nn-vietnamese-bwd_jcat.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Vietnamese ULMFiT from scratch (backwards)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2\n",
    "%matplotlib inline\n",
    "\n",
    "from fastai import *\n",
    "from fastai.text import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs=128"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_path = Config.data_path()\n",
    "lang = 'vi'\n",
    "name = f'{lang}wiki'\n",
    "path = data_path/name\n",
    "dest = path/'docs'\n",
    "lm_fns = [f'{lang}_wt_bwd', f'{lang}_wt_vocab_bwd']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vietnamese wikipedia model\n",
    "use `backwards = True` to reverse word order in reviews."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = (TextList.from_folder(dest)\n",
    "            .split_by_rand_pct(0.1, seed=42)\n",
    "            .label_for_lm()\n",
    "            .databunch(bs=bs, num_workers=1, backwards=True))\n",
    "\n",
    "data.save(f'{lang}_databunch_bwd')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/jhoward/anaconda3/lib/python3.7/site-packages/torch/serialization.py:493: SourceChangeWarning: source code of class 'torch.nn.modules.loss.CrossEntropyLoss' has changed. you can retrieve the original source code by accessing the object's source attribute or set `torch.nn.Module.dump_patches = True` and use the patch tool to revert the changes.\n",
      "  warnings.warn(msg, SourceChangeWarning)\n"
     ]
    }
   ],
   "source": [
    "data = load_data(dest, f'{lang}_databunch_bwd', bs=bs, backwards=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "learn = language_model_learner(data, AWD_LSTM, drop_mult=0.5, pretrained=False).to_fp16()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr = 3e-3\n",
    "lr *= bs/48  # Scale learning rate by batch size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>3.445849</td>\n",
       "      <td>3.424579</td>\n",
       "      <td>0.401327</td>\n",
       "      <td>32:56</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>3.420865</td>\n",
       "      <td>3.383994</td>\n",
       "      <td>0.402841</td>\n",
       "      <td>33:31</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>3.374694</td>\n",
       "      <td>3.330634</td>\n",
       "      <td>0.407800</td>\n",
       "      <td>33:26</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>3.273197</td>\n",
       "      <td>3.257108</td>\n",
       "      <td>0.416047</td>\n",
       "      <td>32:54</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>3.223044</td>\n",
       "      <td>3.200649</td>\n",
       "      <td>0.422695</td>\n",
       "      <td>32:56</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>3.134357</td>\n",
       "      <td>3.132859</td>\n",
       "      <td>0.430725</td>\n",
       "      <td>31:35</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>3.135637</td>\n",
       "      <td>3.057030</td>\n",
       "      <td>0.439737</td>\n",
       "      <td>31:41</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>3.080461</td>\n",
       "      <td>2.992323</td>\n",
       "      <td>0.447939</td>\n",
       "      <td>31:45</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>3.075036</td>\n",
       "      <td>2.943683</td>\n",
       "      <td>0.454494</td>\n",
       "      <td>31:39</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>2.947997</td>\n",
       "      <td>2.929258</td>\n",
       "      <td>0.456500</td>\n",
       "      <td>31:46</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.unfreeze()\n",
    "learn.fit_one_cycle(10, lr, moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "mdl_path = path/'models'\n",
    "mdl_path.mkdir(exist_ok=True)\n",
    "learn.to_fp32().save(mdl_path/lm_fns[0], with_opt=False)\n",
    "learn.data.vocab.save(mdl_path/(lm_fns[1] + '.pkl'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Vietnamese sentiment analysis"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "heading_collapsed": true
   },
   "source": [
    "### Language model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "hidden": true
   },
   "outputs": [],
   "source": [
    "train_df = pd.read_csv(path/'train.csv')\n",
    "train_df.loc[pd.isna(train_df.comment),'comment']='NA'\n",
    "\n",
    "test_df = pd.read_csv(path/'test.csv')\n",
    "test_df.loc[pd.isna(test_df.comment),'comment']='NA'\n",
    "test_df['label'] = 0\n",
    "\n",
    "df = pd.concat([train_df,test_df])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "hidden": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data_lm = (TextList.from_df(df, path, cols='comment')\n",
    "    .split_by_rand_pct(0.1, seed=42)\n",
    "    .label_for_lm()           \n",
    "    .databunch(bs=bs, num_workers=1, backwards=True))\n",
    "\n",
    "learn_lm = language_model_learner(data_lm, AWD_LSTM, config={**awd_lstm_lm_config, 'n_hid': 1152},\n",
    "                                  pretrained_fnames=lm_fns, drop_mult=1.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "hidden": true
   },
   "outputs": [],
   "source": [
    "lr = 1e-3\n",
