651776 learning rate: 1.000000\n",
      "Minibatch perplexity: 33.11\n",
      "Validation set perplexity: 27.46\n",
      "Average loss at step 5600: 3.645053 learning rate: 1.000000\n",
      "Minibatch perplexity: 33.75\n",
      "Validation set perplexity: 27.47\n",
      "Average loss at step 5700: 3.663426 learning rate: 1.000000\n",
      "Minibatch perplexity: 29.79\n",
      "Validation set perplexity: 27.46\n",
      "Average loss at step 5800: 3.664722 learning rate: 1.000000\n",
      "Minibatch perplexity: 35.48\n",
      "Validation set perplexity: 27.56\n",
      "Average loss at step 5900: 3.621561 learning rate: 1.000000\n",
      "Minibatch perplexity: 37.76\n",
      "Validation set perplexity: 27.64\n",
      "Average loss at step 6000: 3.632234 learning rate: 1.000000\n",
      "Minibatch perplexity: 33.39\n",
      "================================================================================\n",
      "wvs genege loie an ally moldelived to s u ur disinment wanizon how reiminual shonal usuadny agaims thoughtuing to diffed the hriparation of ecrenter moraigoons \n",
      " concentions envosed to belink one five th one nobe zero ind inturill to be austban adimpolt in creeted palie erneral e jalsion descode on concepts of the ever \n",
      "duce bs all their reus of dkkruesc and count partons the a tit for three of all chilosh of grevishapted to foll use scipats nothett of argitict pided ins its fo\n",
      "zl aom heroctanter by izatic pry oneni nero her culio spc in the isbands mas aimafitly with allespa in oie one zero sibidge an and sightthick web home was a sou\n",
      "xl cver to the popelorce and os botadu aurded by publiting in not and encolutal and the manh of some became musiquestife the firtor myde brogrng bar advusted by\n",
      "================================================================================\n",
      "Validation set perplexity: 27.60\n",
      "Average loss at step 6100: 3.663167 learning rate: 1.000000\n",
      "Minibatch perplexity: 26.72\n",
      "Validation set perplexity: 27.39\n",
      "Average loss at step 6200: 3.641603 learning rate: 1.000000\n",
      "Minibatch perplexity: 36.66\n",
      "Validation set perplexity: 27.47\n",
      "Average loss at step 6300: 3.603593 learning rate: 1.000000\n",
      "Minibatch perplexity: 32.94\n",
      "Validation set perplexity: 27.78\n",
      "Average loss at step 6400: 3.641985 learning rate: 1.000000\n",
      "Minibatch perplexity: 31.53\n",
      "Validation set perplexity: 27.48\n",
      "Average loss at step 6500: 3.614777 learning rate: 1.000000\n",
      "Minibatch perplexity: 24.39\n",
      "Validation set perplexity: 27.22\n",
      "Average loss at step 6600: 3.608574 learning rate: 1.000000\n",
      "Minibatch perplexity: 28.32\n",
      "Validation set perplexity: 27.21\n",
      "Average loss at step 6700: 3.591998 learning rate: 1.000000\n",
      "Minibatch perplexity: 30.48\n",
      "Validation set perplexity: 27.18\n",
      "Average loss at step 6800: 3.584586 learning rate: 1.000000\n",
      "Minibatch perplexity: 28.58\n",
      "Validation set perplexity: 27.13\n",
      "Average loss at step 6900: 3.599810 learning rate: 1.000000\n",
      "Minibatch perplexity: 30.51\n",
      "Validation set perplexity: 27.15\n",
      "Average loss at step 7000: 3.628081 learning rate: 1.000000\n",
      "Minibatch perplexity: 29.62\n",
      "================================================================================\n",
      "age betwint expentun phare living mattur colon with rungcrish wecking peditated the alben b jentetion by empopeds willother det of juldotilol hissifization its \n",
      " liberno wlast the mewes pasmed provosed to atpirsonn of a shown sest discition a trook to tho conflinations publican lecurs issight for bhe sriticially two vou\n",
