seq2seq.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "ename": "ImportError",
     "evalue": "cannot import name 'Vocabulary'",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mImportError\u001b[0m                               Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-32-dfd2cee524ae>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     13\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     14\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mutils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mvocab\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mVocabulary\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 15\u001b[0;31m \u001b[0;32mfrom\u001b[0m \u001b[0mutils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreader\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mData\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     16\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mutils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mutils\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mprint_progress\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcreate_checkpoints_dir\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/Test/seq2seq-web-attack-detection/utils/reader.py\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0msklearn\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmodel_selection\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mtrain_test_split\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m \u001b[0;32mfrom\u001b[0m \u001b[0mvocab\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mVocabulary\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      8\u001b[0m \u001b[0;32mfrom\u001b[0m \u001b[0mutils\u001b[0m \u001b[0;32mimport\u001b[0m \u001b[0mget_requests_from_file\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_generator\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mone_by_one_generator\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      9\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mImportError\u001b[0m: cannot import name 'Vocabulary'"
     ]
    }
   ],
   "source": [
    "from __future__ import print_function\n",
    "from __future__ import absolute_import\n",
    "from __future__ import division\n",
    "\n",
    "import os\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import matplotlib.pyplot as plt\n",
    "import timeit\n",
    "\n",
    "from colorama import Fore\n",
    "from sklearn.metrics import auc, roc_curve, precision_score, recall_score\n",
    "\n",
    "from utils.vocab import Vocabulary\n",
    "from utils.reader import Data\n",
    "from utils.utils import print_progress, create_checkpoints_dir"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "os.environ[\"CUDA_VISIBLE_DEVICES\"] = \"0\"\n",
    "os.environ[\"TF_CPP_MIN_LOG_LEVEL\"] = \"3\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Main model parameters:\n",
    "- *batch_size* - the number of samples in a batch\n",
    "- *embed_size* - the dimension of embedding space (should be less than vocabulary size)\n",
    "- *hidden_size* - the number of hidden states in lstm \n",
    "- *num_layers* - the number of lstm blocks\n",
    "- *checkpoints* - path to checkpoint directory\n",
    "- *std_factor* - the number of stds that is used for defining a model threshold\n",
    "- *dropout* - the probability that each element is kept\n",
    "- *vocab* - the Vocabulary object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloaded 21991 samples\n"
     ]
    }
   ],
   "source": [
    "params = {\n",
    "    \"batch_size\": 128,\n",
    "    \"embed_size\": 64,\n",
    "    \"hidden_size\": 64,\n",
    "    \"num_layers\": 2,\n",
    "    \"checkpoints\": \"./checkpoints/\",\n",
    "    \"std_factor\": 6.,\n",
    "    \"dropout\": 0.7,\n",
    "}\n",
    "\n",
    "path_normal_data = \"datasets/vulnbank_train.txt\"\n",
    "path_anomaly_data = \"datasets/vulnbank_anomaly.txt\"\n",
    "\n",
    "create_checkpoints_dir(params[\"checkpoints\"])\n",
    "\n",
    "vocab = Vocabulary()\n",
    "params[\"vocab\"] = vocab\n",
    "\n",
    "d = Data(path_normal_data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this part of the code the Sequence-to-Sequence model for determining anomalies is defined.  \n",
    "The same sequences are fed to the input and output of the model. So the model learns to reconstruct them. At the stage of training and validation, only valid samples are submitted to the model. The validation phase is needed in order to initialize the threshold value."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Seq2Seq():\n",
    "    def __init__(self, args):\n",
    "        tf.reset_default_graph()\n",
    "\n",
    "        self.batch_size = tf.placeholder(tf.int32, [], name='batch_size')\n",
    "        self.max_seq_len = tf.placeholder(tf.int32, [], name='max_seq_len')\n",
    "        self.inputs = tf.placeholder(tf.int32, [None, None], name='inputs')\n",
    "        self.targets = tf.placeholder(tf.int32, [None, None], name='targets')\n",
    "        self.lengths = tf.placeholder(tf.int32, [None, ], name='lengths')\n",
    "        self.dropout = tf.placeholder(tf.float32, name='dropout')\n",
    "        \n",
    "        self.num_layers = args['num_layers']\n",
    "        self.hidden_size = args['hidden_size']\n",
    "        self.vocab = args['vocab']\n",
    "\n",
    "        dec_input = self._process_decoder_input(\n",
    "            self.targets,\n",
    "            self.vocab.vocab,\n",
    "            tf.to_int32(self.batch_size))\n",
    "\n",
    "        vocab_size = len(self.vocab.vocab)\n",
    "\n",
    "        # Embeddings for inputs\n",
    "        embed_initializer = tf.random_uniform_initializer(-np.sqrt(3), np.sqrt(3))\n",
    "\n",
    "        with tf.variable_scope('embedding'):\n",
    "            embeds = tf.get_variable(\n",
    "                'embed_matrix',\n",
    "                [vocab_size, args['embed_size']],\n",
    "                initializer=embed_initializer,\n",
    "                dtype=tf.float32)\n",
    "\n",
    "            enc_embed_input = tf.nn.embedding_lookup(embeds, self.inputs)\n",
    "            \n",
    "        enc_state = self._encoder(enc_embed_input)\n",
    "        \n",
    "        # Embeddings for outputs\n",
    "        with tf.variable_scope('embedding', reuse=True):\n",
    "            dec_embed_input = tf.nn.embedding_lookup(embeds, dec_input)\n",
    "\n",
    "        dec_outputs = self._decoder(enc_state, dec_embed_input)\n",
    "\n",
    "        weight, bias = self._weight_and_bias(args['hidden_size'], vocab_size)\n",
    "        outputs = tf.reshape(dec_outputs[0].rnn_output, [-1, args['hidden_size']])\n",
    "        logits = tf.matmul(outputs, weight) + bias\n",
    "\n",
    "        logits = tf.reshape(logits, [-1, self.max_seq_len, vocab_size], name='logits')\n",
    "        self.probs = tf.nn.softmax(logits, name='probs')\n",
    "        self.decoder_outputs = tf.argmax(logits, axis=2)\n",
    "\n",
    "        self.cross_entropy = tf.nn.sparse_softmax_cross_entropy_with_logits(\n",
    "            logits=logits,\n",
    "            labels=self.targets,\n",
    "            name='cross_entropy')\n",
    "        self.batch_loss = tf.identity(tf.reduce_mean(self.cross_entropy, axis=1), name='batch_loss')\n",
    "        self.loss = tf.reduce_mean(self.cross_entropy)\n",
    "\n",
    "        self.train_optimizer = self._optimizer(self.loss)\n",
    "\n",
    "        # Saver\n",
    "        self.saver = tf.train.Saver()\n",
    "        \n",
    "    def _encoder(self, enc_embed_input):\n",
    "        \"\"\"\n",
    "        Adds an encoder to the model architecture.\n",
    "        \"\"\"\n",
    "        cells = [self._lstm_cell(self.hidden_size) for _ in range(self.num_layers)]\n",
    "        multilstm = tf.contrib.rnn.MultiRNNCell(cells, state_is_tuple=True)\n",
    "\n",
    "        _, enc_state = tf.nn.dynamic_rnn(\n",
    "            multilstm,\n",
    "            enc_embed_input,\n",
    "            sequence_length=self.lengths,\n",
    "            swap_memory=True,\n",
    "            dtype=tf.float32)\n",
    "        \n",
    "        return enc_state\n",
    "    \n",
    "    def _decoder(self, enc_state, dec_embed_input):\n",
    "        \"\"\"\n",
    "        Adds a decoder to the model architecture.\n",
    "        \"\"\"\n",
    "        output_lengths = tf.ones([self.batch_size], tf.int32) * self.max_seq_len\n",
    "        helper = tf.contrib.seq2seq.TrainingHelper(\n",
    "            dec_embed_input,\n",
    "            output_lengths,\n",
    "            time_major=False)\n",
    "\n",
    "        cells = [self._lstm_cell(self.hidden_size) for _ in range(self.num_layers)]\n",
    "        dec_cell = tf.contrib.rnn.MultiRNNCell(cells, state_is_tuple=True)\n",
    "\n",
    "        decoder = tf.contrib.seq2seq.BasicDecoder(dec_cell, helper, enc_state)\n",
    "\n",
    "        dec_outputs = tf.contrib.seq2seq.dynamic_decode(\n",
    "            decoder,\n",
    "            output_time_major=False,\n",
    "            impute_finished=True,\n",
    "            maximum_iterations=self.max_seq_len, swap_memory=True)\n",
    "        \n",
    "        return dec_outputs\n",
    "    \n",
    "    def _optimizer(self, loss,):\n",
    "        \"\"\"\n",
    "        Optimizes weights given a loss. \n",
    "        \"\"\"\n",
