03_hasoc_en.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#default_exp hasoc"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "from fastchai import config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%capture\n",
    "!pip install light-the-torch\n",
    "!ltt install torch torchvision\n",
    "!pip install fastai==2.1.4 fastcore==1.3.2 nbdev --upgrade\n",
    "\n",
    "#Ensure that you restart the runtime if it still shows the old version."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "ROOT='/content/gdrive/My Drive/Colab Notebooks/hasoc' if config.GDRIVE else './data' \n",
    "EN_DIR='en_task_a'\n",
    "FILENAME='hasoc_2020_en_train_new.csv'\n",
    "TEST='english_test.csv'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "from fastai.text.all import *\n",
    "from pathlib import Path"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "assert (Path(ROOT)/'en_task_a').exists()\n",
    "assert (Path(ROOT)/'en_task_a'/TEST).exists()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Loading Data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "def load_data(path:Path)->pd.DataFrame:\n",
    "    pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df = pd.read_csv(Path(ROOT)/EN_DIR/FILENAME, sep='\\t')\n",
    "assert len(df) == 3708\n",
    "\n",
    "df['tweet_id'] = df['tweet_id'].astype(str)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['is_valid'] = False"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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>tweet_id</th>\n",
       "      <th>text</th>\n",
       "      <th>task1</th>\n",
       "      <th>task2</th>\n",
       "      <th>ID</th>\n",
       "      <th>text_len</th>\n",
       "      <th>is_valid</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1.1237572634271867e+18</td>\n",
       "      <td>hate wen females hit ah nigga with tht bro 😂😂, I’m tryna make u my la sweety , fuck ah bro</td>\n",
       "      <td>HOF</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>hasoc_2020_en_2574</td>\n",
       "      <td>90</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1.1237333013977331e+18</td>\n",
       "      <td>RT @airjunebug: When you're from the Bay but you're really a NY nigga at heart. W/ @supportcaleon https://t.co/mZ8BAYlnlf</td>\n",
       "      <td>HOF</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>hasoc_2020_en_3627</td>\n",
       "      <td>121</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1.1237340941086595e+18</td>\n",
       "      <td>RT @DonaldJTrumpJr: Dear Democrats: The American people aren’t stupid, they know what spying is and no amount of gaslighting will change th…</td>\n",
       "      <td>NOT</td>\n",
       "      <td>NONE</td>\n",
       "      <td>hasoc_2020_en_3108</td>\n",
       "      <td>140</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1.126951188170199e+18</td>\n",
       "      <td>RT @SheLoveTimothy: He ain’t on drugs he just bored. I be doing the same shit when I’m bored 😂 https://t.co/tkdjSbddET</td>\n",
       "      <td>HOF</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>hasoc_2020_en_3986</td>\n",
       "      <td>118</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1.1268635104477102e+18</td>\n",
       "      <td>RT @TavianJordan: Summer ‘19 I’m coming for you ! No boring shit ! Beach days, road trips, kickbacks and HOT DAYS ! I’m ready I’m ready I’m…</td>\n",
       "      <td>NOT</td>\n",
       "      <td>NONE</td>\n",
       "      <td>hasoc_2020_en_5152</td>\n",
       "      <td>140</td>\n",
       "      <td>False</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                 tweet_id  \\\n",
       "0  1.1237572634271867e+18   \n",
       "1  1.1237333013977331e+18   \n",
       "2  1.1237340941086595e+18   \n",
       "3   1.126951188170199e+18   \n",
       "4  1.1268635104477102e+18   \n",
       "\n",
       "                                                                                                                                           text  \\\n",
       "0                                                    hate wen females hit ah nigga with tht bro 😂😂, I’m tryna make u my la sweety , fuck ah bro   \n",
       "1                     RT @airjunebug: When you're from the Bay but you're really a NY nigga at heart. W/ @supportcaleon https://t.co/mZ8BAYlnlf   \n",
       "2  RT @DonaldJTrumpJr: Dear Democrats: The American people aren’t stupid, they know what spying is and no amount of gaslighting will change th…   \n",
       "3                        RT @SheLoveTimothy: He ain’t on drugs he just bored. I be doing the same shit when I’m bored 😂 https://t.co/tkdjSbddET   \n",
       "4  RT @TavianJordan: Summer ‘19 I’m coming for you ! No boring shit ! Beach days, road trips, kickbacks and HOT DAYS ! I’m ready I’m ready I’m…   \n",
       "\n",
       "  task1 task2                  ID  text_len  is_valid  \n",
       "0   HOF  PRFN  hasoc_2020_en_2574        90     False  \n",
       "1   HOF  PRFN  hasoc_2020_en_3627       121     False  \n",
       "2   NOT  NONE  hasoc_2020_en_3108       140     False  \n",
       "3   HOF  PRFN  hasoc_2020_en_3986       118     False  \n",
       "4   NOT  NONE  hasoc_2020_en_5152       140     False  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "True"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_csv(Path(ROOT)/EN_DIR/TEST, sep=',')\n",
    "test_df['is_valid'] = True\n",
    "len(test_df) == 814"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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>tweet_id</th>\n",
       "      <th>text</th>\n",
       "      <th>task1</th>\n",
       "      <th>task2</th>\n",
       "      <th>ID</th>\n",
       "      <th>is_valid</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1130081762154090497</td>\n",
       "      <td>RT @delmiyaa: Samini resetting the show and moving things along as if nothing happened is all you need to know about his greatness.</td>\n",
       "      <td>NOT</td>\n",
       "      <td>NONE</td>\n",
       "      <td>hasoc_2020_en_2713</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1130048316807491584</td>\n",
       "      <td>@Swxnsea how do you know that he’s left?</td>\n",
       "      <td>HOF</td>\n",
       "      <td>NONE</td>\n",
       "      <td>hasoc_2020_en_3874</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1123657766143504386</td>\n",
       "      <td>Tried to get Divock Origi on a free seeing as he was at the club on loan. He accepted the offer, but for no actual… https://t.co/3WrM7NHjwV</td>\n",
       "      <td>NOT</td>\n",
       "      <td>NONE</td>\n",
       "      <td>hasoc_2020_en_281</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1126782963042013186</td>\n",
       "      <td>RT @nutclusteruwu: that....is yalls stupid white ass reactions meeting tom holland in disneyland? are you fucking kidding me i would have d…</td>\n",
