deep_learning/course1/lesson3/transfer-learning-in-nlp-using-fastai.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [],
   "source": [
    "from fastai.text import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [],
   "source": [
    "path = \"../input/review_polarity/txt_sentoken/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "_uuid": "0fe6623a09b11a118a62bf639827511e1c233e92"
   },
   "outputs": [],
   "source": [
    "bs = 48"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "_uuid": "dc4772207f9b72a3d23d17db16f46cd13b416e12"
   },
   "outputs": [],
   "source": [
    "np.random.seed(42)\n",
    "data = (TextList.from_folder(path)\n",
    "                        .split_by_rand_pct(0.1)\n",
    "                        .label_from_folder()\n",
    "                        .databunch(bs=bs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "_uuid": "f9d1c36f500a094caff59a70a9c649ffac2fc1a7"
   },
   "outputs": [],
   "source": [
    "data.save(\"/kaggle/working/data.pkl\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "_uuid": "1353f963c6d30076000c30cd9ce8ebc4e78e2983"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>text</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>xxbos \" say , any of you know how to madison ? \" \\n  --- brad \" asshole \" majors ( barry bostwick ) shows just how unhip he is after watching the kooky xxunk wrap up their ode to the time warp in the cult classic the rocky horror picture show . \\n  the story behind the rocky horror picture show is one of legend . \\n</td>\n",
       "      <td>pos</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>xxbos i remember seeing the trailer of the black cauldron on the pinocchio video . \\n  it really scared me , and i did n't want to go see it . \\n  however , after i heard about how wonderful it was , i wanted to see it . \\n  i did n't get a chance to do so , until now . \\n  the film</td>\n",
       "      <td>pos</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>xxbos natural born killers is really a very simple story that , in essence , has already been told in bonnie &amp; clyde with some major variations in emphasis , mood and degree . \\n  both films xxunk \" outlaws , \" in one case bank robbers and killers and in the other mass murderers . \\n  both relate the story of a couple liberated by love .</td>\n",
       "      <td>pos</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>xxbos * * * the following review contains spoilers * * * \\n  \" it 's just . . . \\n  the way . . . \\n  we are . \" \\n  --- a rapist ( matt xxunk ) responds to supergirl 's ( helen slater ) xxunk as to why she 's being attacked , an example of the mind - numbingly bad dialogue in</td>\n",
       "      <td>neg</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>xxbos seen may 19 , 1999 at 5 p . m . at the crossgates cinema 18 ( guilderland , n . y . ) , theater # 1 , with my brother john for $ 8 . 50 . \\n  [ theater rating : * * * * : excellent seats , sound and picture ] \\n  ( critic 's note : this review is the longest</td>\n",
       "      <td>pos</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data.show_batch()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "_uuid": "67d15480a1a7c08ff1e666bf8fcf22b02e913909"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['xxunk',\n",
       " 'xxpad',\n",
       " 'xxbos',\n",
       " 'xxeos',\n",
       " 'xxfld',\n",
       " 'xxmaj',\n",
       " 'xxup',\n",
       " 'xxrep',\n",
       " 'xxwrep',\n",
       " ',']"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.vocab.itos[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "_uuid": "79d5361fbc2b6f3f2df872407048bd9f55fd6caf"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text xxbos it is hard to imagine that a movie which includes abortion and incest as prominent plot devices could be so sensitive , so delicate , so insightful and yes , even so enjoyable . \n",
       "  somehow , the cider house rules manages to keep an even xxunk in the face of the harsh subject matter and allows a substantial amount of heart to shine through . \n",
       "  set in the 1940 's , the film tells the story of a unique young man named homer wells ( tobey maguire , ride with the devil ) . \n",
       "  an orphan , twice rejected as an xxunk by possible adoptive parents , he grows up in the care and under the tutelage of dr . xxunk larch ( michael caine , little voice ) , father figure to the orphans of st . cloud , obstetrician or xxunk to the unhappily pregnant women who come there seeking his help . \n",
       "  figuring that as long as homer was going to stay at the orphanage , he might as well be of use , dr . larch begins teaching him all he knows about xxunk . \n",
       "  soon homer is delivering babies like a professional . \n",
       "  he refuses to perform xxunk , instead wondering why the couples just did n't behave more xxunk . \n",
       "  larch , too old and experienced to have such youthful ideals , xxunk that homer \" continues to have such high expectations of people . \" \n",
       "  such is the result of an xxunk upbringing . \n",
       "  though skilled in certain medical procedures , homer remains woefully naive and ignorant about most everything else . \n",