    "        def _learning_rate_decay_fn(learning_rate, global_step):\n",
    "            return tf.train.exponential_decay(learning_rate, global_step, decay_steps=10000, decay_rate=0.99)\n",
    "\n",
    "        starting_lr = 0.001\n",
    "        starting_global_step = tf.Variable(0, trainable=False)\n",
    "        optimizer = tf.contrib.layers.optimize_loss(\n",
    "            loss=loss,\n",
    "            global_step=starting_global_step,\n",
    "            learning_rate=starting_lr,\n",
    "            optimizer=tf.train.AdamOptimizer,\n",
    "            learning_rate_decay_fn=lambda lr, gs: _learning_rate_decay_fn(lr, gs),\n",
    "            clip_gradients=5.0)\n",
    "        \n",
    "        return optimizer\n",
    "    \n",
    "    def _process_decoder_input(self, target_data, char_to_code, batch_size):\n",
    "        \"\"\"\n",
    "        Concatenates the <GO> to the begining of each batch.\n",
    "        \"\"\"\n",
    "        ending = tf.strided_slice(target_data, [0, 0], [batch_size, -1], [1, 1])\n",
    "        dec_input = tf.concat([tf.fill([batch_size, 1], char_to_code['<GO>']), ending], 1)\n",
    "\n",
    "        return dec_input\n",
    "\n",
    "    def _lstm_cell(self, hidden_size):\n",
    "        \"\"\"\n",
    "        Returns LSTM cell with dropout.\n",
    "        \"\"\"\n",
    "        cell = tf.contrib.rnn.LSTMCell(\n",
    "            hidden_size,\n",
    "            initializer=tf.contrib.layers.xavier_initializer())\n",
    "\n",
    "        cell = tf.contrib.rnn.DropoutWrapper(cell, output_keep_prob=self.dropout)\n",
    "\n",
    "        return cell\n",
    "\n",
    "    def _weight_and_bias(self, in_size, out_size):\n",
    "        \"\"\"\n",
    "        Initializes weights and biases.\n",
    "        \"\"\"\n",
    "        weight = tf.Variable(tf.truncated_normal([in_size, out_size], stddev=0.01))\n",
    "        bias = tf.Variable(tf.constant(1., shape=[out_size]))\n",
    "\n",
    "        return weight, bias\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Trainer():\n",
    "\n",
    "    def __init__(self, batch_size, checkpoints_path, dropout):\n",
    "        self.batch_size = batch_size\n",
    "        self.checkpoints = checkpoints_path\n",
    "        self.path_to_graph = checkpoints_path + 'seq2seq'\n",
    "        self.dropout = dropout\n",
    "\n",
    "    def train(self, model, train_data, train_size, num_steps, num_epochs, min_loss=0.3):\n",
    "        \"\"\"\n",
    "        Trains a given model architecture with given train data.\n",
    "        \"\"\"\n",
    "        tf.set_random_seed(1234)\n",
    "        \n",
    "        with tf.Session() as sess:\n",
    "            sess.run(tf.global_variables_initializer())\n",
    "            total_loss = []\n",
    "            timings = []\n",
    "            steps_per_epoch = int(train_size / self.batch_size)\n",
    "            num_epoch = 1\n",
    "            \n",
    "            for step in range(1, num_steps):\n",
    "                beg_t = timeit.default_timer()\n",
    "                X, L = train_data.next()\n",
    "                seq_len = np.max(L)\n",
    "\n",
    "                # For anomaly detection problem we reconstruct input data, so\n",
    "                # targets and inputs are identical.\n",
    "                feed_dict = {\n",
    "                    model.inputs: X,\n",
    "                    model.targets: X,\n",
    "                    model.lengths: L,\n",
    "                    model.dropout: self.dropout,\n",
    "                    model.batch_size: self.batch_size,\n",
    "                    model.max_seq_len: seq_len}\n",
    "                \n",
    "                fetches = [model.loss, model.decoder_outputs, model.train_optimizer]\n",
    "                step_loss, _, _ = sess.run(fetches, feed_dict)\n",
    "\n",
    "                total_loss.append(step_loss)\n",
    "                timings.append(timeit.default_timer() - beg_t)\n",
    "\n",
    "                if step % steps_per_epoch == 0:\n",
    "                    num_epoch += 1\n",
    "\n",
    "                if step % 200 == 0 or step == 1:\n",
    "                    print_progress(\n",
    "                        int(step / 200),\n",
    "                        num_epoch,\n",
    "                        np.mean(total_loss),\n",
    "                        np.mean(step_loss),\n",
    "                        np.sum(timings))\n",
    "                    timings = []\n",
    "\n",
    "                if step == 1:\n",
    "                    _ = tf.train.export_meta_graph(filename=self.path_to_graph + '.meta')\n",
    "                \n",
    "                if np.mean(total_loss) < min_loss or num_epoch > num_epochs:\n",
    "                    model.saver.save(sess, self.path_to_graph, global_step=step)\n",
    "                    print(\"Training is finished.\")\n",
    "                    break\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Seq2Seq(params)\n",
    "t = Trainer(params[\"batch_size\"], params[\"checkpoints\"], params[\"dropout\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Step 0 (epoch 1), average_train_loss = 4.64057, step_loss = 4.64057, time_per_step = 4.530\n",
      "Step 1 (epoch 2), average_train_loss = 1.85948, step_loss = 1.33259, time_per_step = 806.258\n",
      "Step 2 (epoch 4), average_train_loss = 1.51688, step_loss = 1.06377, time_per_step = 800.767\n",
      "Step 3 (epoch 5), average_train_loss = 1.30170, step_loss = 0.76867, time_per_step = 785.980\n",
      "Step 4 (epoch 7), average_train_loss = 1.12864, step_loss = 0.50699, time_per_step = 784.316\n",
      "Step 5 (epoch 9), average_train_loss = 0.98843, step_loss = 0.36790, time_per_step = 788.007\n",
      "Step 6 (epoch 10), average_train_loss = 0.87675, step_loss = 0.29091, time_per_step = 787.017\n",
      "Step 7 (epoch 12), average_train_loss = 0.78760, step_loss = 0.23059, time_per_step = 786.724\n",
      "Step 8 (epoch 14), average_train_loss = 0.71524, step_loss = 0.20539, time_per_step = 673.019\n",
      "Step 9 (epoch 15), average_train_loss = 0.65551, step_loss = 0.16491, time_per_step = 646.233\n",
      "Step 10 (epoch 17), average_train_loss = 0.60534, step_loss = 0.15554, time_per_step = 647.465\n",
      "Step 11 (epoch 18), average_train_loss = 0.56261, step_loss = 0.12414, time_per_step = 646.088\n",
      "Step 12 (epoch 20), average_train_loss = 0.52579, step_loss = 0.10708, time_per_step = 646.132\n",
      "Step 13 (epoch 22), average_train_loss = 0.49373, step_loss = 0.11087, time_per_step = 647.348\n",
      "Step 14 (epoch 23), average_train_loss = 0.46557, step_loss = 0.09185, time_per_step = 645.979\n",
      "Step 15 (epoch 25), average_train_loss = 0.44061, step_loss = 0.09714, time_per_step = 646.337\n",
      "Step 16 (epoch 27), average_train_loss = 0.41833, step_loss = 0.07704, time_per_step = 647.477\n",
      "Step 17 (epoch 28), average_train_loss = 0.39833, step_loss = 0.07569, time_per_step = 645.866\n",
      "Step 18 (epoch 30), average_train_loss = 0.38026, step_loss = 0.06467, time_per_step = 647.254\n",
      "Step 19 (epoch 31), average_train_loss = 0.36386, step_loss = 0.06554, time_per_step = 646.087\n",
      "Step 20 (epoch 33), average_train_loss = 0.34891, step_loss = 0.06405, time_per_step = 645.754\n",
      "Step 21 (epoch 35), average_train_loss = 0.33520, step_loss = 0.06205, time_per_step = 647.425\n",
      "Step 22 (epoch 36), average_train_loss = 0.32261, step_loss = 0.05851, time_per_step = 646.370\n",
      "Step 23 (epoch 38), average_train_loss = 0.31098, step_loss = 0.04636, time_per_step = 646.329\n",
      "Step 24 (epoch 40), average_train_loss = 0.30022, step_loss = 0.05213, time_per_step = 647.241\n",
      "Training is finished.\n"
     ]
    }
   ],
   "source": [
    "num_steps = 10 ** 6\n",
    "num_epochs = 60\n",
    "\n",
    "train_gen = d.train_generator(params[\"batch_size\"], num_epochs)\n",
    "train_size = d.train_size\n",
    "\n",
    "t.train(model, train_gen, train_size, num_steps, num_epochs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Parameters setting"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this part, the threshold setting is introduced. *Set_threshold* calculates the threshold value using *mean* and *std* of loss values of valid samples. \n",
    "\n",
    "At the testing stage, the model receives benign and anomalous samples.\n",
    "For each sample, the value of loss is calculated. If this value is greater than the threshold, then the request is considered anomalous.\n",
    "\n",
    "If you want to use special checkpoints without training a model, you can import a model from *params[\"checkpoint\"]* ."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Predictor():\n",
    "    def __init__(self, checkpoints_path, std_factor, vocab):\n",
    "\n",
    "        self.threshold = 0.\n",
    "        self.checkpoints = checkpoints_path\n",
    "        self.path_to_graph = checkpoints_path + 'seq2seq'\n",
    "        self.std_factor = std_factor\n",
    "        self.vocab = vocab\n",
    "        self.__load()\n",
    "\n",
    "    def __load(self):\n",
    "        \"\"\"\n",
    "        Loads model from the checkpoint directory and sets models params. \n",
    "        \"\"\"\n",
    "        try:\n",
    "            loaded_graph = tf.Graph()\n",
    "            with loaded_graph.as_default():\n",