       "      <td>HOF</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>hasoc_2020_en_2026</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1130159113529434113</td>\n",
       "      <td>&amp;amp; IT DID. But a bitch got big girls things to do now 😇</td>\n",
       "      <td>HOF</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>hasoc_2020_en_4023</td>\n",
       "      <td>True</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              tweet_id  \\\n",
       "0  1130081762154090497   \n",
       "1  1130048316807491584   \n",
       "2  1123657766143504386   \n",
       "3  1126782963042013186   \n",
       "4  1130159113529434113   \n",
       "\n",
       "                                                                                                                                           text  \\\n",
       "0           RT @delmiyaa: Samini resetting the show and moving things along as if nothing happened is all you need to know about his greatness.   \n",
       "1                                                                                                      @Swxnsea how do you know that he’s left?   \n",
       "2   Tried to get Divock Origi on a free seeing as he was at the club on loan. He accepted the offer, but for no actual… https://t.co/3WrM7NHjwV   \n",
       "3  RT @nutclusteruwu: that....is yalls stupid white ass reactions meeting tom holland in disneyland? are you fucking kidding me i would have d…   \n",
       "4                                                                                    &amp; IT DID. But a bitch got big girls things to do now 😇   \n",
       "\n",
       "  task1 task2                  ID  is_valid  \n",
       "0   NOT  NONE  hasoc_2020_en_2713      True  \n",
       "1   HOF  NONE  hasoc_2020_en_3874      True  \n",
       "2   NOT  NONE   hasoc_2020_en_281      True  \n",
       "3   HOF  PRFN  hasoc_2020_en_2026      True  \n",
       "4   HOF  PRFN  hasoc_2020_en_4023      True  "
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "all = pd.concat([df, test_df]);\n",
    "len(all)\n",
    "assert len(all) == 4522"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "False    3708\n",
       "True      814\n",
       "Name: is_valid, dtype: int64"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all['is_valid'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(['tweet_id', 'text', 'task1', 'task2', 'ID'],\n",
       " ['tweet_id', 'text', 'task1', 'task2', 'ID'])"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "list(df.columns), list(test_df.columns)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(HOF    1856\n",
       " NOT    1852\n",
       " Name: task1, dtype: int64, HOF    423\n",
       " NOT    391\n",
       " Name: task1, dtype: int64)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['task1'].value_counts(), test_df['task1'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(NONE    1852\n",
       " PRFN    1377\n",
       " OFFN     321\n",
       " HATE     158\n",
       " Name: task2, dtype: int64, NONE    414\n",
       " PRFN    293\n",
       " OFFN     82\n",
       " HATE     25\n",
       " Name: task2, dtype: int64)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df['task2'].value_counts(), test_df['task2'].value_counts()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "all['text_len'] = all['text'].apply(lambda x : len(x))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "98.47014595311809"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all['text_len'].sum()/len(all)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Language Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "def get_dataloaders(all:pd.DataFrame, is_lm:bool, config)->DataLoaders:\n",
    "    pass"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dls_lm = TextDataLoaders.from_df(all, valid_pct=0.1, seed=42, text_col='text', is_lm=True, bs=128, seq_len=140)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>text_</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>xxbos xxup rt xxunk : if this xxunk is trash i m gon na xxup actually cry it 's 8 xxup am xxup in xxup the xxup morning xxup here and xxunk xxunk and asian armys b … xxbos xxup rt @realdonaldtrump : xxmaj as most people know , and for those who would like to know , i am xxunk pro - life , with the three xxunk - xxmaj rape , xxup i … xxbos xxmaj getting out the house now ! xxbos xxup rt xxunk : i feel bad for all the people who got to be xxunk for jin like … … you really think you ’re gon na be able to xxmaj manage hi … xxbos xxunk \\n xxmaj you need to see this 😂 \\n https : / / t.co / xxunk xxbos xxup</td>\n",
       "      <td>xxup rt xxunk : if this xxunk is trash i m gon na xxup actually cry it 's 8 xxup am xxup in xxup the xxup morning xxup here and xxunk xxunk and asian armys b … xxbos xxup rt @realdonaldtrump : xxmaj as most people know , and for those who would like to know , i am xxunk pro - life , with the three xxunk - xxmaj rape , xxup i … xxbos xxmaj getting out the house now ! xxbos xxup rt xxunk : i feel bad for all the people who got to be xxunk for jin like … … you really think you ’re gon na be able to xxmaj manage hi … xxbos xxunk \\n xxmaj you need to see this 😂 \\n https : / / t.co / xxunk xxbos xxup rt</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dls_lm.show_batch(max_n=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "del learn_lm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn_lm = language_model_learner(dls_lm, AWD_LSTM, drop_mult=0.3, metrics=[accuracy, Perplexity()]).to_fp16()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "SuggestedLRs(lr_min=0.10964782238006592, lr_steep=0.02754228748381138)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYIAAAEKCAYAAAAfGVI8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3deXxV9Z3/8dfnZt/IAiFAIOyICrKIiBQVW7Wltm5drFq1ji0y2tVpp+3Mb9rOr9OZzqO/dlqr1qJtrV206riNqGO1LlgUDAqKArJDEiBhCdlDls/vj1xsDAkEyMm5uff9fDzuI8k9597zTpS88z3L95i7IyIiiSsSdgAREQmXikBEJMGpCEREEpyKQEQkwakIREQSnIpARCTBJYcd4FgNGTLEx4wZE3YMEZEBZeXKlXvcvbC7ZQOuCMaMGUNpaWnYMUREBhQz29bTMu0aEhFJcCoCEZEEpyIQEUlwKgIRkQSnIhARSXAqAhGRBJcwRXCwtZ2HXy9D026LiLxfwhTBw6+XccsDq/mnR9bQ2tYedhwRkZgx4C4oO15XnDGK7fsauOOFTeypa+bnV84gPSUp7FgiIqFLmBGBmfGPH5nMv158Ks+u3c3Vdy9nf/3BsGOJiIQuYYrgkOvmjuGOq2byVvkBLrn9r2zYXRt2JBGRUAVaBGaWZ2YPmdk6M1trZmd1WW5mdquZbTSzN81sZpB5DlkwdTj3L5xDY0sbl92xjOfW7u6PzYqIxKSgjxH8DHja3T9pZqlAZpflC4CJ0ceZwC+iHwM3sySfx7/4Ab5wbymfv7eUD00uYlRBBsNz0zl9dAGnj87vjxgiIqELrAjMLBc4B/gcgLsfBLrulL8EuNc7zul8NTqCGO7uO4PK1dnw3AwevHEuP3jyHV7dvI9XNu2h/mAbAAvPGcfXLzyJ1OSE23smIgkmyBHBWKAK+I2ZTQNWAl9x9/pO6xQDOzp9XRZ97n1FYGYLgYUAJSUlfRoyIzWJf7t0KgDuzoHGFn78zLssfmkzy7fs47YrZzCqoOtARkQkfgT5524yMBP4hbvPAOqBbx3PG7n7Ynef5e6zCgu7va9CnzAz8jJT+f6lU7jj6plsrqzjoluX8nbFgcPWfXrNLn7/6jZe27qPA40tvXr/lrZ2dh5o5O2KAyzdUMXanTW6wE1EQhfkiKAMKHP35dGvH+LwIigHRnX6emT0udB9dOpwpozI5YrFr3D9b17j4ZvmMjK/Y2Tws2c38F/Pvvu+9UsKMpk9toAzxxYwoySf/MwUctJTaHfnhfVVLHlrJ8+t3U1DdNfTIROHZnPpjGIWTBnGmMFZRCLWb99jX6pvbqWuuRUzMIz65lbK9jdStr+B/Q0tDMlOpWhQOkWD0hmSnUp+Zmq332tTSxt76w8SsY5ddyISPAvyL1IzWwp83t3Xm9n3gCx3/0an5RcBXwQ+SsdB4lvdffaR3nPWrFnen3coe3d3LZ/4xTKKBqXz0KKz+NXLW/j5XzbyiZkj+er5E9lQWcu6XbWs2l7Niq37qG54/+jADNxhcFYqH54yjKnFueRnppKfmcKGyjoeW1XOa1v3A5CdlsykomzGDMmiuqGFiupGdh5oYkh2KqeMyOWU4YM4fXQ+M0vySE7qfjBXVdvM02t2sq++hbb2dlranYLMVE4alsPkYTnkpKewobKW9btqKdvfSE56MrkZKeRnpjKxKJuSgkzMjlxG7k55dSOvb69m5dZ9vLZ1P+t21dB+DP8rJUWMgqxUUpMiuDtt7jQ0t1Hb3PreOmeMyedTs0Zx0dTh1DW3sqmyji176ynITGXSsBzGDM4i6QSKs7K2iYrqJjJSkshMTXrvZ3G0719kIDKzle4+q9tlARfBdOBuIBXYDFwPXAHg7ndax7+424CPAA3A9e5+xN/y/V0EAK9u3su1v1pBXmYKlbXNfOaMUfz7ZVMP+4u2vd15t7KWdypqqGlsobaplebWduaMG8yccQU9/vIu29/AS+/uYf2uGtbuqmXb3noGZ6VRnJ/BsEHp7KppYu3OGsr2NwKQl5nC/EmFzBk3mJz0FDLTkqhvbuXRNyp4fn0lbZ1+IydHjNZj+A2dl5nCaSPzOClaCqMKMklLTmL7vnq272tgY2Udb2yvprK2GYDM1CRmluRz+uh8hg5Kw72jKDJSkxmZn8HI/AzyMlPZW9fM7ppmdtc0saeumaraZvbWHaS13YkYRMzISE2iMCeNwVmp7Gs4yEOlZWzeU/9emXaVlhxhSHYaduj1KUmMKshkzOBMRuZnkJaSRFLESDKjubWdxpY2Gppb2VhVx+vb97NjX+Nh75mfmcLEohwmDM1mVH4mw3PTGZ6bztjCLAqz01QSMmCFVgRBCKMIAJa8uZMv3fc6V51Zwv+9eEoou3AONLSwbNMe/rx2N8+vq2R/l9FHYU4al88s5lOnj2TskOz3/lreV3+QdbtqWL+rltqmViYVZTOpKIdRBZk0trRxoKGFPXXNrN1Zy+od1awuq2bLnnqaW98/J1NSxCgpyGTayFxmjs5nZkk+k4fl9FhwJ8rdWbltPy+sr6JoUBrjC7MZPSSLfXUHWb+7lvW7athX34LjuENtUyvb99WzbW/DYdk7G5qTxumjO8przOAsmlvbaTjYyoHGFjZV1bFhdx0bq+oOG90VZKUyeVgOJw3LYeLQnPdGb9lpyaQlR1QSEtNUBH3kQEMLgzKSY+IffFu7U1HdSMPBNhoOttLuMG1kbp/9Um5vd6rqmtmxr4GmlnZKCjIZkZce2C/9vtTe7uxrOEhLWztt7U5bu5OWnERGShIZqUm9PiW4vrmVnQeaqKhuZFNVHet21rJ2Vw0bdtfR2NJ22Po5aclcNaeEr3xoIpmpCTONlwwQKgKRPtTe3nGMZGNlHdv21tPY0rHbaXNVHU+8uZMRuel89+JTufCUopj4o0EEjlwE+rNF5BhFIsao6PGTrj43dx//59E13Pi7leSkJTMoI4VBGSmMG5LFgqnD+ODkoRotSMzRiECkj7W0tfNgaRnv7q6lpqmFmsYWVu04wJ66ZtJTInzo5CKumDWKeROGDNjThWXg0YhApB+lJEW46sz3XwHf1u68tnUfS97cyRNvVrDkzZ2UFGRyxRmjOHdSIScPH3RCp8KKnAiNCET6WXNrG0+v2cUfl29n+ZZ9QMc1JDNK8jj/5CIumT6CvMzUkFNKvNHBYpEYVV7dSOnWfZRu3c+rm/eyobKO1KQIF5xSxNVnljB3wpCwI0qc0K4hkRhVnJdB8fRiLpleDMA7FTU8uHIHj62qYMlbO5k7fjBf//BJzCzRtOgSHI0IRGJQU0sb963Yzu3Pb2RP3UHmjh/MtFF5TCrKZvKwQUwelqNTU+WYaNeQyABV39zKPcu28ugb5WzZU//edCFnTxzCdz52ChOLckJOKAOFikAkDrS0tbNtbz0vrK/iZ89toOFgG9eeNZqvXTCJQekpYceTGKdjBCJxICUpwoShOUwYmsNlM4r58Z/f5Z5lW/nLukruuHomp47IDTuiDFCxP3GMiBxmcHYa/37ZVB5adBbNLe1cdscy7l+xXTc6kuOiIhAZwE4fXcCSL8/jzLEFfOvht/inR9563zTkIr2hIhAZ4AZnp3HP9bO5af547luxgy/f9wYHjzANt0hXOkYgEgeSIsY/fmQy+Zmp/ODJtTS2tHHH1TNJT0kKO5oMABoRiMSRL5wzjn+/bCrPr6/kc79ZQX2nW3+K9ERFIBJnrjqzhJ9eMZ3Xtu7n2l+voKap5egvkoSmIhCJQ5dML+a2K2ewekc119y9nAMNKgPpmYpAJE4tmDqcOz97Omt31nLlXa9S3XAw7EgSo1QEInHs/FOKWHzt6WysrGPhvStp6uZeyyIqApE4N/+kofzoU6exYus+/uHB1bTrOgPpQqePiiSAS6YXs/NAEz98ah3FeRn800dPDjuSxBAVgUiCuPGccVRUN7L4pc2MKsjkmjmjw44kMUJFIJIgzIzvfvxUyvY38q+Pv83EodnMGTc47FgSA3SMQCSBJEWMn35mOqMHZ3LTH15nx76GsCNJDFARiCSYQekp3HXtLFra2ln4u5U0HNTVx4lORSCSgMYVZnPbVTNZv6uG7zz2dthxJGSBFoGZbTWzt8xslZkddlsxM8s1s/8xs9Vm9raZXR9kHhH5m3MnFXLjueN5aGUZr2/fH3YcCVF/jAjOc/fpPdwi7WbgHXefBswHfmxmqf2QSUSAL543gaJBafzr42/r+oIEFvauIQdyzMyAbGAfoB2WIv0kKy2Zby2YzOqyA/z362Vhx5GQBF0EDjxjZivNbGE3y28DTgYqgLeAr7j7YXfUMLOFZlZqZqVVVVXBJhZJMJdMK2ZGSR7/+fR6ajVTaUIKugjmuftMYAFws5md02X5h4FVwAhgOnCbmQ3q+ibuvtjdZ7n7rMLCwoAjiySWSMT43sdPZU9dM7c9vzHsOBKCQIvA3cujHyuBR4DZXVa5HnjYO2wEtgCTg8wkIoebNiqPT50+kl8t3cKa8gNhx5F+FlgRmFmWmeUc+hy4EFjTZbXtwIei6xQBJwGbg8okIj3754tOpiArlX94YDXNrZqlNJEEOSIoAl42s9XACmCJuz9tZovMbFF0ne8Dc83sLeA54JvuvifATCLSg7zMVP7zk6exfnct//XnDWHHkX4U2FxD7r4ZmNbN83d2+ryCjpGCiMSA804aypWzR7H4pU1ccMpQTh9dEHYk6Qdhnz4qIjHmny86hRF5GdzywGoaD2oXUSJQEYjI+2SnJfOjT05j294Gbnteu4gSgYpARA5z1vjBXD6zmMUvbWZjZV3YcSRgKgIR6da3F5xMRkoS33lsDe6afiKeqQhEpFuFOWl84yOTWbZpL4+vrgg7jgRIRSAiPbpqdgmnjczl35aspUbTT8QtFYGI9CgpYvzbpVPYU9fMz57VgeN4pSIQkSM6bWQeV8waxW+XbdWB4zilIhCRo/r6h08iIzWJ7z/xjg4cxyEVgYgc1ZDsNL7yoYm8+G4Vz6+vDDuO9DEVgYj0ynVzxzC+MIvvP7GWg62H3TZEBjAVgYj0SkpShO98/FS27KnnnmVbwo4jfUhFICK9du6kQuafVMjtz2/S6aRxREUgIsfk6xeexIHGFu5+SbcOiRcqAhE5JlOKc7lo6nB+9fIW9tY1hx1H+oCKQESO2dcumERjSxt3vrgp7CjSB1QEInLMJgzN5rIZI/ntK9vYdaAp7DhyglQEInJcvnr+RNydn/9FU08MdCoCETkuowoyueKMUTxQuoOy/Q1hx5EToCIQkeN20/wJGMYdL+hYwUCmIhCR4zYiL4MrzhjFgxoVDGgqAhE5IX8/fzyARgUDmIpARE5I51FBeXVj2HHkOKgIROSE3TR/AgB3PL8x5CRyPFQEInLCDo0KdAbRwKQiEJE+cegMotuf17GCgUZFICJ9YkReBlfO7jhWsH2vRgUDSaBFYGZbzewtM1tlZqU9rDM/uvxtM3sxyDwiEqybzptAJGK62niA6Y8RwXnuPt3dZ3VdYGZ5wB3Axe5+KvCpfsgjIgEpGpTOZ88czcNvlLNlT33YcaSXwt41dBXwsLtvB3B33QxVZIBbNH8cKUnGz5/TqGCgCLoIHHjGzFaa2cJulk8C8s3sheg613b3Jma20MxKzay0qqoq0MAicmKG5qRz7VljeHRVOZuq6sKOI70QdBHMc/eZwALgZjM7p8vyZOB04CLgw8C/mNmkrm/i7ovdfZa7zyosLAw4soicqBvPGUdyUoTfLtsadhTphUCLwN3Lox8rgUeA2V1WKQP+193r3X0P8BIwLchMIhK8wdlpXDR1OA+/Xk59c2vYceQoAisCM8sys5xDnwMXAmu6rPYYMM/Mks0sEzgTWBtUJhHpP5+dU0JdcyuPraoIO4ocRZAjgiLgZTNbDawAlrj702a2yMwWAbj7WuBp4M3oOne7e9eyEJEBaGZJPpOH5fCH5dtw97DjyBEkB/XG7r6ZbnbzuPudXb7+EfCjoHKISDjMjM/OGc3/eXQNq3ZUM6MkP+xI0oOwTx8VkTh26YxislKT+P2r28OOIkegIhCRwGSnJXPpjGKeeLOC6oaDYceRHqgIRCRQn50zmubWdh4sLQs7ivRARSAigTp5+CBmjyngnmVbaW1rDzuOdENFICKB+/zZYymvbuSpNbvCjiLd6FURRK8JiEQ/n2RmF5tZSrDRRCRenH9yEWOHZHH30s06lTQG9XZE8BKQbmbFwDPANcA9QYUSkfgSiRg3zBvL6rIDrNiyL+w40kVvi8DcvQG4HLjD3T8FnBpcLBGJN5+YOZKCrFTuWrol7CjSRa+LwMzOAq4GlkSfSwomkojEo4zUJK6ZM5pn1+7WrKQxprdF8FXg28Aj7v62mY0Dng8ulojEo2vOGk1acoS7NSqIKb0qAnd/0d0vdvf/jB403uPuXw44m4jEmSHZaVw+s5hH3ihjf70uMIsVvT1r6I9mNig6i+ga4B0z+0aw0UQkHl03dwxNLe38qXRH2FEkqre7hk5x9xrgUuApYCwdZw6JiByTycMGcda4wfzulW26wCxG9LYIUqLXDVwKPO7uLXTchlJE5JhdN3cM5dWNPLtWtymPBb0tgl8CW4Es4CUzGw3UBBVKROLb+ScPpTgvQ7eyjBG9PVh8q7sXu/tHvcM24LyAs4lInEpOinDNWaN5ZfNe1u3S35Rh6+3B4lwz+4mZlUYfP6ZjdCAiclw+c8Yo0lMi/HbZtrCjJLze7hr6NVALfDr6qAF+E1QoEYl/eZmpXDpdp5LGgt4WwXh3/667b44+/hUYF2QwEYl/139gLE0t7fxxhe5gFqbeFkGjmc079IWZfQBoDCaSiCSKk4blcPbEIdyzbCvNrW1hx0lYvS2CRcDtZrbVzLYCtwE3BpZKRBLGF84eR1VtM0+s3hl2lITV27OGVrv7NOA04DR3nwF8MNBkIpIQzp44hElF2dylexWE5pjuUObuNdErjAFuCSCPiCQYM+Pz88axblctyzbtDTtOQjqRW1Van6UQkYR2yYwRDMlO466lm8OOkpBOpAg0hhORPpGWnMS1Z43mhfVVbNhdG3achHPEIjCzWjOr6eZRC4zop4wikgA+O2c06SkRjQpCcMQicPccdx/UzSPH3ZP7K6SIxL+CrFQ+PWsUj7xRzu6aprDjJJQT2TUkItKnPj9vHG3tzq//qjuY9adAiyB63cFbZrbKzEqPsN4ZZtZqZp8MMo+IxLaSwZl8dOpw/vjqdmqaWsKOkzD6Y0RwnrtPd/dZ3S00syTgP4Fn+iGLiMS4G88ZT21zK/ct17QT/SUWdg19CfhvQHeoEBGmjszlAxMG8+u/btG0E/0k6CJw4BkzW2lmC7suNLNi4DLgF0d6EzNbeGgK7KqqqoCiikisWHTueHbXNPPYGxVhR0kIQRfBPHefCSwAbjazc7os/ynwTXc/4o1L3X2xu89y91mFhYVBZRWRGDFvwhBOHj6Iu1/WtBP9IdAicPfy6MdK4BFgdpdVZgH3Ryey+yRwh5ldGmQmEYl9HdNOjOXd3XUs3bAn7DhxL7AiMLMsM8s59DlwIbCm8zruPtbdx7j7GOAh4CZ3fzSoTCIycHx82giG5mjaif4Q5IigCHjZzFYDK4Al7v60mS0ys0UBbldE4kBqcoTr5o5h6YY9rN+laSeCFFgRRO9kNi36ONXdfxB9/k53v7Ob9T/n7g8FlUdEBp6rzywhIyWJX72sUUGQYuH0URGRbuVlpvLJ00fy6BsVVNU2hx0nbqkIRCSm/d28sbS0t/O7V7aGHSVuqQhEJKaNHZLF+ScXce+r26hvbg07TlxSEYhIzPv7+eOpbmjhvhWadiIIKgIRiXkzS/KZM66Au5Zu1rQTAVARiMiAcPN5E9hd08wjr5eHHSXuqAhEZECYN2EIU4tzufPFTbS1a9qJvqQiEJEBwcy4af54tu5t4Mm3doYdp9/95M/v8tK7wUy6qSIQkQHjw6cOY1xhFne8sCmhJqNraWvntr9soHTb/kDeX0UgIgNGJGIsOnc8a3fW8EJAfx3Hot01TbQ7FOelB/L+KgIRGVAunV7M8Nx07nxhU9hR+k1FdRMAI/IyAnl/FYGIDCipyRFumDeW5Vv28fr2YHaVxJqK6kZARSAi8p4rZ5eQm5GSMKOC8kNFkKsiEBEBICstmevOGs0z7+xmY2X8T1FdUd1IQVYqGalJgby/ikBEBqTr5o4hPSXCnS/G/xTVFdWNjAjoQDGoCERkgBqcncZnzijh0TfK39uHHq8qqpsYHtBuIVARiMgAdsO8sbS