NLP_transfer_learning_text_classification.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pymysql"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# connect database"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "db = pymysql.connect(\"...\",\"...\",\"...\",\"...\",charset='utf8')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cursor = db.cursor()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "sql = \"SELECT * FROM RAW_DATA_train\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "try:\n",
    "   cursor.execute(sql)\n",
    "   results = cursor.fetchall()\n",
    "except:\n",
    "   print (\"Error: unable to fetch data\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'tuple'>\n"
     ]
    }
   ],
   "source": [
    "print(type(results))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(Lateral boosters can pinch in the sides if youve found a gym in the last few years too.)Weve found that the front seats are widely adjustable and we think that most body types wont have a hard time finding a natural position in the drivers seat—provided you can swipe the keys first.Outward visibility in the Q60 is generally good, and bulges over the front tires should help drivers place the front wheels despite the coupes long hood.\n"
     ]
    }
   ],
   "source": [
    "print(results[0][2])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "martir=[]\n",
    "for row in results:\n",
    "    #print(type(row[1]))\n",
    "    label=row[1]\n",
    "    text=row[2]\n",
    "    pair=[]\n",
    "    pair.append(label)\n",
    "    pair.append(text)\n",
    "    martir.append(pair)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "mat_array=np.array(martir)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<U2630\n"
     ]
    }
   ],
   "source": [
    "print(mat_array.dtype)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(101286, 2)\n"
     ]
    }
   ],
   "source": [
    "print(mat_array.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "exterior\n"
     ]
    }
   ],
   "source": [
    "print(mat_array[0][0])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "DIS_SQL=\"SELECT DISTINCT label FROM RAW_DATA_train\"\n",
    "try:\n",
    "   \n",
    "   cursor.execute(DIS_SQL)\n",
    "   \n",
    "   label_results = cursor.fetchall()\n",
    "except:\n",
    "   print (\"Error: unable to fetch data\")\n",
    " \n",
    "\n",
    "db.close()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(('exterior',), ('driving',), ('safety',), ('performance',), ('space',), ('interior',), ('fuel economy',))\n"
     ]
    }
   ],
   "source": [
    "print(label_results)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "mapping_dict={'exterior':'1','driving':'2','safety':'3','performance':'4','space':'5','interior':'6','fuel economy':'7'}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for i in mat_array:\n",
    "    i[0]=mapping_dict[i[0]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(101286, 2)\n"
     ]
    }
   ],
   "source": [
    "print(mat_array.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[['1'\n",
      "  '(Lateral boosters can pinch in the sides if youve found a gym in the last few years too.)Weve found that the front seats are widely adjustable and we think that most body types wont have a hard time finding a natural position in the drivers seat—provided you can swipe the keys first.Outward visibility in the Q60 is generally good, and bulges over the front tires should help drivers place the front wheels despite the coupes long hood.']\n",
      " ['1'\n",
      "  '(Read more about how we rate cars.)By the numbers, the SL-Class has 101.7 inches between the wheels, but most of that space is dedicated to the long hood or power-operated convertible top holder (aka trunk).']\n",
      " ['1'\n",
      "  '(Read more about how we rate cars.)The Cherokees interior materials are a notch above the competition and Jeep has updated the look this year with contrasting lighter colors on lower panels.']\n",
      " ['1'\n",
      "  '(Read more about how we rate cars.)The Continentals power cloth convertible top is a high-quality woven unit that does a great job of damping out ambient road noise while also folding up or down in seconds.']\n",
      " ['1'\n",
      "  \"(We'd suggest skipping the sunroof for long torsos or anyone taller than 6-foot-3.) The seats are firmer and more supportive, however, which should help for long drives.Luckily, in Plus and Exclaim versions, theres a fold-down center armrest that could serve as a useful dividing line for at-odds siblings.\"]\n",
      " ['1'\n",
      "  'A fold-down rear seat extends the usefulness of the trunk—and on the less practical side, BMW also includes a standard windblock, three-setting neck warmers, and more sound-deadening materials for a longer driving season and for a quieter ride than you might expect from a convertible.The Gran Coupe is an oddity in the lineup with its long, sloping roofline.']\n",
      " ['1'\n",
      "  'A handful of owners have reported glitches or freezing up in the graphic display software, and Chevy--just as other makers do with their infotainment systems--has offered minor software updates.Your mother might not ever know she was driving a Volt from behind the wheel, except for the lack of engine noise and perhaps the futuristic humming tone as the car boots up and does its system checks.But without the masking effect of engine and transmission noise, that silence poses a new challenge: Engineers have to work that much harder to make accessories like wipers seem quiet, and either reduce or insulate against wind noise, tire roar, and a host of other sounds.']\n",
      " ['1'\n",
      "  \"A power tailgate, standard this year, makes loading cargo easier.The 2011 MDX is crafted from high-quality materials, and though its busy-looking dash has plenty of buttons, lines, and lights, it's functionally not a difficult place to work.\"]\n",
      " ['1'\n",
      "  'All grades get an acoustic glass windshield, while the cabin itself has been sealed, and silenced, in various ways ranging from below the carpet to within the headliner.From the cabin, the V-6 is less gruff and coarse than it sounded in last years model, and road and wind noise have been significantly cut.']\n",