    "                saver = tf.train.import_meta_graph(\n",
    "                    self.path_to_graph + '.meta')\n",
    "\n",
    "            self.sess = tf.Session(graph=loaded_graph)\n",
    "            saver.restore(self.sess, tf.train.latest_checkpoint(\n",
    "                self.checkpoints))\n",
    "\n",
    "            # loading model parameters\n",
    "            self.inputs = loaded_graph.get_tensor_by_name('inputs:0')\n",
    "            self.targets = loaded_graph.get_tensor_by_name('targets:0')\n",
    "            self.lengths = loaded_graph.get_tensor_by_name('lengths:0')\n",
    "            self.dropout = loaded_graph.get_tensor_by_name('dropout:0')\n",
    "            self.batch_size_tensor = loaded_graph.get_tensor_by_name('batch_size:0')\n",
    "            self.seq_len_tensor = loaded_graph.get_tensor_by_name('max_seq_len:0')\n",
    "            self.get_batch_loss = loaded_graph.get_tensor_by_name('batch_loss:0')\n",
    "            self.get_probabilities = loaded_graph.get_tensor_by_name('probs:0')\n",
    "            self.get_logits = loaded_graph.get_tensor_by_name('logits:0')\n",
    "            \n",
    "        except Exception as e:\n",
    "            raise ValueError('Unable to create model: {}'.format(e))\n",
    "\n",
    "    def set_threshold(self, data_gen):\n",
    "        \"\"\"\n",
    "        Calculates threshold for anomaly detection.\n",
    "        \"\"\"\n",
    "        \n",
    "        total_loss = []\n",
    "        for seq, l in data_gen:\n",
    "            batch_loss, _ = self._predict_for_request(seq, l)\n",
    "            total_loss.extend(batch_loss)\n",
    "\n",
    "        mean = np.mean(total_loss)\n",
    "        std = np.std(total_loss)\n",
    "        self.threshold = mean + self.std_factor * std\n",
    "\n",
    "        print('Validation loss mean: ', mean)\n",
    "        print('Validation loss std: ', std)\n",
    "        print('Threshold for anomaly detection: ', self.threshold)\n",
    "        \n",
    "        return self.threshold\n",
    "\n",
    "    def predict(self, data_gen, visual=True):\n",
    "        \"\"\"\n",
    "        Predicts probabilities and loss for given sequences.\n",
    "        \"\"\"\n",
    "        loss = []\n",
    "        predictions = []\n",
    "        num_displayed = 0\n",
    "        \n",
    "        for seq, l in data_gen:\n",
    "            batch_loss, alphas = self._predict_for_request(seq, l)\n",
    "            loss.extend(batch_loss)\n",
    "            alphas = self._process_alphas(seq, alphas, 1)\n",
    "            mask = np.array([l > self.threshold for l in batch_loss])\n",
    "            final_pred = mask.astype(int)\n",
    "            predictions.extend(final_pred)\n",
    "            \n",
    "            if visual and num_displayed < 10 and final_pred == [1]:\n",
    "                print('\\n\\nPrediction: ', final_pred[0])\n",
    "                print('Loss ', batch_loss[0])\n",
    "                \n",
    "                num_displayed += 1 \n",
    "                self._visual(alphas, seq)\n",
    "        \n",
    "        return predictions, loss\n",
    "\n",
    "    def _predict_for_request(self, X, l):\n",
    "        \"\"\"\n",
    "        Predicts probabilities and loss for given data. \n",
    "        \"\"\"\n",
    "        lengths = [l]\n",
    "        max_seq_len = l\n",
    "        feed_dict = {\n",
    "            self.inputs: X,\n",
    "            self.targets: X,\n",
    "            self.lengths: lengths,\n",
    "            self.dropout: 1.0,\n",
    "            self.batch_size_tensor: 1,\n",
    "            self.seq_len_tensor: max_seq_len}\n",
    "\n",
    "        fetches = [self.get_batch_loss, self.get_probabilities]\n",
    "        batch_loss, alphas = self.sess.run(fetches, feed_dict=feed_dict)\n",
    "\n",
    "        return batch_loss, alphas\n",
    "\n",
    "    def _process_alphas(self, X, alphas, batch_size):\n",
    "        \"\"\"\n",
    "        Counts numbers as probabilities for given data sample.\n",
    "        \"\"\"\n",
    "        processed_alphas = []\n",
    "        for i in range(batch_size):\n",
    "            probs = alphas[i]\n",
    "            coefs = np.array([probs[j][X[i][j]] for j in range(len(X[i]))])\n",
    "            coefs = coefs / coefs.max()\n",
    "            processed_alphas.append(coefs)\n",
    "            \n",
    "        return processed_alphas\n",
    "\n",
    "    def _visual(self, alphas, X):\n",
    "        \"\"\"\n",
    "        Colors sequence of malicious characters.\n",
    "        \"\"\"\n",
    "        for i, x in enumerate(X):\n",
    "            coefs = alphas[i]\n",
    "            tokens = self.vocab.int_to_string(x)\n",
    "            \n",
    "            for j in range(len(x)):\n",
    "                token = tokens[j]\n",
    "                if coefs[j] < 0.09:\n",
    "                    c = Fore.GREEN\n",
    "                else:\n",
    "                    c = Fore.BLACK\n",
    "                if token != '<PAD>' and token != '<EOS>':\n",
    "                    token = ''.join(c + token)\n",
    "                    print(token, end='')\n",
    "                    \n",
    "            print(Fore.BLACK + '', end='')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Restoring parameters from ./checkpoints/seq2seq-4805\n"
     ]
    }
   ],
   "source": [
    "p = Predictor(params[\"checkpoints\"], params[\"std_factor\"], params[\"vocab\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation loss mean:  0.07533215\n",
      "Validation loss std:  0.108923055\n",
      "Threshold for anomaly detection:  0.7288704812526703\n"
     ]
    }
   ],
   "source": [
    "val_gen = d.val_generator()\n",
    "threshold = p.set_threshold(val_gen)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Benign samples "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here FP samples are showed and FP rate is computed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "\n",
      "Prediction:  1\n",
      "Loss  0.946472\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[30m/\u001b[32mP\u001b[32me\u001b[32m-\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32m-\u001b[32m7\u001b[32m-\u001b[32ms\u001b[32mt\u001b[32mr\u001b[30mo\u001b[32mk\u001b[32me\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[30mf\u001b[32m?\u001b[32md\u001b[32m7\u001b[32my\u001b[32mf\u001b[32m1\u001b[30mv\u001b[32m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[30mv\u001b[30me\u001b[30m\n",
      "\u001b[32mO\u001b[32mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[30m4\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m2\u001b[30m8\u001b[30m2\u001b[30m.\u001b[30m1\u001b[30m8\u001b[30m6\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[32m*\u001b[30m/\u001b[30m*\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mp\u001b[30me\u001b[32m-\u001b[32mi\u001b[32mc\u001b[30mo\u001b[30mn\u001b[32m-\u001b[32m7\u001b[32m-\u001b[32ms\u001b[32mt\u001b[32mr\u001b[30mo\u001b[32mk\u001b[30me\u001b[32m.\u001b[32mc\u001b[30ms\u001b[32ms\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mj\u001b[30m1\u001b[32mp\u001b[30ma\u001b[32mv\u001b[32mg\u001b[30ml\u001b[32mp\u001b[32m5\u001b[32mu\u001b[32me\u001b[32m3\u001b[30m0\u001b[30m2\u001b[30m6\u001b[32m6\u001b[32mc\u001b[30m0\u001b[32mj\u001b[30m8\u001b[30m8\u001b[30mg\u001b[30me\u001b[30md\u001b[30m3\u001b[30m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  0.946472\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[30m/\u001b[32mP\u001b[32me\u001b[32m-\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32m-\u001b[32m7\u001b[32m-\u001b[32ms\u001b[32mt\u001b[32mr\u001b[30mo\u001b[32mk\u001b[32me\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[30mf\u001b[32m?\u001b[32md\u001b[32m7\u001b[32my\u001b[32mf\u001b[32m1\u001b[30mv\u001b[32m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[30mv\u001b[30me\u001b[30m\n",
      "\u001b[32mO\u001b[32mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[30m4\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m2\u001b[30m8\u001b[30m2\u001b[30m.\u001b[30m1\u001b[30m8\u001b[30m6\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[32m*\u001b[30m/\u001b[30m*\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mp\u001b[30me\u001b[32m-\u001b[32mi\u001b[32mc\u001b[30mo\u001b[30mn\u001b[32m-\u001b[32m7\u001b[32m-\u001b[32ms\u001b[32mt\u001b[32mr\u001b[30mo\u001b[32mk\u001b[30me\u001b[32m.\u001b[32mc\u001b[30ms\u001b[32ms\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mj\u001b[30m1\u001b[32mp\u001b[30ma\u001b[32mv\u001b[32mg\u001b[30ml\u001b[32mp\u001b[32m5\u001b[32mu\u001b[32me\u001b[32m3\u001b[30m0\u001b[30m2\u001b[30m6\u001b[32m6\u001b[32mc\u001b[30m0\u001b[32mj\u001b[30m8\u001b[30m8\u001b[30mg\u001b[30me\u001b[30md\u001b[30m3\u001b[30m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  1.1683424\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[30m/\u001b[32mg\u001b[32ml\u001b[32my\u001b[32mp\u001b[32mh\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32ms\u001b[32m-\u001b[32mh\u001b[32ma\u001b[32ml\u001b[32mf\u001b[32ml\u001b[32mi\u001b[32mn\u001b[32mg\u001b[30ms\u001b[32m-\u001b[32mr\u001b[32me\u001b[32mg\u001b[32mu\u001b[32ml\u001b[32ma\u001b[32mr\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[32mf\u001b[32m2\u001b[30m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[30m/\u001b[32m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[30mv\u001b[30me\u001b[30m\n",