       "  he finally figures that in order to find his place in the world , he must go out into it . \n",
       "  befriending a young couple who had come to see dr . larch , he xxunk a ride with them and xxunk that into landing a job as an apple xxunk with a group of black migrant workers at the xxunk where they lived . \n",
       "  thus , homer begins his worldly xxunk , with new friend candy xxunk ( charlize theron , the astronaut 's wife ) a willing tutor . \n",
       "  tobey maguire is well cast yet again as an inherently good man who is seeking answers to life 's questions . . . \n",
       "  many of which he was unaware needed to be asked . \n",
       "  he demonstrates a quiet strength and thoughtful curiosity that is both warm and appealing . \n",
       "  michael caine is , in a word , brilliant as the caring caretaker of these orphans , many of whom he helped to bring into the world . \n",
       "  calling them \" xxunk of maine , kings of new england \" in his ritualistic goodnight , mr . caine manages to communicate the love and commitment larch has towards these children without becoming maudlin or saccharine . \n",
       "  delroy lindo ( a life less ordinary ) has many powerful moments as mr . rose , the strong and charismatic leader of the apple xxunk whom homer joins . \n",
       "  director lasse hallstrom ( what 's eating gilbert grape ? ) and screenwriter / novelist john irving ( the world according to xxunk ) , who adapted his own work for this movie deserve much of the credit for the balanced approach to the sensitive subject matter . \n",
       "  regardless of one 's personal stand on the xxunk matter of abortion , it is hard to imagine anyone becoming offended by the xxunk treatment used by these filmmakers . \n",
       "  the title is a reference to a list of rules posted in the quarters of the illiterate apple xxunk . \n",
       "  figuring the rules did n't apply to them since they did n't have a hand in writing them , they decide to ignore the list and make up their own rules as they go along . \n",
       "  dr . larch , in his own way , acts similarly . \n",
       "  there is a danger in that line of reasoning . \n",
       "  god has also given us a set of rules . \n",
       "  unlike the list posted in the cider house , god 's rules , if followed , promise us benefits beyond our imagination . \n",
       "  and yet , many continue to ignore his \" rules \" preferring to make up their own as they go through life . \n",
       "  \" trust in the lord with all xxunk heart ; and lean not unto xxunk own understanding . \" \n",
       "  xxunk 3 : 5 [ kjv ] \n",
       "  just as a builder follows a designer 's blueprint or a conductor follows a composer 's score , why should n't we follow the rules of life given to us by life 's architect for our own blessing ? \n",
       " "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.train_ds[0][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "_uuid": "d97f4cd307104a41f2ac4bdca3fb4dfb0e3e5171"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([   2,   21,   17,  266,   16, 1126,   22,   13,   42,   64])"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data.train_ds[0][0].data[:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "_uuid": "4a57c2720cc7faac08d9b9c886343649bb16e1d0"
   },
   "outputs": [],
   "source": [
    "np.random.seed(42)\n",
    "data_lm = (TextList.from_folder(path)\n",
    "                        .split_by_rand_pct(0.1)\n",
    "                        .label_for_lm()\n",
    "                        .databunch(bs=bs))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "_uuid": "4166fc195fd1c99baef7b5a7b15329a46d6baacb"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>idx</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>the 1940 's , the film tells the story of a unique young man named homer wells ( tobey maguire , ride with the devil ) . \\n  an orphan , twice rejected as an xxunk by possible adoptive parents , he grows up in the care and under the tutelage of dr . xxunk larch ( michael caine , little voice ) , father figure to the orphans</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>single building exploded . \\n  this , hartley 's seventh feature , is contemplative about \" the artist \" ; fortunately , we never get a sampling of simon 's work , and it 's a testament to the actors and the screenplay that by the end we can surmise what sort of thoughts would pour from these writers ' xxunk . \\n  xxbos kevin smith is like</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>wife 's tuna xxunk . \\n  \" so , \" you ask , \" where is he now ? \" \\n  mr . kaplan actually spends much of his time these days playing professional high - stakes poker . \\n  that 's right . \\n  i 'm not sure if it 's his main source of income , but from what i 've heard , the</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>was neat . \\n  having said that , i did n't think the first hour needed to be as long as it was , and the film did have me tinkering on the bridge of believability when the \" fake \" jackal met with the \" real \" jackal 's girlfriend without much xxunk on her part , but all in all , not too many complaints in a</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>everyone involved . \\n  let 's just hope they make a thousand sequels . \\n  [ sorry , i had to get the lame joke in . . . \\n  forgive me . ] \\n  xxbos originally entitled dancing about architecture ( i . e . \\n  \" talking about love is like dancing about architecture \" ) , but changed because it sounded too</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data_lm.show_batch()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "_uuid": "f4fff0930723ca805facb8baa08a1b92e283e7e3"