789tXtoYdJVAV1Y0UB3SgGFQEIjKAjSrIZMGU4dy3fHvcTlFd09RCbXOrdg2JiPTk7+aNpaaplYdWloUdJRBBnzoKKgIRGeBOH53P9FF5/OavW2iPw8noVAQiIr3w+bPHsnVvA8+tqww7Sp8rj15VrGMEIiJH8JFTh1Gcl8HdS+PvVNKd1Y2kJBmF2WmBbUNFICIDXnJShM/NHcPyLftYU34g7Dh9qqK6kWG56UQiFtg2VAQiEheumD2K7LRkFr8UX6OCiuqmwKaWOERFICJxYVB6ClefWcITb1awfW9D2HH6THnA1xCAikBE4sjfzRtLciTCXXFyrKCt3dlV08TwAK8hABWBiMSRokHpXD6zmAdKd1BV2xx2nBNWWdtEW7sHeuooBFwEZrbVzN4ys1VmVtrN8qvN7M3oOsvMbFqQeUQk/i08ZxwH29q5Z9mWsKOcsP64hgD6Z0RwnrtPd/dZ3SzbApzr7lOB7wOL+yGPiMSxcYXZLJgyjHtf2UZtU0vYcU5If1xDACHvGnL3Ze5+6BZDrwIjw8wjIvFh0bnjqW1q5Q/Lt4cd5YQcGhEMzx3YxwgceMbMVprZwqOsewPwVMB5RCQBnDYyj3kThnD30i00tbSFHee4VVQ3Mig9mZz0lEC3E3QRzHP3mcAC4GYzO6e7lczsPDqK4Js9LF9oZqVmVlpVVRVcWhGJGzefN4E9dc08ULoj7CjHreOGNMHuFoKAi8Ddy6MfK4FHgNld1zGz04C7gUvcfW8P77PY3We5+6zCwsIgI4tInJgzroBZo/P55YubOdjaHnac41Je3RT48QEIsAjMLMvMcg59DlwIrOmyTgnwMHCNu78bVBYRSTxmxs0fnEB5dSOPvlEedpzjEg8jgiLgZTNbDawAlrj702a2yMwWRdf5DjAYuKOnU0xFRI7X/EmFTCkexC9e3ETbAJuiur65lQONLf1SBMlBvbG7bwYOuy7A3e/s9Pnngc8HlUFEEpuZ8cXzJrDo96+z5K2dXDxtRNiReu1v1xAEe8YQ6MpiEYlzF54yjIlDs7n9LxsH1I1rtkbnSxpVkBn4tlQEIhLXIhHjix+cwPrdtTy5ZmfYcXptU1UdAOMLswPflopAROLex04bwcSh2fz02Q0D5ljBpso6CnPSyM0I9hoCUBGISAJIihhfPX8SGyvr+J/VFWHH6ZVNVXWML8zql22pCEQkISyYMozJw3L42XMbaG2L7esK3J1NVfX9slsIVAQikiAiEeOWCyaxZU89D8f4dQV76w9yoLFFRSAi0tcuOKWIqcW53Prchpi+2nhTZfRA8VAVgYhInzLrGBWU7W/koZVlYcfp0aaqegAmqAhERPre/JMKmT4qj9uf3xizo4JNVXVkpCQxfFDwF5OBikBEEoyZ8bULJlFe3RizM5NuqqpjXGEWkYj1y/ZUBCKScM6ZOISZJR2jgubW2LtfQcepo/2zWwhUBCKSgDqOFZzEzgNN/Om12BoVNLW0Uba/UUUgIhK0D0wYzBlj8rn9+Y0xdRezzVX1uMP4of1zMRmoCEQkQZkZXzt/Ertrmvn9q9vCjvOe/pxj6BAVgYgkrLPGD+acSYXc+twG9tUfDDsO0FEEZjB2iEYEIiKBMzP+5aKTqT/Yxk/+vD7sOEDHNQQj8zNIT0nqt22qCEQkoU0syuGzZ5bwx+XbWberJuw4bKrs3zOGQEUgIsJXz59ETnoK33/iHdzDm6a6vd3ZvEdFICLS7/KzUvna+RP568a9/Pmd3aHlqDjQSFNLu4pARCQMV88ZzaSibL7z2NtUN4Rz4PjQHEP9dR+CQ1QEIiJASlKEn3x6Onvrm/n2w2+Fsoto3c6OYxT9NevoISoCEZGoKcW5fP3Ck3hqzS4eLO3/2UmfW1fJ5GE5DMlO69ftqghERDr5wtnjmDt+MN/7n7fZsqe+37a7t66Z0q37uPDUYf22zUNUBCIinUQixo8/PY2UpAhfuf8NWvrptpbPra2k3eHCU4r6ZXudqQhERLoYnpvBDy+fyptlB/j5cxv6ZZv/+/YuivMyOHXEoH7ZXmcqAhGRbiyYOpxPzBzJbc9vZOW2/YFuq765laUb93DhqUWY9c89CDpTEYiI9OB7F5/CiLwMbnlgFfXNrYFt58V3qzjY2s6HQzg+ACoCEZEe5aSn8JNPT2f7vga+/8Q7gW3nmbd3kZ+ZwqzR+YFt40gCLQIz22pmb5nZKjMr7Wa5mdmtZrbRzN40s5lB5hEROVazxxaw6Nzx3P/aDv6wvO+nqz7Y2s5z6yo5/+QikpPC+ds8uR+2cZ677+lh2QJgYvRxJvCL6EcRkZhxywWTWL+rln95dA2Ds1L5yJThffbey7fspbapNZTTRg8Je9fQJcC93uFVIM/M+u4nLCLSB1KSItx+1Uymjcrjy/et4pVNe/vkfZta2rjnr1vJSEni7IlD+uQ9j0fQReDAM2a20swWdrO8GOh8w9Cy6HPvY2YLzazUzEqrqqoCiioi0rOM1CR+fd0ZlAzOZOG9pazeUX1C77dlTz2X37GM59ZV8qUPTejX+w90FXQRzHP3mXTsArrZzM45njdx98XuPsvdZxUWFvZtQhGRXsrPSuXev5tNbmYKV971Ki++e3x/mC55cycfu3UpFQca+fXnZnHT/Al9nPTYBFoE7l4e/VgJPALM7rJKOTCq09cjo8+JiMSkEXkZPPz3cxkzOIsb7nmNh18/tjmJNlfV8ZX73+CkYTk8+eWz+eDk/r+SuKvAisDMssws59DnwIXAmi6rPQ5cGz17aA5wwN13BpVJRFe0JaEAAAhwSURBVKQvDB2Uzp9unMPssQXc8sBqFr+0qdev/eFT60hLjnDnNaczIi8jwJS9F+SIoAh42cxWAyuAJe7+tJktMrNF0XWeBDYDG4G7gJsCzCMi0mdy0lP4zfVn8LHThvPvT67jP55ae9Spq1/ZtJdn3tnNTedNYGhOej8lPbrATh91983AtG6ev7PT5w7cHFQGEZEgpSUn8bPPzCAvM4VfvriZ6voWfnDZlG6vB2hvd37w5DuMyE3nhnljQ0jbs/64jkBEJG4lRYzvXzKFgqw0bn1uA9v21XPNnDF86OSh7zsT6JE3yllTXsPPPjM91DOEuqMiEBE5QWbGLRdMomhQGj97dgM3//F1stOSmTdhCOkpERx4ecMepo3K4+OnjQg77mFUBCIifeTqM0fzmTNKWL55L4+tqmD5lr04YMCw3HR+cOkUIpH+n130aFQEIiJ9KClizJ0whLkTwrtS+FiFPcWEiIiETEUgIpLgVAQiIglORSAikuBUBCIiCU5FICKS4FQEIiIJTkUgIpLg7Giz5cUaM6sCDt1BOhc4cITPuz6XAvR0/+SedH6P3izr+lxvMx76OOQYM/ZXvkPP6WcYW/kGQsZYz3ciGY/0XKz9DEe7e/d39nL3AfsAFh/p867PAaUnso3eLOv6XG8zdvp4TBn7K59+hrGZbyBkjPV8J5LxKFlj6md4pMdA3zX0P0f5vKflx7uN3izr+lxvM8Z6vqNt60j0Mzz6do7kaK+L9Yyxnq+n5b3JeLTnjkXQP8MeDbhdQyfCzErdfVbYOY4k1jPGej6I/Yyxng9iP2Os54OBkfGQgT4iOFaLww7QC7GeMdbzQexnjPV8EPsZYz0fDIyMQIKNCERE5HCJNiIQEZEuVAQiIglORSAikuBUBFFmdraZ3Wlmd5vZsrDzdMfMImb2AzP7uZldF3aersxsvpktjf4c54edpztmlmVmpWb2sbCzdMfMTo7+/B4ys78PO093zOxSM7vLzP5kZheGnacrMxtnZr8ys4fCznJI9P+730Z/bleHnaeruCgCM/u1mVWa2Zouz3/EzNab2UYz+9aR3sPdl7r7IuAJ4LexmBG4BBgJtABlMZjPgTogPUbzAXwTeKAvs/VlRndfG/3/8NPAB2I046Pu/gVgEXBFDObb7O439GWu7hxj1suBh6I/t4uDznbMjuXKt1h9AOcAM4E1nZ5LAjYB44BUYDVwCjCVjl/2nR9DO73uASAnFjMC3wJujL72oRjMF4m+rgj4QwzmuwD4DPA54GOx+N84+pqLgaeAq2I1Y/R1PwZmxnC+Pv03coJZvw1Mj67zxyBzHc8jLm5e7+4vmdmYLk/PBja6+2YAM7sfuMTd/wPodreAmZUAB9y9NhYzmlkZcDD6ZVus5etkP5AWa/miu6uy6PiH2WhmT7p7eyxljL7P48DjZrYE+GNf5eurjGZmwA+Bp9z99VjL11+OJSsdI+SRwCpicE9MXBRBD4qBHZ2+LgPOPMprbgB+E1iiwx1rxoeBn5vZ2cBLQQaLOqZ8ZnY58GEgD7gt2GjAMeZz938GMLPPAXv6sgSO4Fh/hvPp2I2QBjwZaLK/Odb/D78EnA/kmtkEd78zyHAc+89wMPADYIaZfTtaGP2lp6y3AreZ2UUc/xQUgYnnIjhm7v7dsDMcibs30FFWMcndH6ajrGKau98TdoaeuPsLwAshxzgid7+Vjl9sMcnd99Jx/CJmuHs9cH3YOXoSc0OUPlQOjOr09cjoc7Ek1jMq34lTxhMX6/k6G0hZ3xPPRfAaMNHMxppZKh0HCR8POVNXsZ5R+U6cMp64WM/X2UDK+jdhH63uo6P39wE7+dtplTdEn/8o8C4dR/H/WRmVTxljO2Os5xuoWY/20KRzIiIJLp53DYmISC+oCEREEpyKQEQkwakIREQSnIpARCTBqQhERBKcikDigpnV9fP2+uSeFdZxD4cDZrbKzNaZ2f/rxWsuNbNT+mL7IqAiEOmWmR1xHi53n9uHm1vq7tOBGcDHzOxo9yG4lI4ZVEX6hIpA4paZjTezp81spXXcOW1y9PmPm9lyM3vDzJ41s6Lo898zs9+Z2V+B30W//rWZvWBmm83sy53euy76cX50+UPRv+j/EJ2mGTP7aPS5lWZ2q5k9caS87t5IxzTFxdHXf8HMXjOz1Wb232aWaWZz6bhfwY+io4jxPX2fIr2lIpB4thj4krufDnwduCP6/MvAHHefAdwP/GOn15wCnO/uV0a/nkzH1Nqzge+aWUo325kBfDX62nHAB8wsHfglsCC6/cKjhTWzfGAif5ti/GF3P8PdpwFr6ZjCYBkdc9d8w92nu/umI3yfIr2iaaglLplZNjAXeDD6Bzr87WY5I4E/mdlwOu4itaXTSx+P/mV+yBJ3bwaazaySjruvdb0N5wp3L4tudxUwho5bdm5290PvfR+wsIe4Z5vZajpK4Kfuviv6/BQz+zc67u+QDfzvMX6fIr2iIpB4FQGqo/veu/o58BN3fzx6I5jvdVpW32Xd5k6ft9H9v5nerHMkS939Y2Y2FnjVzB5w91XAPcCl7r46ejOd+d289kjfp0ivaNeQxCV3rwG2mNmnoOP2imY2Lbo4l7/NEX9dQBHWA+M63crwqDd5j44efgh8M/pUDrAzujvq6k6r1kaXHe37FOkVFYHEi0wzK+v0uIWOX543RHe7vE3HvWOhYwTwoJmtBPYEESa6e+km4OnodmqBA7146Z3AOdEC+RdgOfBXYF2nde4HvhE92D2enr9PkV7RNNQiATGzbHevi55FdDuwwd3/K+xcIl1pRCASnC9EDx6/TcfuqF+GnEekWxoRiIgkOI0IREQSnIpARCTBqQhERBKcikBEJMGpCEREEpyKQEQkwf1/f1g/ixk5pUQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_lm.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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>perplexity</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>5.805091</td>\n",
       "      <td>4.806961</td>\n",
       "      <td>0.222743</td>\n",
       "      <td>122.359207</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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>perplexity</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>4.570975</td>\n",
       "      <td>4.121800</td>\n",
       "      <td>0.309896</td>\n",
       "      <td>61.670147</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>4.248092</td>\n",
       "      <td>3.735633</td>\n",
       "      <td>0.342014</td>\n",
       "      <td>41.914543</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>4.016548</td>\n",
       "      <td>3.582706</td>\n",
       "      <td>0.351997</td>\n",
       "      <td>35.970737</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>3.835278</td>\n",
       "      <td>3.455148</td>\n",
       "      <td>0.364063</td>\n",
       "      <td>31.662966</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>3.685046</td>\n",
       "      <td>3.412324</td>\n",
       "      <td>0.368056</td>\n",