      " ['1'\n",
      "  'All trim levels of the Titan offer a damped tailgate that softly lowers without banging on the bumper—or can be raised with a single hand—thanks to a strut hidden behind the taillight.Nissan offers in-box storage bins, called Titan Box, that hold tools when needed, or removed when the full bed width is needed.']]\n"
     ]
    }
   ],
   "source": [
    "print(mat_array[0:10])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array(['1',\n",
       "       '(Lateral boosters can pinch in the sides if youve found a gym in the last few years too.)Weve found that the front seats are widely adjustable and we think that most body types wont have a hard time finding a natural position in the drivers seat—provided you can swipe the keys first.Outward visibility in the Q60 is generally good, and bulges over the front tires should help drivers place the front wheels despite the coupes long hood.'],\n",
       "      dtype='<U2630')"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "mat_array[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "mat=np.array([[1,2,3],[1,2,3]])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "np.savetxt('test.csv', mat, delimiter = ',')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "np.savetxt('input.csv',mat_array,fmt='%s',delimiter=',')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# import libraries\n",
    "import fastai\n",
    "from fastai import *\n",
    "from fastai.text import * \n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "from functools import partial\n",
    "import io\n",
    "import os"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df=DataFrame(mat_array)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df=df.sample(frac=1).reset_index(drop=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>0</th>\n",
       "      <th>1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>The instant recognizability of the silhouette ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>4</td>\n",
       "      <td>An optional 6.8-liter V10 produces 362 hp and ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>6</td>\n",
       "      <td>The L features 16-inch steel wheels, air-condi...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5</td>\n",
       "      <td>It provides minimal legroom even for kids and ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3</td>\n",
       "      <td>And in the new, revised federal tests for 2011...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5</td>\n",
       "      <td>With seating for 12 passengers or room for a r...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>4</td>\n",
       "      <td>The 2008 Subaru Legacy lineup spans a price ra...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>2</td>\n",
       "      <td>The SE's recalibrated suspension gives it even...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>2</td>\n",
       "      <td>An optional package bundles stability and trac...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>7</td>\n",
       "      <td>We find Subaru's CVT to be one of the best of ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   0                                                  1\n",
       "0  1  The instant recognizability of the silhouette ...\n",
       "1  4  An optional 6.8-liter V10 produces 362 hp and ...\n",
       "2  6  The L features 16-inch steel wheels, air-condi...\n",
       "3  5  It provides minimal legroom even for kids and ...\n",
       "4  3  And in the new, revised federal tests for 2011...\n",
       "5  5  With seating for 12 passengers or room for a r...\n",
       "6  4  The 2008 Subaru Legacy lineup spans a price ra...\n",
       "7  2  The SE's recalibrated suspension gives it even...\n",
       "8  2  An optional package bundles stability and trac...\n",
       "9  7  We find Subaru's CVT to be one of the best of ..."
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df.columns = ['label','text']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>label</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1</td>\n",
       "      <td>The instant recognizability of the silhouette ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>4</td>\n",
       "      <td>An optional 6.8-liter V10 produces 362 hp and ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>6</td>\n",
       "      <td>The L features 16-inch steel wheels, air-condi...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>5</td>\n",
       "      <td>It provides minimal legroom even for kids and ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>3</td>\n",
       "      <td>And in the new, revised federal tests for 2011...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5</td>\n",
       "      <td>With seating for 12 passengers or room for a r...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>4</td>\n",
       "      <td>The 2008 Subaru Legacy lineup spans a price ra...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>2</td>\n",
       "      <td>The SE's recalibrated suspension gives it even...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>2</td>\n",
       "      <td>An optional package bundles stability and trac...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>7</td>\n",
       "      <td>We find Subaru's CVT to be one of the best of ...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "  label                                               text\n",
       "0     1  The instant recognizability of the silhouette ...\n",
       "1     4  An optional 6.8-liter V10 produces 362 hp and ...\n",
       "2     6  The L features 16-inch steel wheels, air-condi...\n",
       "3     5  It provides minimal legroom even for kids and ...\n",
       "4     3  And in the new, revised federal tests for 2011...\n",
       "5     5  With seating for 12 passengers or room for a r...\n",
       "6     4  The 2008 Subaru Legacy lineup spans a price ra...\n",
       "7     2  The SE's recalibrated suspension gives it even...\n",
       "8     2  An optional package bundles stability and trac...\n",
       "9     7  We find Subaru's CVT to be one of the best of ..."