      "\u001b[32mO\u001b[32mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[30m4\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m2\u001b[30m8\u001b[30m2\u001b[30m.\u001b[30m1\u001b[30m8\u001b[30m6\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[32m*\u001b[30m/\u001b[30m*\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mb\u001b[30mo\u001b[32mo\u001b[30mt\u001b[32ms\u001b[32mt\u001b[32mr\u001b[32ma\u001b[32mp\u001b[32m.\u001b[32mm\u001b[32mi\u001b[32mn\u001b[32m.\u001b[32mc\u001b[32ms\u001b[32ms\u001b[32m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mj\u001b[30m1\u001b[32mp\u001b[30ma\u001b[32mv\u001b[32mg\u001b[30ml\u001b[32mp\u001b[32m5\u001b[32mu\u001b[32me\u001b[32m3\u001b[30m0\u001b[30m2\u001b[30m6\u001b[32m6\u001b[32mc\u001b[30m0\u001b[32mj\u001b[30m8\u001b[30m8\u001b[30mg\u001b[30me\u001b[30md\u001b[30m3\u001b[30m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  2.4811156\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[32m/\u001b[32mg\u001b[32ml\u001b[32my\u001b[32mp\u001b[32mh\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32ms\u001b[32m-\u001b[32mh\u001b[32ma\u001b[32ml\u001b[32mf\u001b[32ml\u001b[32mi\u001b[32mn\u001b[32mg\u001b[30ms\u001b[32m-\u001b[32mr\u001b[32me\u001b[32mg\u001b[32mu\u001b[32ml\u001b[32ma\u001b[32mr\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[32mf\u001b[32m2\u001b[30m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[32mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[32mU\u001b[32mb\u001b[30mu\u001b[30mn\u001b[30mt\u001b[30mu\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m;\u001b[30m \u001b[30mr\u001b[30mv\u001b[30m:\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m)\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m/\u001b[30m2\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m1\u001b[30m \u001b[30mF\u001b[30mi\u001b[30mr\u001b[30me\u001b[30mf\u001b[30mo\u001b[30mx\u001b[30m/\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[32mf\u001b[30mo\u001b[32mn\u001b[32mt\u001b[32m-\u001b[30mw\u001b[30mo\u001b[32mf\u001b[32mf\u001b[32m2\u001b[32m;\u001b[32mq\u001b[32m=\u001b[32m1\u001b[32m.\u001b[32m0\u001b[32m,\u001b[32ma\u001b[32mp\u001b[32mp\u001b[32ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[32mi\u001b[32mo\u001b[32mn\u001b[32m/\u001b[32mf\u001b[32mo\u001b[32mn\u001b[32mt\u001b[32m-\u001b[32mw\u001b[32mo\u001b[32mf\u001b[32mf\u001b[32m;\u001b[32mq\u001b[32m=\u001b[32m0\u001b[30m.\u001b[30m9\u001b[32m,\u001b[32m*\u001b[30m/\u001b[32m*\u001b[32m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m8\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[32mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m5\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[32mi\u001b[32md\u001b[32me\u001b[32mn\u001b[32mt\u001b[32mi\u001b[32mt\u001b[32my\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[32m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[32m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mb\u001b[30mo\u001b[32mo\u001b[30mt\u001b[32ms\u001b[32mt\u001b[32mr\u001b[32ma\u001b[32mp\u001b[32m.\u001b[32mm\u001b[32mi\u001b[32mn\u001b[32m.\u001b[32mc\u001b[32ms\u001b[32ms\u001b[32m\n",
      "\u001b[32mC\u001b[32mo\u001b[32mo\u001b[32mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[32mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mm\u001b[32ml\u001b[32ma\u001b[32mc\u001b[32ms\u001b[32m0\u001b[32mu\u001b[32mi\u001b[32mo\u001b[32mu\u001b[32m3\u001b[32m4\u001b[32m4\u001b[32mi\u001b[32m3\u001b[32mf\u001b[32ma\u001b[32m5\u001b[32m3\u001b[32ms\u001b[32m7\u001b[32mr\u001b[32ma\u001b[32mu\u001b[32mt\u001b[32m6\u001b[32m\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[32mv\u001b[30me\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  0.946472\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[30m/\u001b[32mP\u001b[32me\u001b[32m-\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32m-\u001b[32m7\u001b[32m-\u001b[32ms\u001b[32mt\u001b[32mr\u001b[30mo\u001b[32mk\u001b[32me\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[30mf\u001b[32m?\u001b[32md\u001b[32m7\u001b[32my\u001b[32mf\u001b[32m1\u001b[30mv\u001b[32m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[30mv\u001b[30me\u001b[30m\n",
      "\u001b[32mO\u001b[32mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[30m4\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m2\u001b[30m8\u001b[30m2\u001b[30m.\u001b[30m1\u001b[30m8\u001b[30m6\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[32m*\u001b[30m/\u001b[30m*\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mp\u001b[30me\u001b[32m-\u001b[32mi\u001b[32mc\u001b[30mo\u001b[30mn\u001b[32m-\u001b[32m7\u001b[32m-\u001b[32ms\u001b[32mt\u001b[32mr\u001b[30mo\u001b[32mk\u001b[30me\u001b[32m.\u001b[32mc\u001b[30ms\u001b[32ms\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mj\u001b[30m1\u001b[32mp\u001b[30ma\u001b[32mv\u001b[32mg\u001b[30ml\u001b[32mp\u001b[32m5\u001b[32mu\u001b[32me\u001b[32m3\u001b[30m0\u001b[30m2\u001b[30m6\u001b[32m6\u001b[32mc\u001b[30m0\u001b[32mj\u001b[30m8\u001b[30m8\u001b[30mg\u001b[30me\u001b[30md\u001b[30m3\u001b[30m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  1.1683424\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[30m/\u001b[32mg\u001b[32ml\u001b[32my\u001b[32mp\u001b[32mh\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32ms\u001b[32m-\u001b[32mh\u001b[32ma\u001b[32ml\u001b[32mf\u001b[32ml\u001b[32mi\u001b[32mn\u001b[32mg\u001b[30ms\u001b[32m-\u001b[32mr\u001b[32me\u001b[32mg\u001b[32mu\u001b[32ml\u001b[32ma\u001b[32mr\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[32mf\u001b[32m2\u001b[30m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[30m/\u001b[32m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[30mv\u001b[30me\u001b[30m\n",
      "\u001b[32mO\u001b[32mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[30m4\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m2\u001b[30m8\u001b[30m2\u001b[30m.\u001b[30m1\u001b[30m8\u001b[30m6\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[32m*\u001b[30m/\u001b[30m*\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mb\u001b[30mo\u001b[32mo\u001b[30mt\u001b[32ms\u001b[32mt\u001b[32mr\u001b[32ma\u001b[32mp\u001b[32m.\u001b[32mm\u001b[32mi\u001b[32mn\u001b[32m.\u001b[32mc\u001b[32ms\u001b[32ms\u001b[32m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mj\u001b[30m1\u001b[32mp\u001b[30ma\u001b[32mv\u001b[32mg\u001b[30ml\u001b[32mp\u001b[32m5\u001b[32mu\u001b[32me\u001b[32m3\u001b[30m0\u001b[30m2\u001b[30m6\u001b[32m6\u001b[32mc\u001b[30m0\u001b[32mj\u001b[30m8\u001b[30m8\u001b[30mg\u001b[30me\u001b[30md\u001b[30m3\u001b[30m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  1.1683424\n",
      "\u001b[30mG\u001b[30mE\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mf\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30ms\u001b[30m/\u001b[32mg\u001b[32ml\u001b[32my\u001b[32mp\u001b[32mh\u001b[32mi\u001b[32mc\u001b[30mo\u001b[32mn\u001b[32ms\u001b[32m-\u001b[32mh\u001b[32ma\u001b[32ml\u001b[32mf\u001b[32ml\u001b[32mi\u001b[32mn\u001b[32mg\u001b[30ms\u001b[32m-\u001b[32mr\u001b[32me\u001b[32mg\u001b[32mu\u001b[32ml\u001b[32ma\u001b[32mr\u001b[32m.\u001b[32mw\u001b[32mo\u001b[32mf\u001b[32mf\u001b[32m2\u001b[30m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[30m/\u001b[32m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[30mv\u001b[30me\u001b[30m\n",
      "\u001b[32mO\u001b[32mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[30m4\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m2\u001b[30m8\u001b[30m2\u001b[30m.\u001b[30m1\u001b[30m8\u001b[30m6\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[32m*\u001b[30m/\u001b[30m*\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[32ma\u001b[32ms\u001b[32ms\u001b[30me\u001b[30mt\u001b[32ms\u001b[30m/\u001b[32mc\u001b[30ms\u001b[32ms\u001b[32m/\u001b[32mb\u001b[30mo\u001b[32mo\u001b[30mt\u001b[32ms\u001b[32mt\u001b[32mr\u001b[32ma\u001b[32mp\u001b[32m.\u001b[32mm\u001b[32mi\u001b[32mn\u001b[32m.\u001b[32mc\u001b[32ms\u001b[32ms\u001b[32m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mj\u001b[30m1\u001b[32mp\u001b[30ma\u001b[32mv\u001b[32mg\u001b[30ml\u001b[32mp\u001b[32m5\u001b[32mu\u001b[32me\u001b[32m3\u001b[30m0\u001b[30m2\u001b[30m6\u001b[32m6\u001b[32mc\u001b[30m0\u001b[32mj\u001b[30m8\u001b[30m8\u001b[30mg\u001b[30me\u001b[30md\u001b[30m3\u001b[30m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  0.7708773\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ma\u001b[30mp\u001b[30mi\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m \u001b[30mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[32mU\u001b[32mb\u001b[30mu\u001b[30mn\u001b[30mt\u001b[30mu\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m;\u001b[30m \u001b[30mr\u001b[30mv\u001b[30m:\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m)\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m/\u001b[30m2\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m1\u001b[30m \u001b[30mF\u001b[30mi\u001b[30mr\u001b[30me\u001b[30mf\u001b[30mo\u001b[30mx\u001b[30m/\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m5\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30mt\u001b[30mr\u001b[30ma\u001b[30mn\u001b[30ms\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30ms\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mx\u001b[30m-\u001b[30mw\u001b[30mw\u001b[30mw\u001b[30m-\u001b[30mf\u001b[30mo\u001b[30mr\u001b[30mm\u001b[30m-\u001b[30mu\u001b[30mr\u001b[30ml\u001b[30me\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30me\u001b[30md\u001b[30m;\u001b[30m \u001b[30mc\u001b[30mh\u001b[30ma\u001b[30mr\u001b[30ms\u001b[30me\u001b[30mt\u001b[30m=\u001b[30mU\u001b[30mT\u001b[30mF\u001b[30m-\u001b[30m8\u001b[30m\n",