       "      <td>30.335676</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>3.580752</td>\n",
       "      <td>3.399411</td>\n",
       "      <td>0.371788</td>\n",
       "      <td>29.946455</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>3.500516</td>\n",
       "      <td>3.395023</td>\n",
       "      <td>0.370399</td>\n",
       "      <td>29.815334</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_lm.fine_tune(7, base_lr=2e-2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn_lm.save_encoder('fine_tuned_enc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "from fastai.learner import Learner\n",
    "\n",
    "def train(\n",
    "        dls:DataLoaders, \n",
    "        is_lm:bool,\n",
    "        config)->Learner:\n",
    "    \n",
    "    if is_lm:\n",
    "        pass\n",
    "    else:\n",
    "        pass"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Task 1 Classification - Hate Speech Detection"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dls_cls_tsk1 = DataBlock(\n",
    "    blocks=(TextBlock.from_df(text_cols='text', vocab=dls_lm.vocab), CategoryBlock),\n",
    "    get_x=ColReader('text'),\n",
    "    get_y=ColReader('task1'),\n",
    "    splitter=ColSplitter(col='is_valid'),\n",
    ").dataloaders(all, seq_len=140, bs=128)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>category</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>xxbos xxup rt xxunk : xxup dear xxup thanos \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - \\n - …</td>\n",
       "      <td>NOT</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dls_cls_tsk1.show_batch(max_n=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "del learn_cls_tsk1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<fastai.text.learner.TextLearner at 0x7f7d7f763fd0>"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk1 = text_classifier_learner(dls_cls_tsk1, AWD_LSTM, metrics=[accuracy, F1Score(average='macro')], loss_func=LabelSmoothingCrossEntropyFlat()).to_fp16()\n",
    "learn_cls_tsk1.load_encoder('fine_tuned_enc')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "FlattenedLoss of LabelSmoothingCrossEntropy()"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk1.loss_func"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "SuggestedLRs(lr_min=0.03019951581954956, lr_steep=1.5848931980144698e-06)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_cls_tsk1.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.811168</td>\n",
       "      <td>0.601881</td>\n",
       "      <td>0.692875</td>\n",
       "      <td>0.683235</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.653342</td>\n",
       "      <td>0.505722</td>\n",
       "      <td>0.807125</td>\n",
       "      <td>0.806726</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.549306</td>\n",
       "      <td>0.445110</td>\n",
       "      <td>0.853808</td>\n",
       "      <td>0.853782</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.478721</td>\n",
       "      <td>0.452239</td>\n",
       "      <td>0.864865</td>\n",
       "      <td>0.864504</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.424653</td>\n",
       "      <td>0.474120</td>\n",
       "      <td>0.875921</td>\n",
       "      <td>0.875822</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.382380</td>\n",
       "      <td>0.456109</td>\n",
       "      <td>0.869779</td>\n",
       "      <td>0.869772</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.346060</td>\n",
       "      <td>0.470737</td>\n",
       "      <td>0.859951</td>\n",
       "      <td>0.859943</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.317610</td>\n",
       "      <td>0.462400</td>\n",
       "      <td>0.868550</td>\n",
       "      <td>0.868541</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_cls_tsk1.fine_tune(7, 3e-2) #Label Smoothing works great and also we changed the validation set to the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.770102</td>\n",
       "      <td>0.640420</td>\n",
       "      <td>0.630728</td>\n",
       "      <td>0.614553</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.667972</td>\n",
       "      <td>0.566275</td>\n",
       "      <td>0.719677</td>\n",
       "      <td>0.718939</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.647416</td>\n",
       "      <td>0.542833</td>\n",
       "      <td>0.700809</td>\n",
       "      <td>0.698275</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.626427</td>\n",
       "      <td>0.530974</td>\n",
       "      <td>0.716981</td>\n",
       "      <td>0.716775</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.597920</td>\n",
       "      <td>0.510023</td>\n",
       "      <td>0.733154</td>\n",
       "      <td>0.732376</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.576861</td>\n",
       "      <td>0.478575</td>\n",
       "      <td>0.746631</td>\n",
       "      <td>0.746216</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.552693</td>\n",
       "      <td>0.489319</td>\n",
       "      <td>0.768194</td>\n",
       "      <td>0.767584</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.527822</td>\n",
       "      <td>0.474808</td>\n",
       "      <td>0.768194</td>\n",
       "      <td>0.767449</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.506994</td>\n",
       "      <td>0.477738</td>\n",
       "      <td>0.760108</td>\n",
       "      <td>0.759661</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.486780</td>\n",
       "      <td>0.472012</td>\n",
       "      <td>0.768194</td>\n",
       "      <td>0.768058</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.478131</td>\n",
       "      <td>0.466463</td>\n",
       "      <td>0.768194</td>\n",
       "      <td>0.768192</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#learn_cls_tsk1.fine_tune(10, 1.4e-3) #Old result not needed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "preds_raw, _, preds_tsk1 = learn_cls_tsk1.get_preds(dl=dls_cls_tsk1.valid, with_decoded=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(#814) ['NOT','NOT','NOT','HOF','HOF','NOT','HOF','HOF','HOF','HOF'...]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk1.dls.categorize.vocab[preds_tsk1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission_task1_df = test_df.drop(columns=['text', 'task1', 'task2', 'is_valid'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission_task1_df['task1'] = learn_cls_tsk1.dls.categorize.vocab[preds_tsk1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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>tweet_id</th>\n",
       "      <th>ID</th>\n",
       "      <th>task1</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>1130081762154090497</td>\n",
       "      <td>hasoc_2020_en_2713</td>\n",
       "      <td>NOT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1130048316807491584</td>\n",