   },
   "outputs": [],
   "source": [
    "learn = language_model_learner(data_lm, arch=AWD_LSTM, drop_mult=0.3, model_dir=\"/tmp/model/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "_uuid": "9c18fffe4192b6719926e5ffd0edc9358611880c"
   },
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    }
   ],
   "source": [
    "learn.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "_uuid": "cb3e4ac4e89bb5e2280128973099fdf065d3d1b6"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.recorder.plot(skip_end=15)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "_uuid": "fe847f3a9ff946d284016104cad11820c115c721"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 00:49 <p><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>4.627384</td>\n",
       "      <td>4.365340</td>\n",
       "      <td>0.278990</td>\n",
       "      <td>00:49</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(1, 3e-02, moms=(0.8, 0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "_uuid": "ba7135a868ca8b6904e2a38bf9af8af022bf2468"
   },
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.unfreeze()\n",
    "learn.lr_find()\n",
    "learn.recorder.plot(skip_end=15)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "_uuid": "7a98b109ec617a97ba55d6163810cecf8bd7d128"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='6' class='' max='10', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      60.00% [6/10 05:52<03:54]\n",
       "    </div>\n",
       "    \n",
       "<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>4.148184</td>\n",
       "      <td>4.326283</td>\n",
       "      <td>0.281718</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>4.059402</td>\n",
       "      <td>4.266865</td>\n",
       "      <td>0.286838</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>4.003819</td>\n",
       "      <td>4.222210</td>\n",
       "      <td>0.291916</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>3.894475</td>\n",
       "      <td>4.190765</td>\n",
       "      <td>0.295098</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>3.844278</td>\n",
       "      <td>4.166241</td>\n",
       "      <td>0.297328</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>3.756064</td>\n",
       "      <td>4.157764</td>\n",
       "      <td>0.298305</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='16' class='' max='435', style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      3.68% [16/435 00:02<00:54 3.7406]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(10, 3e-04, moms=(0.8, 0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "_uuid": "fc2e385e7172f60dbbfbdba2bcf17b947f74de2f"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Avengers is a  series of pictures that have been showing on television for over a century . \n",
      "  like the first one , \" scream 2 \" 's title is \" cape city\n",
      "Avengers is a  film that takes place in the near future when the avengers is being released . \n",
      "  at first , the camera is so utterly unbelievable that it 's just too\n"
     ]
    }
   ],
   "source": [
    "TEXT = \"Avengers is a \"\n",
    "N_WORDS = 30\n",
    "N_SENTENCES = 2\n",
    "\n",
    "print(\"\\n\".join(learn.predict(TEXT, N_WORDS, temperature=0.75) for _ in range(N_SENTENCES)))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "_uuid": "12809af4e0171480114d17afec38a3daac8be92d"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "My sister watched that movie because it 's a great film . \n",
      "  this film opens\n",
      "My sister watched that movie because of her desire to have a comedy and a story\n"
     ]
    }
   ],
   "source": [
    "TEXT = \"My sister watched that movie because\"\n",
    "N_WORDS = 10\n",
    "N_SENTENCES = 2\n",
    "\n",
    "print(\"\\n\".join(learn.predict(TEXT, N_WORDS, temperature=0.75) for _ in range(N_SENTENCES)))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "_uuid": "d4dae393095949386ebcd160b58a09a0ee58bb5f"
   },
   "outputs": [],
   "source": [
    "learn.save('/kaggle/working/fine_tuned')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "_uuid": "2eebc8ac4113b9c26ddb292b82d3026638b12c7e"
   },
   "outputs": [],
   "source": [
    "learn.save_encoder('fine_tuned_enc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "_uuid": "feb00416e4d6df9fa0eae2df2e019a312bbcb151"
   },
   "outputs": [],
   "source": [
    "learn = text_classifier_learner(data, arch=AWD_LSTM, drop_mult=0.3, model_dir=\"/tmp/model/\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "_uuid": "b8540211def31c158a90250490bad50f10906cca"
   },
   "outputs": [],
   "source": [
    "learn.load_encoder('fine_tuned_enc')\n",
    "learn.freeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "_uuid": "6406683d85a2c834f8d9abf36130b9c21efb5ec7"
   },
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    }
   ],
   "source": [