      "\u001b[32mX\u001b[30m-\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30me\u001b[30md\u001b[30m-\u001b[30mW\u001b[30mi\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30mX\u001b[30mM\u001b[30mL\u001b[30mH\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30m1\u001b[32m1\u001b[32m2\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[30mc\u001b[32mh\u001b[32m1\u001b[32mu\u001b[32mp\u001b[32mt\u001b[32m4\u001b[30ml\u001b[32mk\u001b[32m9\u001b[30mo\u001b[32mn\u001b[30mm\u001b[30m6\u001b[32me\u001b[32m2\u001b[32m6\u001b[30ml\u001b[32mo\u001b[30m4\u001b[32ml\u001b[32mk\u001b[32mc\u001b[30m8\u001b[30m9\u001b[30m3\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[32mv\u001b[30me\u001b[30m\n",
      "\u001b[32m\n",
      "\u001b[30mt\u001b[30my\u001b[30mp\u001b[30me\u001b[30m=\u001b[30mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m&\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m=\u001b[30mc\u001b[30mh\u001b[30me\u001b[30mc\u001b[30mk\u001b[30m&\u001b[30mf\u001b[30mi\u001b[30mr\u001b[30ms\u001b[30mt\u001b[30mn\u001b[30ma\u001b[30mm\u001b[30me\u001b[30m=\u001b[32m&\u001b[30ml\u001b[30ma\u001b[30ms\u001b[30mt\u001b[30mn\u001b[30ma\u001b[30mm\u001b[30me\u001b[30m=\u001b[32mP\u001b[32ml\u001b[30ma\u001b[32mn\u001b[32mt\u001b[30m&\u001b[32mc\u001b[30mr\u001b[30me\u001b[30md\u001b[30mi\u001b[30mt\u001b[30mc\u001b[30ma\u001b[30mr\u001b[30md\u001b[30m=\u001b[32m1\u001b[32m2\u001b[30m3\u001b[32m4\u001b[30m-\u001b[32m6\u001b[32m5\u001b[30m7\u001b[32m9\u001b[30m-\u001b[32m3\u001b[32m5\u001b[32m1\u001b[30m4\u001b[30m-\u001b[30m6\u001b[32m4\u001b[32m7\u001b[32m5\u001b[30m&\u001b[32mr\u001b[30me\u001b[30mc\u001b[30mi\u001b[30mp\u001b[30mi\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m=\u001b[30mD\u001b[30mE\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m1\u001b[32m6\u001b[32m5\u001b[32m6\u001b[32m5\u001b[32m6\u001b[30m7\u001b[32m4\u001b[32m7\u001b[32m4\u001b[32m7\u001b[32m8\u001b[32m3\u001b[32m8\u001b[32m3\u001b[32m8\u001b[30m"
     ]
    }
   ],
   "source": [
    "test_gen = d.test_generator()\n",
    "valid_preds, valid_loss = p.predict(test_gen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of FP:  8\n",
      "Number of samples:  2200\n",
      "FP rate: 0.0036\n"
     ]
    }
   ],
   "source": [
    "print('Number of FP: ', np.sum(valid_preds))\n",
    "print('Number of samples: ', len(valid_preds))\n",
    "print('FP rate: {:.4f}'.format(np.sum(valid_preds) / len(valid_preds)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Anomalous samples "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here TP samples are showed and TP rate is computed."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloaded 1097 samples\n",
      "\n",
      "\n",
      "Prediction:  1\n",
      "Loss  2.16642\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ma\u001b[30mp\u001b[30mi\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m \u001b[30mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[32mU\u001b[32mb\u001b[30mu\u001b[30mn\u001b[30mt\u001b[30mu\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m;\u001b[30m \u001b[30mr\u001b[30mv\u001b[30m:\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m)\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m/\u001b[30m2\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m1\u001b[30m \u001b[30mF\u001b[30mi\u001b[30mr\u001b[30me\u001b[30mf\u001b[30mo\u001b[30mx\u001b[30m/\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m5\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ml\u001b[30mo\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mx\u001b[30m-\u001b[30mw\u001b[30mw\u001b[30mw\u001b[30m-\u001b[30mf\u001b[30mo\u001b[30mr\u001b[30mm\u001b[30m-\u001b[30mu\u001b[30mr\u001b[30ml\u001b[30me\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30me\u001b[30md\u001b[30m;\u001b[30m \u001b[30mc\u001b[30mh\u001b[30ma\u001b[30mr\u001b[30ms\u001b[30me\u001b[30mt\u001b[30m=\u001b[30mU\u001b[30mT\u001b[30mF\u001b[30m-\u001b[30m8\u001b[30m\n",
      "\u001b[32mX\u001b[30m-\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30me\u001b[30md\u001b[30m-\u001b[30mW\u001b[30mi\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30mX\u001b[30mM\u001b[30mL\u001b[30mH\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[32m2\u001b[32m0\u001b[32m9\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[32mk\u001b[32mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mm\u001b[32ml\u001b[30ma\u001b[32mc\u001b[32ms\u001b[32m0\u001b[32mu\u001b[32mi\u001b[32mo\u001b[32mu\u001b[32m3\u001b[32m4\u001b[32m4\u001b[32mi\u001b[32m3\u001b[32mf\u001b[32ma\u001b[32m5\u001b[32m3\u001b[32ms\u001b[32m7\u001b[32mr\u001b[32ma\u001b[32mu\u001b[32mt\u001b[32m6\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[32me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[32mv\u001b[30me\u001b[30m\n",
      "\u001b[32m\n",
      "\u001b[30mt\u001b[30my\u001b[30mp\u001b[30me\u001b[30m=\u001b[30mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m&\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m=\u001b[32mc\u001b[32mr\u001b[30me\u001b[32ma\u001b[30mt\u001b[32me\u001b[32m&\u001b[32mu\u001b[32ms\u001b[30me\u001b[30mr\u001b[32mn\u001b[32ma\u001b[32mm\u001b[32me\u001b[30m=\u001b[32mJ\u001b[30ma\u001b[32mc\u001b[32mk\u001b[32m'\u001b[32m+\u001b[32ma\u001b[32mn\u001b[32md\u001b[32m+\u001b[32me\u001b[32mx\u001b[32mt\u001b[30mr\u001b[32ma\u001b[32mc\u001b[32mt\u001b[32mv\u001b[32ma\u001b[32ml\u001b[32mu\u001b[32me\u001b[32m(\u001b[32m0\u001b[32mx\u001b[32m0\u001b[32ma\u001b[32m,\u001b[32mc\u001b[32mo\u001b[32mn\u001b[32mc\u001b[32ma\u001b[32mt\u001b[32m(\u001b[30m0\u001b[32mx\u001b[32m0\u001b[32ma\u001b[32m,\u001b[32m(\u001b[32ms\u001b[32me\u001b[32ml\u001b[32me\u001b[32mc\u001b[32mt\u001b[32m \u001b[32mv\u001b[32me\u001b[32mr\u001b[32ms\u001b[32mi\u001b[32mo\u001b[32mn\u001b[32m(\u001b[32m)\u001b[32m)\u001b[32m)\u001b[32m)\u001b[32m+\u001b[32ma\u001b[32mn\u001b[32md\u001b[32m+\u001b[32m'\u001b[30m1\u001b[32m'\u001b[32m=\u001b[32m'\u001b[32m1\u001b[32m&\u001b[32mp\u001b[32ma\u001b[30ms\u001b[30ms\u001b[32mw\u001b[32mo\u001b[30mr\u001b[32md\u001b[30m=\u001b[32mp\u001b[32ma\u001b[32ms\u001b[32ms\u001b[32mw\u001b[32m0\u001b[32mr\u001b[32md\u001b[32m&\u001b[32mf\u001b[32mi\u001b[32mr\u001b[30ms\u001b[30mt\u001b[32mn\u001b[30ma\u001b[32mm\u001b[32me\u001b[30m=\u001b[32mf\u001b[32mi\u001b[32mr\u001b[32ms\u001b[32mt\u001b[32m&\u001b[32ml\u001b[32ma\u001b[32ms\u001b[32mt\u001b[32mn\u001b[32ma\u001b[32mm\u001b[32me\u001b[32m=\u001b[32ml\u001b[32ma\u001b[32ms\u001b[32mt\u001b[32m&\u001b[32mb\u001b[30mi\u001b[30mr\u001b[30mt\u001b[32mh\u001b[32md\u001b[30ma\u001b[32mt\u001b[30me\u001b[32m=\u001b[32m3\u001b[32m0\u001b[32m-\u001b[32m0\u001b[32m8\u001b[32m-\u001b[32m2\u001b[32m0\u001b[30m1\u001b[32m7\u001b[32m&\u001b[32me\u001b[32mm\u001b[32ma\u001b[32mi\u001b[32ml\u001b[32m=\u001b[32me\u001b[32me\u001b[32me\u001b[32m%\u001b[32m4\u001b[32m0\u001b[32mm\u001b[32ma\u001b[32mi\u001b[32ml\u001b[32m.\u001b[32mc\u001b[32mo\u001b[32mm\u001b[32m&\u001b[32mp\u001b[32mh\u001b[30mo\u001b[30mn\u001b[30me\u001b[32m=\u001b[32m7\u001b[32m4\u001b[32m7\u001b[32m4\u001b[32m7\u001b[32m4\u001b[32m7\u001b[32m4\u001b[32m7\u001b[32m&\u001b[32ma\u001b[32mc\u001b[32mc\u001b[32mo\u001b[32mu\u001b[30mn\u001b[30mt\u001b[30m=\u001b[32mD\u001b[32mE\u001b[32m4\u001b[32m4\u001b[32m4\u001b[32m0\u001b[32m4\u001b[32m4\u001b[32m1\u001b[32m9\u001b[32m5\u001b[30m6\u001b[32m9\u001b[32m7\u001b[32m5\u001b[32m0\u001b[32m5\u001b[32m5\u001b[32m3\u001b[32m3\u001b[32m4\u001b[30m0\u001b[32m&\u001b[30mc\u001b[30mr\u001b[30me\u001b[30md\u001b[30mi\u001b[30mt\u001b[30mc\u001b[30ma\u001b[30mr\u001b[30md\u001b[30m=\u001b[32m4\u001b[30m5\u001b[32m5\u001b[32m6\u001b[30m-\u001b[32m9\u001b[30m3\u001b[30m7\u001b[32m3\u001b[30m-\u001b[32m3\u001b[30m9\u001b[32m1\u001b[32m3\u001b[32m-\u001b[32m6\u001b[32m5\u001b[30m1\u001b[32m0\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  2.2834342\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ma\u001b[30mp\u001b[30mi\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m \u001b[30mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[32mU\u001b[32mb\u001b[30mu\u001b[30mn\u001b[30mt\u001b[30mu\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m;\u001b[30m \u001b[30mr\u001b[30mv\u001b[30m:\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m)\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m/\u001b[30m2\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m1\u001b[30m \u001b[30mF\u001b[30mi\u001b[30mr\u001b[30me\u001b[30mf\u001b[30mo\u001b[30mx\u001b[30m/\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m5\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ml\u001b[30mo\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mx\u001b[30m-\u001b[30mw\u001b[30mw\u001b[30mw\u001b[30m-\u001b[30mf\u001b[30mo\u001b[30mr\u001b[30mm\u001b[30m-\u001b[30mu\u001b[30mr\u001b[30ml\u001b[30me\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30me\u001b[30md\u001b[30m\n",