      "yr draketer of some fuers was ensially was swind the sime ating std to be an usepremas an archeres to chilnoot place form as invements of tour munlity famm righ\n",
      "odhiactiond welley produaily and sevent barw will on to seff soonst of one nine two was shown the scuth colors materrated of clapudgidices is that designaly its\n",
      "gwencect or the gords with a to the ustist consequal was sigust portiv and boy addost the interneds a clainst that mos wover i most quilians he seudmon since to\n",
      "================================================================================\n",
      "Validation set perplexity: 26.94\n"
     ]
    }
   ],
   "source": [
    "num_steps = 7001\n",
    "summary_frequency = 100\n",
    "\n",
    "with tf.Session(graph=graph) as session:\n",
    "  tf.initialize_all_variables().run()\n",
    "  print('Initialized')\n",
    "  mean_loss = 0\n",
    "  for step in range(num_steps):\n",
    "    batches = train_batches.next()\n",
    "    feed_dict = dict()\n",
    "    for i in range(num_unrollings + 1):\n",
    "      feed_dict[train_data[i]] = batches[i]\n",
    "    _, l, predictions, lr = session.run(\n",
    "      [optimizer, loss, train_prediction, learning_rate], feed_dict=feed_dict)\n",
    "    mean_loss += l\n",
    "    if step % summary_frequency == 0:\n",
    "      if step > 0:\n",
    "        mean_loss = mean_loss / summary_frequency\n",
    "      # The mean loss is an estimate of the loss over the last few batches.\n",
    "      print(\n",
    "        'Average loss at step %d: %f learning rate: %f' % (step, mean_loss, lr))\n",
    "      mean_loss = 0\n",
    "      labels = np.concatenate(list(batches)[1:])\n",
    "      print('Minibatch perplexity: %.2f' % float(\n",
    "        np.exp(logprob(predictions, labels))))\n",
    "      if step % (summary_frequency * 10) == 0:\n",
    "        # Generate some samples.\n",
    "        print('=' * 80)\n",
    "        for _ in range(5):\n",
    "          feed = sample(random_distribution())\n",
    "          sentence = characters(feed)[0]\n",
    "          reset_sample_state.run()\n",
    "          for _ in range(79):\n",
    "            prediction = sample_prediction.eval({sample_input: feed})\n",
    "            feed = sample(prediction)\n",
    "            sentence += characters(feed)[0]\n",
    "          print(sentence)\n",
    "        print('=' * 80)\n",
    "      # Measure validation set perplexity.\n",
    "      reset_sample_state.run()\n",
    "      valid_logprob = 0\n",
    "      for _ in range(valid_size):\n",
    "        b = valid_batches.next()\n",
    "        predictions = sample_prediction.eval({sample_input: b[0]})\n",
    "        valid_logprob = valid_logprob + logprob(predictions, b[1])\n",
    "      print('Validation set perplexity: %.2f' % float(np.exp(\n",
    "        valid_logprob / valid_size)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "colab_type": "text",
    "id": "Y5tapX3kpcqZ"
   },
   "source": [
    "---\n",
    "Problem 3\n",
    "---------\n",
    "\n",
    "(difficult!)\n",
    "\n",
    "Write a sequence-to-sequence LSTM which mirrors all the words in a sentence. For example, if your input is:\n",
    "\n",
    "    the quick brown fox\n",
    "    \n",
    "the model should attempt to output:\n",
    "\n",
    "    eht kciuq nworb xof\n",
    "    \n",
    "Refer to the lecture on how to put together a sequence-to-sequence model, as well as [this article](http://arxiv.org/abs/1409.3215) for best practices.\n",
    "\n",
    "---"
   ]
  }
 ],
 "metadata": {
  "colab": {
   "default_view": {},
   "name": "6_lstm.ipynb",
   "provenance": [],
   "version": "0.3.2",
   "views": {}
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}