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df['text'] = df['text'].str.replace(\"[^a-zA-Z]\", \" \")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[nltk_data] Downloading package stopwords to\n",
      "[nltk_data]     /home/I348655/nltk_data...\n",
      "[nltk_data]   Package stopwords is already up-to-date!\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import nltk\n",
    "nltk.download('stopwords')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from nltk.corpus import stopwords \n",
    "stop_words = stopwords.words('english')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# tokenization \n",
    "tokenized_doc = df['text'].apply(lambda x: x.split())\n",
    "\n",
    "# remove stop-words \n",
    "tokenized_doc = tokenized_doc.apply(lambda x: [item for item in x if item not in stop_words]) \n",
    "\n",
    "# de-tokenization \n",
    "detokenized_doc = [] \n",
    "for i in range(len(df)): \n",
    "    t = ' '.join(tokenized_doc[i]) \n",
    "    detokenized_doc.append(t) \n",
    "df['text'] = detokenized_doc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "# split data into training and validation set\n",
    "df_trn, df_val = train_test_split(df, stratify = df['label'], test_size = 0.3, random_state = 12)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((70900, 2), (30386, 2))"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_trn.shape, df_val.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Language model data\n",
    "data_lm = TextLMDataBunch.from_df(train_df=df_trn, valid_df=df_val, path = \"\")\n",
    "\n",
    "# Classifier model data\n",
    "data_clas = TextClasDataBunch.from_df(path = \"\", train_df = df_trn, valid_df = df_val, vocab=data_lm.train_ds.vocab, bs=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn = language_model_learner(data_lm, pretrained_model=URLs.WT103, drop_mult=0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "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": 39,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYsAAAEKCAYAAADjDHn2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nIpLU/wFCzGCujSidagAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.save_encoder('ft_enc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn = text_classifier_learner(data_clas, drop_mult=0.5)\n",
    "learn.load_encoder('ft_enc')\n",
    "learn.freeze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "gc.collect();"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEKCAYAAAD9xUlFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nKCmIiEgLJQUREWnx/wG1SRQMNIfzrwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.lr_find()\n",
    "learn.recorder.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 01:54 <p><table style='width:300px; margin-bottom:10px'>\n",
       "  <tr>\n",
       "    <th>epoch</th>\n",
       "    <th>train_loss</th>\n",
       "    <th>valid_loss</th>\n",
       "    <th>accuracy</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>1</th>\n",
       "    <th>1.165452</th>\n",
       "    <th>0.653785</th>\n",
       "    <th>0.782367</th>\n",
       "  </tr>\n",
       "</table>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit_one_cycle(1, 2e-2, moms=(0.8,0.7))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.save('first')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.load('first');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 02:05 <p><table style='width:300px; margin-bottom:10px'>\n",
       "  <tr>\n",
       "    <th>epoch</th>\n",
       "    <th>train_loss</th>\n",
       "    <th>valid_loss</th>\n",
       "    <th>accuracy</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>1</th>\n",
       "    <th>0.622753</th>\n",
       "    <th>0.369122</th>\n",
       "    <th>0.880438</th>\n",
       "  </tr>\n",
       "</table>\n"
      ],
      "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": 49,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.save('second')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.load('second');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 02:34 <p><table style='width:300px; margin-bottom:10px'>\n",
       "  <tr>\n",
       "    <th>epoch</th>\n",
       "    <th>train_loss</th>\n",
       "    <th>valid_loss</th>\n",
       "    <th>accuracy</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>1</th>\n",
       "    <th>0.426709</th>\n",
       "    <th>0.282734</th>\n",
       "    <th>0.906799</th>\n",
       "  </tr>\n",
       "</table>\n"
      ],
      "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": 52,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.save('third')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.load('third');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 06:36 <p><table style='width:300px; margin-bottom:10px'>\n",
       "  <tr>\n",
       "    <th>epoch</th>\n",
       "    <th>train_loss</th>\n",
       "    <th>valid_loss</th>\n",
       "    <th>accuracy</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>1</th>\n",
       "    <th>0.353056</th>\n",
       "    <th>0.256581</th>\n",
       "    <th>0.915092</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>2</th>\n",
       "    <th>0.355785</th>\n",
       "    <th>0.241961</th>\n",
       "    <th>0.920292</th>\n",
       "  </tr>\n",
       "</table>\n"
      ],
      "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": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category 2,\n",
       " tensor(1),\n",
       " tensor([0.0281, 0.6674, 0.0317, 0.1439, 0.0739, 0.0043, 0.0507]))"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"This car is very motivated\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category 2,\n",
       " tensor(1),\n",
       " tensor([0.0102, 0.5740, 0.0239, 0.1115, 0.1097, 0.0028, 0.1679]))"
      ]
     },
     "execution_count": 56,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"The outlook of the car is not as good as expected\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category 5,\n",
       " tensor(4),\n",
       " tensor([0.3817, 0.0392, 0.0076, 0.0210, 0.4930, 0.0015, 0.0559]))"
      ]
     },
     "execution_count": 57,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"Between the wheels there are 118.1 inches, and there are 200.8 inches bumper-to-bumper, which puts it on the small side of minivans on the road today. \")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(Category 1,\n",
       " tensor(0),\n",
       " tensor([9.8041e-01, 2.8604e-04, 1.5792e-03, 4.8620e-04, 1.7658e-03, 1.4486e-02,\n",
       "         9.8991e-04]))"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.predict(\"A chrome moustache dips under the rear window for a dressier look, replacing the former black plastic panel. \")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "learn.save(\"last\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}