      "\u001b[32mX\u001b[30m-\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30me\u001b[30md\u001b[30m-\u001b[30mW\u001b[30mi\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30mX\u001b[30mM\u001b[30mL\u001b[30mH\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[32m7\u001b[30m6\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32mm\u001b[32ml\u001b[32ma\u001b[32mc\u001b[32ms\u001b[32m0\u001b[32mu\u001b[32mi\u001b[32mo\u001b[32mu\u001b[32m3\u001b[32m4\u001b[32m4\u001b[32mi\u001b[32m3\u001b[32mf\u001b[32ma\u001b[32m5\u001b[32m3\u001b[32ms\u001b[32m7\u001b[32mr\u001b[32ma\u001b[32mu\u001b[32mt\u001b[32m6\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[32me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[32mv\u001b[30me\u001b[30m\n",
      "\u001b[32m\n",
      "\u001b[30mt\u001b[30my\u001b[30mp\u001b[30me\u001b[30m=\u001b[30mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m&\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m=\u001b[32ml\u001b[30mo\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m&\u001b[30mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30mn\u001b[30ma\u001b[30mm\u001b[30me\u001b[30m=\u001b[32mn\u001b[32mo\u001b[32mn\u001b[32me\u001b[32m'\u001b[32m+\u001b[32mu\u001b[32mn\u001b[32mi\u001b[32mo\u001b[32mn\u001b[32m+\u001b[32ms\u001b[32me\u001b[32ml\u001b[32me\u001b[32mc\u001b[32mt\u001b[32m+\u001b[32m1\u001b[32m,\u001b[32m2\u001b[32m,\u001b[32ml\u001b[32mo\u001b[32mg\u001b[32mi\u001b[32mn\u001b[32m,\u001b[32mp\u001b[30ma\u001b[32ms\u001b[32ms\u001b[32mw\u001b[32mo\u001b[32mr\u001b[32md\u001b[32m,\u001b[32m5\u001b[32m,\u001b[32m6\u001b[32m,\u001b[32m7\u001b[32m,\u001b[32mN\u001b[32mU\u001b[32mL\u001b[32mL\u001b[32m,\u001b[32mN\u001b[32mU\u001b[32mL\u001b[32mL\u001b[32m,\u001b[32m1\u001b[32m0\u001b[32m,\u001b[32m1\u001b[32m1\u001b[32m,\u001b[32m1\u001b[32m2\u001b[32m,\u001b[32m1\u001b[32m3\u001b[32m,\u001b[32m1\u001b[32m4\u001b[32m,\u001b[32m1\u001b[32m5\u001b[32m,\u001b[32m1\u001b[32m6\u001b[32m,\u001b[32m1\u001b[32m7\u001b[32m+\u001b[32mf\u001b[32mr\u001b[32mo\u001b[32mm\u001b[32m+\u001b[32mu\u001b[32ms\u001b[32me\u001b[32mr\u001b[32ms\u001b[32m+\u001b[32ml\u001b[32mi\u001b[32mm\u001b[32mi\u001b[32mt\u001b[32m+\u001b[32m1\u001b[32m+\u001b[32m-\u001b[32m-\u001b[32m1\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  0.9289187\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ma\u001b[30mp\u001b[30mi\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m \u001b[30mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m\n",
      "\u001b[32mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[32mU\u001b[32mb\u001b[30mu\u001b[30mn\u001b[30mt\u001b[30mu\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m;\u001b[30m \u001b[30mr\u001b[30mv\u001b[30m:\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m)\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m/\u001b[30m2\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m1\u001b[30m \u001b[30mF\u001b[30mi\u001b[30mr\u001b[30me\u001b[30mf\u001b[30mo\u001b[30mx\u001b[30m/\u001b[30m5\u001b[32m9\u001b[30m.\u001b[30m0\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m5\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[30m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30mt\u001b[30mr\u001b[30ma\u001b[30mn\u001b[30ms\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30ms\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mx\u001b[30m-\u001b[30mw\u001b[30mw\u001b[30mw\u001b[30m-\u001b[30mf\u001b[30mo\u001b[30mr\u001b[30mm\u001b[30m-\u001b[30mu\u001b[30mr\u001b[30ml\u001b[30me\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30me\u001b[30md\u001b[30m;\u001b[30m \u001b[30mc\u001b[30mh\u001b[30ma\u001b[30mr\u001b[30ms\u001b[30me\u001b[30mt\u001b[30m=\u001b[30mU\u001b[30mT\u001b[30mF\u001b[30m-\u001b[30m8\u001b[30m\n",
      "\u001b[32mX\u001b[30m-\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30me\u001b[30md\u001b[30m-\u001b[30mW\u001b[30mi\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30mX\u001b[30mM\u001b[30mL\u001b[30mH\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m6\u001b[30m2\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mo\u001b[30mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[30mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[30mc\u001b[32mh\u001b[32m1\u001b[32mu\u001b[32mp\u001b[32mt\u001b[32m4\u001b[32ml\u001b[32mk\u001b[32m9\u001b[30mo\u001b[32mn\u001b[30mm\u001b[30m6\u001b[32me\u001b[32m2\u001b[32m6\u001b[30ml\u001b[30mo\u001b[30m4\u001b[32ml\u001b[32mk\u001b[32mc\u001b[30m8\u001b[30m9\u001b[30m3\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[30me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[30mi\u001b[32mv\u001b[30me\u001b[30m\n",
      "\u001b[32m\n",
      "\u001b[30mt\u001b[30my\u001b[30mp\u001b[30me\u001b[30m=\u001b[30mt\u001b[30mr\u001b[30ma\u001b[30mn\u001b[30ms\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m&\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m=\u001b[30ms\u001b[30me\u001b[30mn\u001b[30md\u001b[30m&\u001b[30ms\u001b[30me\u001b[30mn\u001b[30md\u001b[30me\u001b[30mr\u001b[30m=\u001b[30mD\u001b[30mE\u001b[30m0\u001b[30m0\u001b[30m0\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m2\u001b[30m2\u001b[30m2\u001b[30m2\u001b[30m2\u001b[30m3\u001b[30m3\u001b[30m3\u001b[30m3\u001b[30m3\u001b[30m&\u001b[30mr\u001b[30me\u001b[30mc\u001b[30mi\u001b[30mp\u001b[30mi\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m=\u001b[30mD\u001b[30mE\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m1\u001b[30m1\u001b[32m6\u001b[32m5\u001b[32m6\u001b[32m5\u001b[32m6\u001b[30m7\u001b[32m4\u001b[32m7\u001b[32m4\u001b[32m7\u001b[32m8\u001b[32m3\u001b[32m8\u001b[32m3\u001b[32m8\u001b[32m&\u001b[30mc\u001b[32mr\u001b[30me\u001b[30md\u001b[30mi\u001b[30mt\u001b[30mc\u001b[30ma\u001b[30mr\u001b[30md\u001b[30m=\u001b[32m1\u001b[32m2\u001b[30m3\u001b[32m4\u001b[32m-\u001b[32m6\u001b[32m5\u001b[30m7\u001b[32m9\u001b[30m-\u001b[30m3\u001b[32m5\u001b[32m1\u001b[32m4\u001b[32m-\u001b[30m6\u001b[32m4\u001b[32m7\u001b[32m5\u001b[30m&\u001b[30ma\u001b[30mm\u001b[30mo\u001b[30mu\u001b[30mn\u001b[30mt\u001b[30m=\u001b[32m2\u001b[30m&\u001b[30mc\u001b[30mo\u001b[30mm\u001b[30mm\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m=\u001b[32m<\u001b[32ms\u001b[32mv\u001b[32mg\u001b[32m \u001b[30mo\u001b[32mn\u001b[32ml\u001b[32mo\u001b[32ma\u001b[32md\u001b[32m=\u001b[30ma\u001b[32ml\u001b[32me\u001b[32mr\u001b[32mt\u001b[32m(\u001b[32m1\u001b[32m)\u001b[32m>\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  8.67965\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[32mb\u001b[30ma\u001b[32mn\u001b[32mk\u001b[32m/\u001b[32ma\u001b[30mp\u001b[30mi\u001b[32m.\u001b[30mp\u001b[32mh\u001b[30mp\u001b[32m?\u001b[32mx\u001b[32mm\u001b[32ml\u001b[32m \u001b[32mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[32m1\u001b[32m\n",
      "\u001b[32mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[32mv\u001b[32mu\u001b[32ml\u001b[32mn\u001b[32mb\u001b[32ma\u001b[32mn\u001b[32mk\u001b[30m.\u001b[32mc\u001b[32mo\u001b[32mm\u001b[32m\n",
      "\u001b[32mU\u001b[32ms\u001b[30me\u001b[30mr\u001b[32m-\u001b[32mA\u001b[32mg\u001b[30me\u001b[32mn\u001b[32mt\u001b[32m:\u001b[32m \u001b[32mM\u001b[30mo\u001b[32mz\u001b[32mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[32m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[30mW\u001b[30mi\u001b[30mn\u001b[30md\u001b[30mo\u001b[30mw\u001b[30ms\u001b[30m \u001b[30mN\u001b[30mT\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m;\u001b[30m \u001b[30mW\u001b[30mi\u001b[30mn\u001b[30m6\u001b[30m4\u001b[30m;\u001b[30m \u001b[30mx\u001b[30m6\u001b[30m4\u001b[32m;\u001b[30m \u001b[30mr\u001b[30mv\u001b[30m:\u001b[30m5\u001b[32m2\u001b[30m.\u001b[30m0\u001b[30m)\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m/\u001b[30m2\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m0\u001b[30m1\u001b[30m0\u001b[30m1\u001b[30m \u001b[30mF\u001b[30mi\u001b[30mr\u001b[30me\u001b[30mf\u001b[30mo\u001b[30mx\u001b[30m/\u001b[30m5\u001b[32m2\u001b[30m.\u001b[30m0\u001b[30m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[30m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[32mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[32mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[32mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[32mx\u001b[32mm\u001b[32ml\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[32m9\u001b[32m4\u001b[32m5\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[32mn\u001b[30m:\u001b[30m \u001b[32mc\u001b[30ml\u001b[30mo\u001b[32ms\u001b[30me\u001b[32m\n",