    "lr *= bs/48"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "hidden": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>4.797052</td>\n",
       "      <td>4.025901</td>\n",
       "      <td>0.323326</td>\n",
       "      <td>00:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>4.275975</td>\n",
       "      <td>3.914450</td>\n",
       "      <td>0.333719</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_lm.fit_one_cycle(2, lr*10, moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "hidden": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>3.996770</td>\n",
       "      <td>3.809489</td>\n",
       "      <td>0.346052</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>3.856959</td>\n",
       "      <td>3.664919</td>\n",
       "      <td>0.363239</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>3.726143</td>\n",
       "      <td>3.584303</td>\n",
       "      <td>0.369685</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>3.608569</td>\n",
       "      <td>3.531390</td>\n",
       "      <td>0.375307</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>3.514265</td>\n",
       "      <td>3.500826</td>\n",
       "      <td>0.379701</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>3.446292</td>\n",
       "      <td>3.486931</td>\n",
       "      <td>0.380859</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>3.392542</td>\n",
       "      <td>3.479732</td>\n",
       "      <td>0.382520</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>3.357502</td>\n",
       "      <td>3.478930</td>\n",
       "      <td>0.382520</td>\n",
       "      <td>00:09</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_lm.unfreeze()\n",
    "learn_lm.fit_one_cycle(8, lr, moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "hidden": true
   },
   "outputs": [],
   "source": [
    "learn_lm.save(f'{lang}fine_tuned_bwd')\n",
    "learn_lm.save_encoder(f'{lang}fine_tuned_enc_bwd')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classifier"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data_clas = (TextList.from_df(train_df, path, vocab=data_lm.vocab, cols='comment')\n",
    "    .split_by_rand_pct(0.1, seed=42)\n",
    "    .label_from_df(cols='label')\n",
    "    .databunch(bs=bs, num_workers=1, backwards=True))\n",
    "\n",
    "data_clas.save(f'{lang}_textlist_class_bwd')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_clas = load_data(path, f'{lang}_textlist_class_bwd', bs=bs, num_workers=1, backwards=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import f1_score\n",
    "\n",
    "@np_func\n",
    "def f1(inp,targ): return f1_score(targ, np.argmax(inp, axis=-1))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn_c = text_classifier_learner(data_clas, AWD_LSTM, drop_mult=0.5, metrics=[accuracy,f1]).to_fp16()\n",
    "learn_c.load_encoder(f'{lang}fine_tuned_enc_bwd')\n",
    "learn_c.freeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "lr=2e-2\n",
    "lr *= bs/48"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>f1</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.369300</td>\n",
       "      <td>0.363769</td>\n",
       "      <td>0.834577</td>\n",
       "      <td>0.826098</td>\n",
       "      <td>00:03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.328192</td>\n",
       "      <td>0.278986</td>\n",
       "      <td>0.874378</td>\n",
       "      <td>0.851747</td>\n",
       "      <td>00:02</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_c.fit_one_cycle(2, lr, moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>f1</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.337875</td>\n",
       "      <td>0.306132</td>\n",
       "      <td>0.876866</td>\n",
       "      <td>0.860107</td>\n",
       "      <td>00:03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.276982</td>\n",
       "      <td>0.237260</td>\n",
       "      <td>0.906095</td>\n",
       "      <td>0.886427</td>\n",
       "      <td>00:03</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_c.freeze_to(-2)\n",
    "learn_c.fit_one_cycle(2, slice(lr/(2.6**4),lr), moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>f1</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.292297</td>\n",
       "      <td>0.252393</td>\n",
       "      <td>0.896144</td>\n",
       "      <td>0.877916</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.255284</td>\n",
       "      <td>0.213655</td>\n",
       "      <td>0.912313</td>\n",
       "      <td>0.892551</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_c.freeze_to(-3)\n",
    "learn_c.fit_one_cycle(2, slice(lr/2/(2.6**4),lr/2), moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>f1</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.167376</td>\n",
       "      <td>0.266633</td>\n",
       "      <td>0.904851</td>\n",
       "      <td>0.885386</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_c.unfreeze()\n",
    "learn_c.fit_one_cycle(1, slice(lr/10/(2.6**4),lr/10), moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn_c.save(f'{lang}clas_bwd')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {},
   "number_sections": false,
   "sideBar": true,
   "skip_h1_title": true,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}