    "learn.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "_uuid": "7fae5d963bef55e2dc8efd3a0e6275096a55612e"
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "_uuid": "aa442e08590278c8007cc3ccf5818749f8b219f2"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 00:52 <p><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>0.542260</td>\n",
       "      <td>0.555401</td>\n",
       "      <td>0.780000</td>\n",
       "      <td>00:25</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.456235</td>\n",
       "      <td>0.376684</td>\n",
       "      <td>0.830000</td>\n",
       "      <td>00:26</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(2, 2e-2, moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "_uuid": "75b9a2551a2ee029eebd7c4a65355241a2acceb0"
   },
   "outputs": [],
   "source": [
    "learn.save('/kaggle/working/first')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "_uuid": "9b21094d226add656dfbcbf1c2030cc3b97af910"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 00:31 <p><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>0.409211</td>\n",
       "      <td>0.365711</td>\n",
       "      <td>0.845000</td>\n",
       "      <td>00:31</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.freeze_to(-2)\n",
    "learn.fit_one_cycle(1, slice(1e-2/(2.6**4),1e-2), moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "_uuid": "9b63a8e0e986cb60e10058b80d2e9ea5d29b4fd8"
   },
   "outputs": [],
   "source": [
    "learn.save('/kaggle/working/second')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "_uuid": "7f9446313890da72e88be3148dd2b03c97ea2f32"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 00:46 <p><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>0.305756</td>\n",
       "      <td>0.284292</td>\n",
       "      <td>0.895000</td>\n",
       "      <td>00:46</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.freeze_to(-3)\n",
    "learn.fit_one_cycle(1, slice(5e-3/(2.6**4),5e-3), moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "_uuid": "7e3aee714a684d210d8f929b4f66e3e725aa8a53"
   },
   "outputs": [],
   "source": [
    "learn.save('/kaggle/working/third')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "_uuid": "d80f11b5a42d2531e2eb4169feecd41acff72f04"
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 01:55 <p><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>0.250906</td>\n",
       "      <td>0.247524</td>\n",
       "      <td>0.910000</td>\n",
       "      <td>00:58</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.224148</td>\n",
       "      <td>0.236824</td>\n",
       "      <td>0.925000</td>\n",
       "      <td>00:56</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(2, slice(1e-3/(2.6**4),1e-3), moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "_uuid": "4add4282a3e4dceb67a731540d83a5afb46d9cbe"
   },
   "outputs": [],
   "source": [
    "learn.save('/kaggle/working/fourth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "_uuid": "b01e47bef415de68d5962f7e70c6ff4b39b6e912"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category pos, tensor(1), tensor([0.0081, 0.9919]))"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"I really loved that movie, it was awesome!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "_uuid": "116528dbd274bc87a64aff50786e08cfdad77c46"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.9002, 0.0998]))"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"The movie is disappointing\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "_uuid": "a8e59a923ed9d348e4239fe470afbdb5f6a9cc93"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.9976, 0.0024]))"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"I would not recommend this movie to anyone\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "_uuid": "e24c17105764010726f955ba313e3d50f94aa608"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.9243, 0.0757]))"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"The movie was okay\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "_uuid": "3cceed0e5be91901ea230f9cf8ab2f86f14e0d29"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.9694, 0.0306]))"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"I would feel sorry for anyone watching this movie\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "_uuid": "8e7bf28384f848a02c9adba6237d9193838640e1"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.9609, 0.0391]))"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"The direction was horrible\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "_uuid": "04f5d14fcd9e42ec17de2a50d54996c51cd77899"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.9915, 0.0085]))"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"One would not be disappointed watching the movie.\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "_uuid": "84a6bf6cda901ed5c1ce175001609976b431713e"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category neg, tensor(0), tensor([0.8470, 0.1530]))"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"One would not be disappointed watching the movie. I especially loved the part when the protagonist goes for a retreat and finds all his answers\")"
   ]
  }
 ],
 "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": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}