      "\u001b[32m\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[32m<\u001b[32m?\u001b[32mx\u001b[32mm\u001b[32ml\u001b[32m \u001b[32mv\u001b[32me\u001b[32mr\u001b[32ms\u001b[32mi\u001b[32mo\u001b[32mn\u001b[32m=\u001b[32m\"\u001b[32m1\u001b[32m.\u001b[32m0\u001b[32m\"\u001b[32m \u001b[32me\u001b[32mn\u001b[32mc\u001b[32mo\u001b[32md\u001b[32mi\u001b[32mn\u001b[32mg\u001b[32m=\u001b[32m\"\u001b[32mu\u001b[32mt\u001b[30mf\u001b[32m-\u001b[30m8\u001b[32m\"\u001b[32m?\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mD\u001b[32mO\u001b[32mC\u001b[30mT\u001b[32mY\u001b[32mP\u001b[32mE\u001b[32m \u001b[32ml\u001b[30mo\u001b[32ml\u001b[32mz\u001b[32m \u001b[32m[\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[32m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m \u001b[32m\"\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mL\u001b[32mE\u001b[32mM\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32m \u001b[32ml\u001b[32mo\u001b[32mg\u001b[32mi\u001b[32mn\u001b[32m \u001b[32m(\u001b[32m#\u001b[32mP\u001b[32mC\u001b[32mD\u001b[32mA\u001b[32mT\u001b[32mA\u001b[32m)\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[30m \u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m1\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[30m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m1\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[32m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m2\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[30m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m3\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[30m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m4\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[30m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m5\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[30m \u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m7\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m6\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[32m \u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m7\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m<\u001b[32m!\u001b[32mE\u001b[32mN\u001b[32mT\u001b[32mI\u001b[32mT\u001b[32mY\u001b[32m \u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m9\u001b[30m \u001b[32m\"\u001b[32m&\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m&\u001b[32ml\u001b[32mo\u001b[32ml\u001b[32m8\u001b[32m;\u001b[32m\"\u001b[32m>\u001b[32m]\u001b[32m>\u001b[32m<\u001b[30ma\u001b[32mp\u001b[32mi\u001b[32m>\u001b[32m<\u001b[32mt\u001b[32my\u001b[32mp\u001b[32me\u001b[32m>\u001b[32mu\u001b[32ms\u001b[30me\u001b[32mr\u001b[32m<\u001b[32m/\u001b[32mt\u001b[32my\u001b[32mp\u001b[32me\u001b[32m>\u001b[32m<\u001b[32ma\u001b[32mc\u001b[32mt\u001b[32mi\u001b[32mo\u001b[32mn\u001b[32m>\u001b[32mf\u001b[32mo\u001b[32mr\u001b[32mg\u001b[32mo\u001b[32mt\u001b[32mp\u001b[32ma\u001b[32ms\u001b[32ms\u001b[32m<\u001b[32m/\u001b[32ma\u001b[32mc\u001b[32mt\u001b[32mi\u001b[32mo\u001b[32mn\u001b[32m>\u001b[32m<\u001b[32mu\u001b[32ms\u001b[32me\u001b[32mr\u001b[32mn\u001b[32ma\u001b[32mm\u001b[32me\u001b[32m>\u001b[32mj\u001b[32m.\u001b[32md\u001b[30mo\u001b[32me\u001b[32m<\u001b[32m/\u001b[32mu\u001b[32ms\u001b[32me\u001b[32mr\u001b[32mn\u001b[32ma\u001b[32mm\u001b[32me\u001b[32m>\u001b[32m<\u001b[32mp\u001b[32ma\u001b[32ms\u001b[32ms\u001b[32mw\u001b[30mo\u001b[32mr\u001b[32md\u001b[32m>\u001b[32m&\u001b[32ml\u001b[30mo\u001b[32ml\u001b[32m9\u001b[32m;\u001b[32m<\u001b[32m/\u001b[32mp\u001b[32ma\u001b[30ms\u001b[32ms\u001b[32mw\u001b[32mo\u001b[32mr\u001b[32md\u001b[32m>\u001b[32m<\u001b[32mc\u001b[32mo\u001b[32md\u001b[32me\u001b[32m>\u001b[32m0\u001b[32m0\u001b[32m0\u001b[32m<\u001b[32m/\u001b[32mc\u001b[32mo\u001b[32md\u001b[32me\u001b[32m>\u001b[32m<\u001b[32m/\u001b[32ma\u001b[32mp\u001b[32mi\u001b[32m>\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  2.2545607\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ma\u001b[30mp\u001b[30mi\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m \u001b[30mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[32me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[32mi\u001b[32mv\u001b[30me\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30m1\u001b[32m3\u001b[32m0\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[32mO\u001b[30mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mX\u001b[30m-\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30me\u001b[30md\u001b[30m-\u001b[30mW\u001b[30mi\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30mX\u001b[30mM\u001b[30mL\u001b[30mH\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[32m6\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m3\u001b[32m5\u001b[32m9\u001b[30m.\u001b[30m1\u001b[30m3\u001b[32m9\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mx\u001b[30m-\u001b[30mw\u001b[30mw\u001b[30mw\u001b[30m-\u001b[30mf\u001b[30mo\u001b[30mr\u001b[30mm\u001b[30m-\u001b[30mu\u001b[30mr\u001b[30ml\u001b[30me\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30me\u001b[30md\u001b[30m;\u001b[30m \u001b[30mc\u001b[30mh\u001b[30ma\u001b[30mr\u001b[30ms\u001b[30me\u001b[30mt\u001b[30m=\u001b[30mU\u001b[30mT\u001b[30mF\u001b[30m-\u001b[30m8\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[32mD\u001b[32mN\u001b[32mT\u001b[32m:\u001b[30m \u001b[30m1\u001b[32m\r",
      "\u001b[32m\n",
      "\u001b[32mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[32mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30mi\u001b[30mn\u001b[30mf\u001b[30mo\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[32m\r",
      "\u001b[30m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[32m\r",
      "\u001b[32m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[32m,\u001b[32mr\u001b[32mu\u001b[32m;\u001b[32mq\u001b[32m=\u001b[30m0\u001b[30m.\u001b[30m8\u001b[32m\r",
      "\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[32mo\u001b[32mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[32mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32m7\u001b[32mb\u001b[32mg\u001b[32mg\u001b[32m5\u001b[32mt\u001b[32ma\u001b[32ml\u001b[32md\u001b[32m4\u001b[32ma\u001b[32me\u001b[32ma\u001b[32m8\u001b[32m0\u001b[32m7\u001b[32m9\u001b[32m0\u001b[32mi\u001b[32mt\u001b[32m7\u001b[32ms\u001b[32m1\u001b[32mc\u001b[32mc\u001b[32m1\u001b[32m\r",
      "\u001b[32m\n",
      "\u001b[32m\r",
      "\u001b[32m\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[30mt\u001b[30my\u001b[30mp\u001b[30me\u001b[32m=\u001b[30mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m&\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m=\u001b[32mi\u001b[32mn\u001b[32mf\u001b[32mo\u001b[32mu\u001b[32mp\u001b[32md\u001b[32ma\u001b[32mt\u001b[32me\u001b[32m&\u001b[32mf\u001b[30mi\u001b[32mr\u001b[32ms\u001b[32mt\u001b[32mn\u001b[30ma\u001b[32mm\u001b[30me\u001b[30m=\u001b[32mJ\u001b[32mo\u001b[32mh\u001b[32mn\u001b[32m&\u001b[30ml\u001b[30ma\u001b[30ms\u001b[30mt\u001b[30mn\u001b[30ma\u001b[30mm\u001b[30me\u001b[30m=\u001b[32mD\u001b[32mo\u001b[32me\u001b[32m&\u001b[32mp\u001b[32mh\u001b[32mo\u001b[32mn\u001b[32me\u001b[32m=\u001b[32m7\u001b[32m9\u001b[32m0\u001b[32m3\u001b[32m5\u001b[32m2\u001b[32m2\u001b[32m2\u001b[32m2\u001b[32m1\u001b[32m2\u001b[32m&\u001b[32me\u001b[30mm\u001b[32ma\u001b[32mi\u001b[32ml\u001b[30m=\u001b[32mj\u001b[32m.\u001b[32md\u001b[30mo\u001b[32me\u001b[32m%\u001b[32m4\u001b[32m0\u001b[32mv\u001b[32mu\u001b[32ml\u001b[32mn\u001b[32mb\u001b[32ma\u001b[32mn\u001b[32mk\u001b[32m.\u001b[32md\u001b[32me\u001b[32m&\u001b[32mb\u001b[32mi\u001b[32mr\u001b[30mt\u001b[32mh\u001b[32md\u001b[32ma\u001b[32mt\u001b[32me\u001b[32m=\u001b[32m1\u001b[32m9\u001b[32m8\u001b[32m4\u001b[30m-\u001b[32m0\u001b[32m4\u001b[32m-\u001b[32m0\u001b[32m4\u001b[32m&\u001b[32ma\u001b[32mb\u001b[32mo\u001b[32mu\u001b[30mt\u001b[32m=\u001b[32mH\u001b[32mi\u001b[32m!\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  2.2545607\n",
      "\u001b[30mP\u001b[30mO\u001b[30mS\u001b[30mT\u001b[30m \u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[30ma\u001b[30mp\u001b[30mi\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[30m \u001b[30mH\u001b[30mT\u001b[30mT\u001b[30mP\u001b[30m/\u001b[30m1\u001b[30m.\u001b[30m1\u001b[32m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[32mk\u001b[30me\u001b[32me\u001b[30mp\u001b[32m-\u001b[30ma\u001b[30ml\u001b[32mi\u001b[32mv\u001b[30me\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30me\u001b[30mn\u001b[30mg\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30m1\u001b[32m3\u001b[32m0\u001b[32m\n",
      "\u001b[30mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mj\u001b[30ms\u001b[30mo\u001b[30mn\u001b[30m,\u001b[30m \u001b[30mt\u001b[30me\u001b[30mx\u001b[30mt\u001b[30m/\u001b[30mj\u001b[30ma\u001b[30mv\u001b[30ma\u001b[30ms\u001b[30mc\u001b[30mr\u001b[30mi\u001b[30mp\u001b[30mt\u001b[30m,\u001b[30m \u001b[30m*\u001b[30m/\u001b[30m*\u001b[30m;\u001b[30m \u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m1\u001b[32m\n",
      "\u001b[30mO\u001b[30mr\u001b[30mi\u001b[30mg\u001b[30mi\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\n",
      "\u001b[30mX\u001b[30m-\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[30me\u001b[30md\u001b[30m-\u001b[30mW\u001b[30mi\u001b[30mt\u001b[30mh\u001b[30m:\u001b[30m \u001b[30mX\u001b[30mM\u001b[30mL\u001b[30mH\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30mR\u001b[30me\u001b[30mq\u001b[30mu\u001b[30me\u001b[30ms\u001b[30mt\u001b[32m\n",
      "\u001b[30mU\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m-\u001b[30mA\u001b[30mg\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m:\u001b[30m \u001b[30mM\u001b[30mo\u001b[30mz\u001b[30mi\u001b[30ml\u001b[30ml\u001b[30ma\u001b[30m/\u001b[30m5\u001b[30m.\u001b[30m0\u001b[30m \u001b[30m(\u001b[32mX\u001b[30m1\u001b[30m1\u001b[30m;\u001b[30m \u001b[30mL\u001b[30mi\u001b[30mn\u001b[30mu\u001b[30mx\u001b[30m \u001b[30mx\u001b[30m8\u001b[30m6\u001b[30m_\u001b[30m6\u001b[30m4\u001b[30m)\u001b[30m \u001b[30mA\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30me\u001b[30mW\u001b[30me\u001b[30mb\u001b[30mK\u001b[30mi\u001b[30mt\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[30m \u001b[30m(\u001b[30mK\u001b[30mH\u001b[30mT\u001b[30mM\u001b[30mL\u001b[30m,\u001b[30m \u001b[30ml\u001b[30mi\u001b[30mk\u001b[30me\u001b[30m \u001b[30mG\u001b[30me\u001b[30mc\u001b[30mk\u001b[30mo\u001b[30m)\u001b[30m \u001b[30mC\u001b[30mh\u001b[30mr\u001b[30mo\u001b[30mm\u001b[30me\u001b[30m/\u001b[30m6\u001b[32m6\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m3\u001b[30m3\u001b[32m5\u001b[32m9\u001b[30m.\u001b[30m1\u001b[30m3\u001b[32m9\u001b[30m \u001b[30mS\u001b[30ma\u001b[30mf\u001b[30ma\u001b[30mr\u001b[30mi\u001b[30m/\u001b[30m5\u001b[30m3\u001b[30m7\u001b[30m.\u001b[30m3\u001b[30m6\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[30mT\u001b[30my\u001b[30mp\u001b[30me\u001b[30m:\u001b[30m \u001b[30ma\u001b[30mp\u001b[30mp\u001b[30ml\u001b[30mi\u001b[30mc\u001b[30ma\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m/\u001b[30mx\u001b[30m-\u001b[30mw\u001b[30mw\u001b[30mw\u001b[30m-\u001b[30mf\u001b[30mo\u001b[30mr\u001b[30mm\u001b[30m-\u001b[30mu\u001b[30mr\u001b[30ml\u001b[30me\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30me\u001b[30md\u001b[30m;\u001b[30m \u001b[30mc\u001b[30mh\u001b[30ma\u001b[30mr\u001b[30ms\u001b[30me\u001b[30mt\u001b[30m=\u001b[30mU\u001b[30mT\u001b[30mF\u001b[30m-\u001b[30m8\u001b[32m\n",
      "\u001b[32mD\u001b[32mN\u001b[32mT\u001b[32m:\u001b[30m \u001b[30m1\u001b[32m\n",
      "\u001b[30mR\u001b[30me\u001b[30mf\u001b[30me\u001b[30mr\u001b[30me\u001b[30mr\u001b[30m:\u001b[30m \u001b[30mh\u001b[30mt\u001b[30mt\u001b[30mp\u001b[30m:\u001b[30m/\u001b[30m/\u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[30m/\u001b[30mv\u001b[30mu\u001b[30ml\u001b[30mn\u001b[30mb\u001b[30ma\u001b[30mn\u001b[30mk\u001b[30m/\u001b[30mo\u001b[30mn\u001b[30ml\u001b[30mi\u001b[30mn\u001b[30me\u001b[30m/\u001b[32mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30mi\u001b[30mn\u001b[30mf\u001b[30mo\u001b[30m.\u001b[30mp\u001b[30mh\u001b[30mp\u001b[32m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mE\u001b[30mn\u001b[30mc\u001b[30mo\u001b[30md\u001b[30mi\u001b[30mn\u001b[30mg\u001b[30m:\u001b[30m \u001b[30mg\u001b[30mz\u001b[30mi\u001b[30mp\u001b[30m,\u001b[30m \u001b[30md\u001b[30me\u001b[30mf\u001b[30ml\u001b[30ma\u001b[30mt\u001b[30me\u001b[32m\n",
      "\u001b[32mA\u001b[30mc\u001b[30mc\u001b[30me\u001b[30mp\u001b[30mt\u001b[30m-\u001b[30mL\u001b[30ma\u001b[30mn\u001b[30mg\u001b[30mu\u001b[30ma\u001b[30mg\u001b[30me\u001b[30m:\u001b[30m \u001b[30me\u001b[30mn\u001b[30m-\u001b[30mU\u001b[30mS\u001b[30m,\u001b[30me\u001b[30mn\u001b[30m;\u001b[30mq\u001b[30m=\u001b[30m0\u001b[30m.\u001b[30m9\u001b[32m,\u001b[32mr\u001b[32mu\u001b[32m;\u001b[32mq\u001b[32m=\u001b[30m0\u001b[30m.\u001b[30m8\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[32mo\u001b[32mk\u001b[30mi\u001b[30me\u001b[30m:\u001b[30m \u001b[32mP\u001b[30mH\u001b[30mP\u001b[30mS\u001b[30mE\u001b[30mS\u001b[30mS\u001b[30mI\u001b[30mD\u001b[30m=\u001b[32m7\u001b[32mb\u001b[32mg\u001b[32mg\u001b[32m5\u001b[32mt\u001b[32ma\u001b[32ml\u001b[32md\u001b[32m4\u001b[32ma\u001b[32me\u001b[32ma\u001b[32m8\u001b[32m0\u001b[32m7\u001b[32m9\u001b[32m0\u001b[32mi\u001b[32mt\u001b[32m7\u001b[32ms\u001b[32m1\u001b[32mc\u001b[32mc\u001b[32m1\u001b[32m\n",
      "\u001b[32m\n",
      "\u001b[30mt\u001b[30my\u001b[30mp\u001b[30me\u001b[32m=\u001b[30mu\u001b[30ms\u001b[30me\u001b[30mr\u001b[30m&\u001b[30ma\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m=\u001b[32mi\u001b[32mn\u001b[32mf\u001b[32mo\u001b[32mu\u001b[32mp\u001b[32md\u001b[32ma\u001b[32mt\u001b[32me\u001b[32m&\u001b[32mf\u001b[30mi\u001b[32mr\u001b[32ms\u001b[32mt\u001b[32mn\u001b[30ma\u001b[32mm\u001b[30me\u001b[30m=\u001b[32mJ\u001b[32mo\u001b[32mh\u001b[32mn\u001b[32m&\u001b[30ml\u001b[30ma\u001b[30ms\u001b[30mt\u001b[30mn\u001b[30ma\u001b[30mm\u001b[30me\u001b[30m=\u001b[32mD\u001b[32mo\u001b[32me\u001b[32m&\u001b[32mp\u001b[32mh\u001b[32mo\u001b[32mn\u001b[32me\u001b[32m=\u001b[32m7\u001b[32m9\u001b[32m0\u001b[32m3\u001b[32m5\u001b[32m2\u001b[32m2\u001b[32m2\u001b[32m2\u001b[32m1\u001b[32m2\u001b[32m&\u001b[32me\u001b[30mm\u001b[32ma\u001b[32mi\u001b[32ml\u001b[30m=\u001b[32mj\u001b[32m.\u001b[32md\u001b[30mo\u001b[32me\u001b[32m%\u001b[32m4\u001b[32m0\u001b[32mv\u001b[32mu\u001b[32ml\u001b[32mn\u001b[32mb\u001b[32ma\u001b[32mn\u001b[32mk\u001b[32m.\u001b[32md\u001b[32me\u001b[32m&\u001b[32mb\u001b[32mi\u001b[32mr\u001b[30mt\u001b[32mh\u001b[32md\u001b[32ma\u001b[32mt\u001b[32me\u001b[32m=\u001b[32m1\u001b[32m9\u001b[32m8\u001b[32m4\u001b[30m-\u001b[32m0\u001b[32m4\u001b[32m-\u001b[32m0\u001b[32m4\u001b[32m&\u001b[32ma\u001b[32mb\u001b[32mo\u001b[32mu\u001b[30mt\u001b[32m=\u001b[32mH\u001b[32mi\u001b[32m!\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  5.8993344\n",
      "\u001b[32mP\u001b[32mU\u001b[32mT\u001b[30m \u001b[32m/\u001b[32mc\u001b[32mv\u001b[32mp\u001b[32mh\u001b[32mq\u001b[32mf\u001b[32my\u001b[32mm\u001b[32ml\u001b[32mt\u001b[32m.\u001b[32mt\u001b[32mx\u001b[32mt\u001b[32m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[30mP\u001b[32m/\u001b[32m1\u001b[32m.\u001b[32m1\u001b[32m\n",
      "\u001b[30mH\u001b[30mo\u001b[32ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[32mt\u001b[30me\u001b[30mn\u001b[30mt\u001b[30m-\u001b[32mL\u001b[32me\u001b[32mn\u001b[30mg\u001b[32mt\u001b[32mh\u001b[30m:\u001b[30m \u001b[30m1\u001b[32m0\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mc\u001b[30ml\u001b[30mo\u001b[30ms\u001b[30me\u001b[32m\n",
      "\u001b[32m\n",
      "\u001b[32mm\u001b[32mb\u001b[32m4\u001b[32m7\u001b[32m0\u001b[32mo\u001b[32mr\u001b[32m8\u001b[32ms\u001b[32ml\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  6.080594\n",
      "\u001b[32mT\u001b[32mR\u001b[32mA\u001b[32mC\u001b[32mE\u001b[32m \u001b[32m/\u001b[32m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[32m/\u001b[30m1\u001b[32m.\u001b[32m1\u001b[32m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[32mo\u001b[32mk\u001b[32mi\u001b[30me\u001b[30m:\u001b[30m \u001b[32mw\u001b[32md\u001b[32m1\u001b[32m2\u001b[32mk\u001b[32mg\u001b[32mj\u001b[32m0\u001b[32mo\u001b[32m3\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[32mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mc\u001b[30ml\u001b[30mo\u001b[30ms\u001b[30me\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  4.7534595\n",
      "\u001b[32mG\u001b[32mE\u001b[32mT\u001b[32m \u001b[30m/\u001b[32mc\u001b[32ml\u001b[32mi\u001b[32me\u001b[32mn\u001b[32mt\u001b[32ma\u001b[32mc\u001b[32mc\u001b[32me\u001b[32ms\u001b[30ms\u001b[32mp\u001b[32mo\u001b[32ml\u001b[32mi\u001b[32mc\u001b[32my\u001b[32m.\u001b[32mx\u001b[32mm\u001b[32ml\u001b[30m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[32m/\u001b[32m1\u001b[32m.\u001b[32m1\u001b[32m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[32m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\n",
      "\u001b[30mC\u001b[30mo\u001b[30mn\u001b[30mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mc\u001b[30ml\u001b[30mo\u001b[30ms\u001b[30me\u001b[30m\n",
      "\n",
      "Prediction:  1\n",
      "Loss  5.9863434\n",
      "\u001b[32mD\u001b[30mE\u001b[32mB\u001b[32mU\u001b[32mG\u001b[30m \u001b[32m/\u001b[32mD\u001b[32me\u001b[32mf\u001b[32ma\u001b[32mu\u001b[32ml\u001b[32mt\u001b[32m.\u001b[32ma\u001b[32ms\u001b[32mp\u001b[32mx\u001b[32m \u001b[32mH\u001b[32mT\u001b[32mT\u001b[32mP\u001b[32m/\u001b[32m1\u001b[32m.\u001b[32m1\u001b[32m\n",
      "\u001b[30mH\u001b[30mo\u001b[30ms\u001b[30mt\u001b[30m:\u001b[30m \u001b[30m1\u001b[30m0\u001b[30m.\u001b[30m0\u001b[30m.\u001b[30m2\u001b[30m1\u001b[30m2\u001b[30m.\u001b[30m2\u001b[30m5\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[32mm\u001b[32mm\u001b[32ma\u001b[30mn\u001b[32md\u001b[32m:\u001b[30m \u001b[32ms\u001b[32mt\u001b[32ma\u001b[32mr\u001b[32mt\u001b[32m-\u001b[32md\u001b[32me\u001b[32mb\u001b[32mu\u001b[32mg\u001b[32m\n",
      "\u001b[32mC\u001b[30mo\u001b[30mn\u001b[32mn\u001b[30me\u001b[30mc\u001b[30mt\u001b[30mi\u001b[30mo\u001b[30mn\u001b[30m:\u001b[30m \u001b[30mc\u001b[30ml\u001b[30mo\u001b[30ms\u001b[32me\u001b[30m"
     ]
    }
   ],
   "source": [
    "pred_data = Data(path_anomaly_data, predict=True)\n",
    "pred_gen = pred_data.predict_generator()\n",
    "anomaly_preds, anomaly_loss = p.predict(pred_gen)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Number of TP:  1097\n",
      "Number of samples:  1097\n",
      "TP rate: 1.0000\n"
     ]
    }
   ],
   "source": [
    "print('Number of TP: ', np.sum(anomaly_preds))\n",
    "print('Number of samples: ', len(anomaly_preds))\n",
    "print('TP rate: {:.4f}'.format(np.sum(anomaly_preds) / len(anomaly_preds)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Testing "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "To evaluate the results, let's compute metrics of quality: precision, recall, ROC-AUC."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_true = np.concatenate(([0] * len(valid_preds), [1] * len(anomaly_preds)), axis=0)\n",
    "preds = np.concatenate((valid_preds, anomaly_preds), axis=0)\n",
    "loss_pred = np.concatenate((valid_loss, anomaly_loss), axis=0)\n",
    "assert len(y_true)==len(loss_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Precision: 0.9928\n",
      "Recall: 1.0000\n"
     ]
    }
   ],
   "source": [
    "precision = precision_score(y_true, preds)\n",
    "recall = recall_score(y_true, preds)\n",
    "print('Precision: {:.4f}'.format(precision))\n",
    "print('Recall: {:.4f}'.format(recall))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fdc25daf1d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fpr, tpr, _ = roc_curve(y_true, loss_pred)\n",
    "roc_auc = auc(fpr, tpr)\n",
    "\n",
    "plt.figure()\n",
    "plt.plot(fpr, tpr, label='ROC curve (area = %0.4f)' % roc_auc)\n",
    "plt.plot([0, 1], [0, 1], color='navy', linestyle='--')\n",
    "plt.xlim([-0.05, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('ROC curve')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}