       "      <td>hasoc_2020_en_3874</td>\n",
       "      <td>NOT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1123657766143504386</td>\n",
       "      <td>hasoc_2020_en_281</td>\n",
       "      <td>NOT</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1126782963042013186</td>\n",
       "      <td>hasoc_2020_en_2026</td>\n",
       "      <td>HOF</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1130159113529434113</td>\n",
       "      <td>hasoc_2020_en_4023</td>\n",
       "      <td>HOF</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "              tweet_id                  ID task1\n",
       "0  1130081762154090497  hasoc_2020_en_2713   NOT\n",
       "1  1130048316807491584  hasoc_2020_en_3874   NOT\n",
       "2  1123657766143504386   hasoc_2020_en_281   NOT\n",
       "3  1126782963042013186  hasoc_2020_en_2026   HOF\n",
       "4  1130159113529434113  hasoc_2020_en_4023   HOF"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "submission_task1_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission_task1_df.to_csv('submission_EN_A.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#export\n",
    "from fastai.learner import Learner\n",
    "from fastai.data.all import DataLoaders\n",
    "\n",
    "from typing import List\n",
    "\n",
    "def predict(learn:Learner, dls:DataLoaders)->List[str]:\n",
    "    pass"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Classification - Task 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "dls_cls_tsk2 = DataBlock(\n",
    "    blocks=(TextBlock.from_df(text_cols='text', vocab=dls_lm.vocab), CategoryBlock),\n",
    "    get_x=ColReader('text'),\n",
    "    get_y=ColReader('task2'),\n",
    "    splitter=ColSplitter(col='is_valid')\n",
    ").dataloaders(all, seq_len=140, bs=128)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<fastai.text.learner.TextLearner at 0x7f7d7f784a20>"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "del learn_cls_tsk2\n",
    "learn_cls_tsk2 = text_classifier_learner(dls_cls_tsk2, AWD_LSTM, metrics=[accuracy, F1Score(average='macro')], loss_func=LabelSmoothingCrossEntropyFlat()).to_fp16()\n",
    "learn_cls_tsk2.load_encoder('fine_tuned_enc')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Without Weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "SuggestedLRs(lr_min=0.025118863582611083, lr_steep=0.10000000149011612)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEKCAYAAAAIO8L1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+WH4yJAAAgAElEQVR4nO3dd3hVVfr28e+TTgqhJPQSepcWUBQBK9hRsWAvI+JYxvE31nHU15nRUaY4ig42xHHGjl0sY6GIBQKCdOkdIZQEkpOc5GS9f+SAAU9CgJyW3J/rykWy9z573znGPFlr7b2WOecQERE5UEy4A4iISGRSgRARkYBUIEREJCAVCBERCUgFQkREAlKBEBGRgOLCHaAmZWRkuKysrHDHEBGJGnPmzMl1zmUG2lerCkRWVhY5OTnhjiEiEjXMbG1l+9TFJCIiAalAiIhIQCoQIiISkAqEiIgEpAIhIiIBqUCIiEhAKhAiIlFs0aY8pv24LSjnVoEQEYli//l2Hb97Y35Qzq0CISISxQq9pSQnxAbl3CoQIiJRrNDro168CoSIiBzA4/WRkhicWZNUIEREopi6mEREJCB1MYmISECFXp9aECIi8kuFXh/1EjQGISIiB/B4S0lRC0JERCpyzlFYoi4mERE5QFFJGc6hLiYREdlfobcUQC0IERHZX6HXB6hAiIjIATwlewuEuphERKSCgmJ1MYmISAAefxdTPRUIERGpaO8YRIq6mEREpKLCErUgREQkgEKNQYiISCBRe5urmU00s61mtrCS/beb2Tz/x0Iz85lZI/++EWa2zMxWmNldwcooIhLNovk210nAiMp2OufGOef6OOf6AHcD05xzO8wsFngSOA3oDow2s+5BzCkiEpUKvaXExRgJccH5VR60AuGcmw7sqObho4FX/J8PBFY451Y557zAq8A5QYgoIhLVCop9QRughggYgzCzZMpbGpP9m1oC6yscssG/TUREKvAEcbEgiIACAZwFzHTOVbe1sR8zG2NmOWaWs23bthqOJiISuQpLfEF7BgIio0BczM/dSwAbgdYVvm7l3xaQc+4Z51y2cy47MzMzSBFFRCKPx1tae7uYzCwdGAq8W2HzbKCTmbUzswTKC8h74cgnIhLJCoqD28UUtLaJmb0CDAMyzGwDcD8QD+Ccm+A/7FzgU+dcwd7XOedKzewm4BMgFpjonFsUrJwiItGqsMRHer34oJ0/aAXCOTe6GsdMovx22AO3TwGm1HwqEZHaw+MtpXn9pKCdPxLGIERE5DAUen0kJ9bSMQgRETl8hXXgNlcRETkMhd7SoE2zASoQIiJRqazMUVRSRr14tSBERKSCvRP1pWgMQkREKirct9youphERKSCQq9/sSB1MYmISEXBXiwIVCBERKLSvgKRqC4mERGpwKMWhIiIBFLgH4PQba4iIrIftSBERCSgvWMQKRqDEBGRivbe5lprFwwSEZHDs+8uJo1BiIhIRYVeHwmxMcTFBu/XuAqEiEgU8nhLg7oWBKhAiIhEpUKvL6jdS6ACISISlQq9vqAOUIMKhIhIVAr2YkGgAiEiEpXUghARkYA8JT5SVCBERORABcXqYhIRkQA86mISEZFACkt8QZ2oD1QgRESiUqHXpy4mERHZX6mvDG9pWfS2IMxsopltNbOFVRwzzMzmmdkiM5tWYfsaM1vg35cTrIwiItGosCT4a0EABLN9MgkYD/w70E4zawA8BYxwzq0zsyYHHHKCcy43iPlERKLS3sWConaQ2jk3HdhRxSGXAG8559b5j98arCwiIrXJvsWCavEYRGegoZlNNbM5ZnZFhX0O+NS/fUxVJzGzMWaWY2Y527ZtC2pgEZFIEIrFgiC4XUzVuXZ/4CSgHvCNmX3rnPsRGOyc2+jvdvqfmS31t0h+wTn3DPAMQHZ2tgtRdhGRsCkMwXrUEN4WxAbgE+dcgX+sYTrQG8A5t9H/71bgbWBg2FKKiESYulAg3gUGm1mcmSUDRwNLzCzFzNIAzCwFOBWo9E4oEZG6xuPvYgr2cxBBO7uZvQIMAzLMbANwPxAP4Jyb4JxbYmYfAz8AZcBzzrmFZtYeeNvM9uZ72Tn3cbByiohEm1C1IIJWIJxzo6txzDhg3AHbVuHvahIRkV8qiPbbXEVEJDhC1cWkAhFBCr2l+25fExGpzN4upnpak7pu+HLpVoY8OpUT/zqNnDVVPV8oInWdx+sjKT6G2BgL6nVUIMLM4/Vx37sLuXrSbDJSE0iKj+GiZ77lmekrcU6PdYjILxWEYD1qCO+DcrXKIx8vBeA3J3UiqZrNvl2FXi6Y8A3Lt+7hV4Pb8bvhXfD6yrjzzR94aMpSZq3eybhRR9EwJSGY0UUkyhR6fUHvXgK1IGrEjgIvT09byb+mruSsJ75i4ca8ar3uX9NWsmLbHiZdPYB7z+xOUnws9ZPieerSftx3Znem/biV0/45g29Wbg/ydyAi0cTjDf5iQaACUSO+XLqVMgd3ndaVPE8J5z41kye/XEFZWeVdRFvzi3jx6zWM7NOSYV32n8jWzLhmcDveuuE46iXEcslz3/LXT5ZR4isL9rciIlGg0OsjOTH4HUAqEDXg86U/0SQtkTHHt+eTW4dwavdmjPtkGde8OJtdhd6Ar3lq6kpKfI5bT+5U6Xl7tUrng5sHM6pfK8Z/uYKzx888pAHsPE8J//l2LbuLSg75e9qrrMzx+ZKf2FOsu6tEIkWht5RkdTFFvuJSH9OWbeOkbk2JiTEapiQw/pK+/HFkT2auyOWMx79iwYb9u5w27vLw8nfruDC7FW0bp1R5/pTEOMZd0JsJl/Unr9DLqAnfcNvr89i4y4PH68NbWhawpfLDhl2c+cQM7n1nIZc8+x3b9xQf8vdWVua45+0FXPtiDpc+9x15hYdfaESk5hSGqItJg9QHmL9+F2/MWU/35un0bp1Ol6ZpxMVWXke/W7WDAq+Pk7v93E1kZlx+TFt6tUzn1/+Zw/kTvmbs0A5cfWwWDVMSePyz5QDcfGLlrYcDjejZjCGdMxj/xQqenbGKt+ZurHA96N2qAcN7NOPUHk2Z8eM2/jxlCZmpidx7RjfGfbKMC5/+hpeuPZoWDeoB5UsWFpb4qJ8UH/B6e4vDq7PXc+ZRzfl00U+MfvZbXrp2II1TE6ude6+iEh9xMbbfe1niK2P++l2s3LaH03o1rzSLiOzP4/UF/SlqAKtNt1JmZ2e7nJwjW6H0lle+5735m/Z9XS8+lguyW/HrYR1plp70i+Pve3chr+esZ959pwa8e2lHgZd731nAlAVbSE6I5bx+LXll1nquGNSW+8/qcVgZV+cW8PmSnygtc/jKHAXFpcxckcv8Ci2Vk7o24W8X9qZBcgKz1+zgmkmzSUuM48IBrZmzdidz1+7EU+JjzJAO3Hry/ndelZU57n5rAa/lrOeWkzrx25M7MX15LmP+nUObRsk8f+UAWjWsR0yM4ZxjxdY9TPtxGzNX5NIuI5UbhnUgM628iBSV+Hh62iqenLoC5xxtGiXTLiOV0rIyZq3ese+Bn67N0ph09cCA77GI7O+Yhz5naOdMHhl11BGfy8zmOOeyA+5Tgdjf4Ee+oGeLdO4+vSvz1u9ixvJc3vl+IzExxmVHt93vl59zjsGPfEm35vV57sqA7+8+y7bs5l9TV/De/E0kxsUy7Y5hNEmr2V+Gm3Z5+GzJT6QkxHFev5b4JzwEYNGmPK6cOIvcPV66NktjQFYjCrylvDV3I+0zUxg36ijSkuL5aMEWpizYzLKfdvObkzrx21M67zvHNyu3c+2Lsyn0+ogxaJCcQIwZuf7uq3YZKazbUUhCbAzXDM6iT+uGPDRlCatzCzijV3PaNE5m9bYCVucW4HAc074xx3ZojJlx22vzSK8Xz4vXDKRT07Qae088Xh9z1+3EW2GAv1fLdDIOoxUkEil6PfAJ5/drxQNnH94fmRWpQFTT1t1FDPzz5/z+9G5cN6T9vu3rdxTyxBfLmTx3IxmpCbw59lhaN0pmyeZ8TvvnDB45vxcXDWhTrWus215IYUkpXZvVP+ych6uoxEdxSRnpyT935cxYvo27Ji9g4y4PUN5dld22IRdmt+aC7Na/OMeKrbuZsTyXHQVedhR48ZT4GJDViCGdM2nZoB6rcwv4+/9+5H1/K6xdRgoPntOD4ztlVplt4cY8rp40m+ISH09fns2gDo332+8rc0yeu4G12wvwlUGZcxiQGB9LvfhY6sXH0DAlgcy0RDJTE1m7vZD35m/isyU/7Wul7FU/KY4Hz+nJOX1a7FdERaJFx3umMGZIe+4Y0fWIz6UCUU0fL9zC2P/MYfINg+jfttEv9i/cmMclz35Lo5QEXh87iNdmredv//uRWb8/qcZbA6G0p7iUF79eQ/168Qzv3pQm9Y/8e1m8KZ9lP+VzWs/m1X5wcP2OQq58YRarcwsYM6Q9t53SmcS4WDbt8nDb6/P4dtUOYmOMWDNiYsA5KC6t/NbfBsnxnNazOcN7NCW9XnlR9JT4+NunPzJn7U5O69mMP43seVhjKiLhUlzqo8u9H3P78C7ceELHIz5fVQVCg9QVfL9uJ/GxRo8W6QH392yZzgtXD+Sy577jiudnAdC7dYOoLg4AqYlxNfKDVlH3FvXp3uLQWkmtGyXz3k2D+fOHi3l62iqmLt3GRQNa89hnP1Ja5hg36ihG9W+131/9ZWUOr6+MguJSdhZ62bbby7Y9xaTXi+fYDo2JD3CDwdHXN+bZGav4+6c/Mmv1dO4c0ZVR/VsRE+R5bURqQr6n/Jbz+vWCf1OHbnOtYO66nfRokV7lX7z92zbk6cv7s3LbHpZu2c0p3ZpUeqwcutTEOB4+7yheuGoAOwu9PPjBYtpnpjLlluO5ILv1L7qEYmKMpPhYGqcm0rFJGoM6NObs3i0Y2jkzYHEAiI0xxg7twPs3DyYrI4U7Jv/AuU/N5Pt1O0PxLYockTxP+e3m9ZP0oFzIeEvL+GFDHv3aNDzosUM6Z/LPi/uS1TiZs3q3CEG6uueErk349LdDeOrSfrwxdhBZGVU/L3I4ujRL482xg3jsoj5szivi3Ke+5jevfs+67YU1fi2RmpLvf/A1FC0IdTH5LdmcT3FpGf3aNqjW8af3as7pvZoHOVXd1iA5IejvsZkxsm9LTu7elKe+XMHEmauZsmAzlwxsw40ndoz67kOpffa2INJVIEJnrr97oTotCKl9UhPjuGNEV648NovHPlvOf75bx7+/XUvPFukc26Exx3XMYFAlYxoioZS/r4tJBSJk5q7bRbP6SfueNJa6qWn9JB4+rxfXHd+O9+Zv4uuV25k4czVPT19FRmoC5/drxYUDWtMhMzXcUaWOylcLIvTmrt1Z7e4lqf3aZ6Zy68mdufXk8onRZq7Yzhs563nuq/JiMSCrIRf0b83pRzUntcKsms45PVshQZVftPcuJi0YFBJb84vYuMvD1cdlhTuKRKDkhDhO6d6UU7o3ZevuIibP2cgbOeu5Y/IP3P/eIo7vlMGe4lI27fKwKa+I9hkpXJjdmpF9W9JIiz1JDcvzlJAUH0NinGZzDYm94w99Nf4gB9EkLYkbhnXg8/8byuQbjmVk3xYs3bIbT4mPni3TufyYtiTExfDgB4s5+qHPuPmV71m/Q3dFSc3J95SEbGJLtSAoH39IiI2hZ8vQT38h0cnM6N+2If3bBv6jYumWfF6fvYFXZ6/j00VbuGFYB8YO7VDtp8pFKpNfVBKS8QdQCwIoH3/o0bJ+SJpsUjd0bVaf+87qzuf/N5RTujflsc+Wc8o/pjFj+bZwR5Mol+cpCckzEKACUf6A3MbqPSAncqiap9dj/CX9ePm6o0mIjeHy52dx/7sL8RwwgaBIdeV7SiOrBWFmKWYW4/+8s5mdbWZVJjSziWa21cwWVnHMMDObZ2aLzGxahe0jzGyZma0ws7uq+80cjrgY48ObB3PloKxgXkbquGM7ZPDhLcdzzXHtePGbtZzxxAxmr9lBbZosU0Ijz1MSkmk2oPotiOlAkpm1BD4FLgcmHeQ1k4ARle00swbAU8DZzrkewAX+7bHAk8BpQHdgtJl1r2bOQxYTY3RqmkabxsnBuoQIAEnxsdx3Vnde/tXRFHl9XDDhG05//CsmzVxd6drlIgfKL4q8LiZzzhUC5wFPOecuAKpcqcI5Nx3YUcUhlwBvOefW+Y/f6t8+EFjhnFvlnPMCrwLnVDOnSMQ7tmMGn942lD+N7ElcjPHA+4s5+qHPeW32unBHkwhXVubI90TeILWZ2SDgUuBD/7YjHdHtDDQ0s6lmNsfMrvBvbwmsr3DcBv+2yoKNMbMcM8vZtk0DgBIdUhPjuOyYtrx/82A+uHkwA7IacefkBTz4/mJKfZWvcSF1W4G3lDIXmmk2oPoF4lbgbuBt59wiM2sPfHmE144D+gNnAMOBP5hZ56pf8kvOuWecc9nOuezMzKpXLROJRD1bpjPp6gFcdWwWE2eu5poXc/ZNyCZSUSgn6oNqPgfhnJsGTAPwD1bnOuduOcJrbwC2O+cKgAIzmw709m+vuNZlK2DjEV5LJKLFxcbwwNk96Nw0jfveXciIx6bz+zO6cUav5pq6Q/b5ebGgCBqkNrOXzay+maUAC4HFZnb7EV77XWCwmcWZWTJwNLAEmA10MrN2ZpYAXAy8d4TXEokKlxzdhtfHDqJhcgI3vfw9lzz7HT/+tDvcsSRC7FssKMLGILo75/KBkcBHQDvK72SqlJm9AnwDdDGzDWZ2rZmNNbOxAM65JcDHwA/ALOA559xC51wpcBPwCeUF43Xn3KLD+N5EolK/Ng15/+bB/HFkTxZvzueMx2fwRs76g79Qar19iwVF2FQb8f7nHkYC451zJWZW5Q3czrnRBzupc24cMC7A9inAlGpmE6l1YmOMy49pyxm9mnPLK99z+5s/sHZ7Ibed0llrZ9dhoZzqG6rfgngaWAOkANPNrC2QH6xQIlKuUUoCL1w9gNEDWzP+yxXc8ur3egq7Dgt1F1N1B6kfBx6vsGmtmZ0QnEgiUlF8bAwPnduLrMYpPPzRUqYu28apPZpydu8WHNcxQ6vc1SH5RaWYQVpiaAapq3UVM0sH7geG+DdNAx4E8oKUS0QqMDOuH9qB/m0b8trs9Xy8aAtvzd1I+8wU3rnxuJD1SUt45XtKSEuMC1k3Y3X/9JgI7AYu9H/kAy8EK5SIBJad1YhxF/Qm596TeeyiPqzdXsj97+oejroiP4QzuUL1B6k7OOfOr/D1/zOzecEIJCIHlxgXy8i+LVmdW8A/P1/OSd2acOZRLcIdS4IsL4TTbED1WxAeMxu89wszOw7wBCeSiFTXTSd2pHfrBvz+7YVsySsKdxwJsvyi0K0mB9UvEGOBJ81sjZmtAcYD1wctlYhUS3xsDI9d1AdvaRm/e2M+ZWWaPrw2i8gWhHNuvnOuN3AUcJRzri9wYlCTiUi1tMtI4d4zu/HVilyGPzadp6auYNMuNfBro3xPacim2YBDXFHOOZfvf6Ia4LYg5BGRw3DJwDY8ev5R1K8Xz6MfL+O4R77gun/nsFGFolYJdQviSEqRHucUiRBmxoUDWnPhgNas3V7A5DkbeHbGak75+zTuGN6FywdlEasnsKOat7QMT4kvIscgAlFnp0gEats4hdtO7cKnvx3CgKxGPPD+YkZN+JqlWzT5QTTb7Z+HKT05QgqEme02s/wAH7sB3VMnEsFaN0pm0tUD+OfF5c9LnPn4V4z7ZClFJZqqIxrtm2YjUloQzrk051z9AB9pzrnQjZSIyGExM87p05LPbhvKOX1a8uSXKxnx2HS+Xpkb7mhyiPKLQrsWBBxZF5OIRIlGKQn87cLe/PdXR+OAS579jrsm/6CV66JIqFeTAxUIkTrluI4ZfHLrEK4f2p7Xc9Zzyt+n8b/FP4U7llRDfqR1MYlI7ZMUH8vdp3Xj3RsH0zg1kTEv5bBgg+bdjHRqQYhIyPRqlc7r1x9Do+QE/vjBYpzTjYmRbN9qcioQIhIKaUnx3HZqZ2at2cHHC7eEO45UIc9TQkJcDEnxsSG7pgqESB13UXZrujRN46GPllBcqltgI1W+pzTk636oQIjUcXGxMdx7ZjfW7/Dwwsw14Y4jlcj3lJAewltcQQVCRIDjO2VyYtcmjP9iBbl7isMdRwLILwrtYkGgAiEifvec3o2iEh93TV6gacMjUL4ntGtBgAqEiPh1bJLKPad347MlPzH+yxXhjiMHCPVMrqACISIVXH1cFuf2bck/PvuRz5foAbpIkl8U2rUgQAVCRCowMx46txfdm9fn1lfnsWrbnnBHEsA5pxaEiIRfvYRYJlzWn7hY4+pJs1m8SdOEh1uh14evzNWeMQgzm2hmW81sYSX7h5lZnpnN83/cV2HfGjNb4N+eE6yMIhJY60bJPH/VADxeHyOfmsl/vl2rJ63DKBzTbEBwWxCTgBEHOWaGc66P/+PBA/ad4N+eHZx4IlKVfm0aMuU3xzOofWPufWchN738PYXe0nDHqpPCMc0GBLFAOOemAzuCdX4RCb6M1EReuGoAd47oykcLN3PX5AVqSYRBXmHta0FUxyAzm29mH5lZjwrbHfCpmc0xszFVncDMxphZjpnlbNu2LbhpReqgmBjjhmEduO2Uzrw3fxMvfbs23JHqnH2LBdWWMYhqmAu0dc71Bp4A3qmwb7Bzrh9wGnCjmQ2p7CTOuWecc9nOuezMzMzgJhapw349rCMndm3CHz9YzPfrdoY7Tp1SG8cgquScy3fO7fF/PgWIN7MM/9cb/f9uBd4GBoYrp4iUi4kx/n5hb5rWT+LG/85lR4E33JHqjF2F5e91nSkQZtbMzMz/+UB/lu1mlmJmaf7tKcCpQMA7oUQktBokJ/DUpf3I3ePlppfnavbXEFm3o5C0pLja86Ccmb0CfAN0MbMNZnatmY01s7H+Q0YBC81sPvA4cLErH/1qCnzl3z4L+NA593GwcorIoTmqVQMePq8XX6/czv+9Pl/zNoXA6twCshqn4P+bOmSCVo6cc6MPsn88MD7A9lVA72DlEpEjd37/VuTuKebhj5aSkZrI/Wd1D/kvr7pkzfYC+rRuGPLrhra9IiK1xpgh7dm2u5jnvlpNZloiN57QMdyRaiVvaRkbd3o4t0/LkF9bBUJEDouZcc/p3cjdU8y4T5YRY+W3w0rNWr+zkDIHWRkpIb+2CoSIHLaYGGPcBb3xOXjk46XkF5Vwx/Au6m6qQWtyCwBo21gFQkSiTHxsDI9d1Ie0pDj+NXUlu4tKePDsnsTEqEjUhNX+AtFOLQgRiUaxMcafR/YkLSmOp6etIiE2lvvO6h7uWLXC2u2F1E+Ko2FyaJ+BABUIEakhZsZdI7pSXFLGxJmrGdShMad0bxruWFFvzfYC2mWE/hZXCP9cTCJSi5gZd5/elZ4t63P7m/PZnOcJd6Sotzq3ICzjD6ACISI1LDEulidG96OktIzfvDKPUl9ZuCNFreJSH5t2ecJyBxOoQIhIELTLSOFP5/Zk1podPP7FinDHiVrrd3goc9AuIzks19cYhIgExbl9W/HV8u08/vlyUhJiuX6onpE4VHtvcc0KUxeTCoSIBM1D5/WkuNTHwx8tJXdPMXef1k23vx6CNdtVIESklkqMi+Xxi/uSkZrIszNWs213MY+O6k1CnHq3q2N1bgHp9eJpmJIQluurQIhIUMXEGPef1Z3MtETGfbKMJZt389B5vejfNvSTz0WbtdsLwzZADRqkFpEQMDNuPKEjz12Rze6iEkZN+Jp731mwb6U0CWx1bgHtGodngBpUIEQkhE7u3pRPbxvK1ce24+Xv1nHBhK8pKtGiQ4EUlfjYlOcJ2zMQoAIhIiGWmhjHfWd159krsvnxpz2M122wAa3fUYhz4ZmDaS8VCBEJi5O6NeX8fq2YMG0lizblhTtOxFmzvRAIzzTfe6lAiEjY/OHMbjRITuDOyT/oiesD/PwMhMYgRKQOapCcwB/P6cHCjfk8O2N1uONElNXbC2iQHE+D5PDc4goqECISZqf1as6IHs34x2c/8vWK3HDHiRhrcgvC9oDcXioQIhJ2D47sQeuG9bjs+e94etpKnHPhjhR263YU0jaM3UugAiEiEaBJWhLv3jSYET2b8fBHS/n1f+eyp7g03LHCxlfm2JJXRIsG9cKaQwVCRCJCamIcT17Sj3tO78oni7Zw5cRZeEvr5sB17p5iSsucCoSIyF5mxpghHfjnxX2Zs3YnD7y/KNyRwmLTrvKFllqkJ4U1h+ZiEpGIc1bvFizalM+EaSvp1TKd0QPbhDtSSG3OKwKgebpaECIiv3D78C4c3ymD+95dyJy1O8MdJ6T2tSAahLcFEbQCYWYTzWyrmS2sZP8wM8szs3n+j/sq7BthZsvMbIWZ3RWsjCISuWJjjCdG96V5ej2ufymH9+ZvoqysbtzdtDmviHrxsaTXiw9rjmC2ICYBIw5yzAznXB//x4MAZhYLPAmcBnQHRptZ9yDmFJEI1SA5geevzCYjNZFbXvmes8Z/xYzl28IdK+g253lo3iAJs/AurhS0AuGcmw7sOIyXDgRWOOdWOee8wKvAOTUaTkSiRqemaUy55Xj+cVFvdhWWcPnzs7j7rR/w1eLWxKZdRbQI8/gDhH8MYpCZzTezj8ysh39bS2B9hWM2+LeJSB0VE2Oc27cVX/xuKGOHduCVWev5zavfU1JL52/anOeheZjvYILw3sU0F2jrnNtjZqcD7wCdDvUkZjYGGAPQpk3dutNBpK5JjIvlrtO60iA5nr98tBSP18eTl/YjKT423NFqTImvjK27i2ke5mcgIIwtCOdcvnNuj//zKUC8mWUAG4HWFQ5t5d9W2Xmecc5lO+eyMzMzg5pZRCLD2KEd+NPInnyxbCu/ejGnVj1Q91N+Ec6F/xkICGOBMLNm5h+BMbOB/izbgdlAJzNrZ2YJwMXAe+HKKSKR6bJj2vLI+Ufx1Ypc/vjB4nDHqTH7noGIgBZE0LqYzOwVYBiQYWYbgPuBeADn3ARgFHCDmZUCHuBiVz5DV6mZ3QR8AsQCE51zdfNxShGp0oXZrVmxdQ/PTF9Fjxb1ubgWPFAXKU9RQxALhHNu9EH2jwfGV7JvCjAlGLlEpBj7pQEAAA1PSURBVHa5c0RXlmzO5w/vLqRT01T6t20U7khHJJJaEOG+i0lE5IjExhjjR/ejRYN6XP/SXH7KLwp3pCOyeZeHtKQ4UhPDPxOSCoSIRL305HievSKbPcUl3PPWgqheT2JTXmQ8AwEqECJSS3RumsbvTu3C50u38u68TeGOc9j2PkUdCVQgRKTWuPq4dvRr04AH3l/E1t3R2dW0eVdR2Gdx3UsFQkRqjdgY49FRvSn0+rj/3ei7+bGoxMf2Am9E3MEEKhAiUst0bJLKrSd34qOFW/jwh83hjnNItkTQHUygAiEitdCY49vTq2U697+3kB0F3nDHqbZNeZHzDASoQIhILRQXG8O4C44iz1PCg1G0bOnmXWpBiIgEXddm9bnxhI68M28Tny3+KdxxqmWzvwURCTO5ggqEiNRivx7Wka7N0rjn7QXkeUrCHeegNuUV0SglIWJmp1WBEJFaKyEuhnGjerO9wMufP4z8Cf027YqMdSD2UoEQkVqtV6t0rh/SntdzNvD29xvCHadKkfQMBKhAiEgdcOvJnRnUvjF3vPkDM1fkhjtOpTbleWgRIU9RgwqEiNQBCXExPH1FfzpkpjL2pTks2Zwf7ki/sKe4lN1FpWpBiIiEWv2keF64egApiXFc9cIsNvrXXQinbbuL+W7VdnLW7GDqsq0AEdWCCP98siIiIdI8vR6TrhnABRO+4fLnv+O1MYPITEsMeY6FG/OY+NVq3v9hEyW+/Weebds4JeR5KqMCISJ1Stdm9Zl41QCueH4Wlz33Ha+OOYaGKQkhuXaht5TrX5rDjOW5JCfEcunRbTmxaxMAfM6RHB9L71bpIclSHSoQIlLnDMhqxPNXZnPVpNlcPvE7/vurY0ivFx/0606YupIZy3O5fXgXLjumbUiueSQ0BiEiddKxHTN4+vL+LNuym2smzabUVxbU623a5eGZGas486jm3HhCx4gvDqACISJ12AldmjBuVG/mrN3Jaznrg3qtRz9eSpmDu07rGtTr1CQVCBGp087p04IBWQ35x/+Ws6e4dL99eZ4SVucWHPE15q3fxTvzNnHd8e1o1TD5iM8XKioQIlKnmRn3nN6N3D3FPDN91b7tOwu8nP+vrznhr1M596mZvD57PQUHFJDKFJf6KPF3WTnn+OMHi8lMS+SGYR2D8j0EiwapRaTO69umIWcc1Zxnp6/i0qPbkJoYx9WTZrNuRyE3DOvAp4u2cMfkH3jwg8X8aWRPRvZtWem5dhV6Gf7YdLbuLiYjNZFGyQks+2k3j5zfi9TE6PqVG11pRUSC5M7hXfl00RYe/XgZ2/YU88OGXfzrsv4M79GMO4Z3Yc7anTz68TJufW0eizfnc+eIrsTG2C/O8/CUpeTu8XL9kA7sLPCyJb+I7i3qM6p/6zB8V0dGBUJEBGjTOJnLj8li4szVADx6/lEM79EMKO+Gys5qxH+vO5o/frCYZ6avYumW3TxxcV/Sk3++G+m7Vdt5LWc91w9tH1WD0ZXRGISIiN/NJ5avH3HvGd24cMAv/+KPj43hwXN68vB5vfhmZS5njp/B9+t2AuXjDne/vYDWjepx60mdQx09KILWgjCzicCZwFbnXM8qjhsAfANc7Jx707/NByzwH7LOOXd2sHKKiOzVMCWBj28dctDjRg9sQ5dmadz88vdcMOEbbh/ehUKvj1XbCnjxmoHUS4iMBX+OVDC7mCYB44F/V3aAmcUCjwCfHrDL45zrE7xoIiJHpl+bhkz5zfHcNfkHHv5oKQBn927B0M6ZYU5Wc4LWxeScmw7sOMhhNwOTga3ByiEiEizp9eJ56tJ+/GlkT45p34g/nNk93JFqVNgGqc2sJXAucAIw4IDdSWaWA5QCf3HOvRPqfCIi1WFmXHZMWy47pm24o9S4cN7F9Bhwp3OuzOwXt4q1dc5tNLP2wBdmtsA5tzLQScxsDDAGoE2bNkENLCJSl4TzLqZs4FUzWwOMAp4ys5EAzrmN/n9XAVOBvpWdxDn3jHMu2zmXnZlZe/r+RETCLWwFwjnXzjmX5ZzLAt4Efu2ce8fMGppZIoCZZQDHAYvDlVNEpK4K5m2urwDDgAwz2wDcD8QDOOcmVPHSbsDTZlZGeQH7i3NOBUJEJMSCViCcc6MP4dirKnz+NdArGJlERKT69CS1iIgEpAIhIiIBqUCIiEhA5pwLd4YaY2bbgLVAOpDn33ywz/f+mwHkHuIlK56vuvsO3F7dfBW3HWrWqnJWtr+qnAfLGqz39EizRuN//2jKqp/Vmssayv/+bZ1zgZ8RcM7Vug/gmep+XuHfnCO5TnX3Hbi9uvmOJGtVOSvbX1XOaryXQXlPjzRrNP73j6as+lmNzp/Vqj5qaxfT+4fwecVtR3Kd6u47cPuh5DvcrAd7XaD9VeU88OsDswbrPa1sf3WzRuN//4qfR3pW/awefH8k/qxWqlZ1MR0JM8txzmWHO0d1REvWaMkJyhos0ZI1WnJCaLPW1hbE4Xgm3AEOQbRkjZacoKzBEi1ZoyUnhDCrWhAiIhKQWhAiIhKQCoSIiASkAiEiIgGpQFSDmR1vZhPM7Dkz+zrceSpjZjFm9mcze8LMrgx3nqqY2TAzm+F/X4eFO8/BmFmKmeWY2ZnhzlIZM+vmfz/fNLMbwp2nKmY20syeNbPXzOzUcOepipm1N7PnzezNcGcJxP+z+aL//by0Js9d6wuEmU00s61mtvCA7SPMbJmZrTCzu6o6h3NuhnNuLPAB8GKk5gTOAVoBJcCGYOSswawO2AMkRUFWgDuB14OTssZ+Tpf4f04vpHwdlUjO+o5z7jpgLHBRhGdd5Zy7NlgZAznE3OcBb/rfz7NrNMjhPF0XTR/AEKAfsLDCtlhgJdAeSADmA90pn2b8gwM+mlR43etAWqTmBO4Crve/9s1Ifk+BGP/rmgL/jfCspwAXA1cBZ0ZqTv9rzgY+Ai6J5Pe0wuv+BvSLkqxB+3/qCHPfDfTxH/NyTeYI55rUIeGcm25mWQdsHgiscOVLmmJmrwLnOOceBgJ2IZhZGyDPObc7UnP6F2by+r/0BSNnTWWtYCeQGIycUGPv6zAghfL/GT1mNsU5VxZpOf3neQ94z8w+BF6uyYw1mdXKF6L/C/CRc25uMHLWVNZwOJTclLfAWwHzqOFeoVpfICrRElhf4esNwNEHec21wAtBSxTYoeZ8C3jCzI4HpgczWACHlNXMzgOGAw2A8cGN9guHlNU593sAM7sKyK3p4lCFQ31Ph1He3ZAITAlqsl861J/Vm4GTgXQz6+iqXmWyph3q+9oY+DPQ18zu9heScKgs9+PAeDM7gyObjuMX6mqBOGTOufvDneFgnHOFlBeyiOece4vyghY1nHOTwp2hKs65qcDUMMeoFufc45T/Yot4zrntlI+VRCTnXAFwdTDOXesHqSuxEWhd4etW/m2RJlpygrIGQ7TkBGUNhZDnrqsFYjbQyczamVkC5QOQ74U5UyDRkhOUNRiiJScoayiEPneoRuXD9QG8Amzm51s/r/VvPx34kfK7An6vnMqqnMoaKR+RkluT9YmISEB1tYtJREQOQgVCREQCUoEQEZGAVCBERCQgFQgREQlIBUJERAJSgZBazcz2hPh6NbJeiJWvl5FnZvPMbKmZ/bUarxlpZt1r4voioAIhckjMrMr5y5xzx9bg5WY45/oAfYEzzexgazyMpHzGWZEaoQIhdY6ZdTCzj81sjpWvatfVv/0sM/vOzL43s8/MrKl/+wNm9pKZzQRe8n890cymmtkqM7ulwrn3+P8d5t//pr8F8F//FNeY2en+bXPM7HEz+6CqvM45D+VTObf0v/46M5ttZvPNbLKZJZvZsZSvBTHO3+roUNn3KVJdKhBSFz0D3Oyc6w/8DnjKv/0r4BjnXF/gVeCOCq/pDpzsnBvt/7or5dOVDwTuN7P4ANfpC9zqf2174DgzSwKeBk7zXz/zYGHNrCHQiZ+ncH/LOTfAOdcbWEL5NAxfUz4vz+3OuT7OuZVVfJ8i1aLpvqVOMbNU4FjgDf8f9PDzgkWtgNfMrDnlK3atrvDS9/x/ye/1oXOuGCg2s62Ur4x34NKps5xzG/zXnQdkUb7M6irn3N5zvwKMqSTu8WY2n/Li8Jhzbot/e08z+xPla2mkAp8c4vcpUi0qEFLXxAC7/H37B3oC+Ltz7j3/4jsPVNhXcMCxxRU+9xH4/6XqHFOVGc65M82sHfCtmb3unJsHTAJGOufm+xcxGhbgtVV9nyLVoi4mqVOcc/nAajO7AMqXvjSz3v7d6fw8v/6VQYqwDGhfYTnJiw72An9r4y/Anf5NacBmf7fWpRUO3e3fd7DvU6RaVCCktks2sw0VPm6j/Jfqtf7um0WUr+sL5S2GN8xsDpAbjDD+bqpfAx/7r7MbyKvGSycAQ/yF5Q/Ad8BMYGmFY14FbvcPsneg8u9TpFo03bdIiJlZqnNuj/+upieB5c65f4Q7l8iB1IIQCb3r/IPWiyjv1no6zHlEAlILQkREAlILQkREAlKBEBGRgFQgREQkIBUIEREJSAVCREQCUoEQEZGA/j9vxiu6MX5uzgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_cls_tsk2.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.395348</td>\n",
       "      <td>0.974703</td>\n",
       "      <td>0.681818</td>\n",
       "      <td>0.377805</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.064528</td>\n",
       "      <td>0.839850</td>\n",
       "      <td>0.793612</td>\n",
       "      <td>0.477176</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.920528</td>\n",
       "      <td>0.746002</td>\n",
       "      <td>0.816953</td>\n",
       "      <td>0.530273</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.827596</td>\n",
       "      <td>0.750408</td>\n",
       "      <td>0.818182</td>\n",
       "      <td>0.543239</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_cls_tsk2.fine_tune(3, 2.5e-2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.349215</td>\n",
       "      <td>1.039876</td>\n",
       "      <td>0.595687</td>\n",
       "      <td>0.318555</td>\n",
       "      <td>00:04</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>0.996707</td>\n",
       "      <td>0.895949</td>\n",
       "      <td>0.654987</td>\n",
       "      <td>0.375300</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>0.890318</td>\n",
       "      <td>0.742220</td>\n",
       "      <td>0.716981</td>\n",
       "      <td>0.456922</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>0.730161</td>\n",
       "      <td>0.688776</td>\n",
       "      <td>0.787062</td>\n",
       "      <td>0.475367</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.603235</td>\n",
       "      <td>0.682857</td>\n",
       "      <td>0.776280</td>\n",
       "      <td>0.472363</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.500495</td>\n",
       "      <td>0.684234</td>\n",
       "      <td>0.776280</td>\n",
       "      <td>0.494802</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#learn_cls_tsk2.fine_tune(5, 1.9e-2) #Not needed anymore Old one"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### With weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "  def get_weights(dls):\n",
    "    \n",
    "      # 0th index would provide the vocab from text\n",
    "      # 1st index would provide the vocab from classes\n",
    "      classes = dls.vocab[1]\n",
    "\n",
    "      #Get label ids from the dataset using map\n",
    "      #train_lb_ids = L(map(lambda x: x[1], dls.train_ds))\n",
    "      # Get the actual labels from the label_ids & the vocab\n",
    "      #train_lbls = L(map(lambda x: classes[x], train_lb_ids))\n",
    "\n",
    "      #Combine the above into a single\n",
    "      train_lbls = L(map(lambda x: classes[x[1]], dls.train_ds))\n",
    "      label_counter = Counter(train_lbls)\n",
    "      n_most_common_class = max(label_counter.values()); \n",
    "      print(f'Occurrences of the most common class {n_most_common_class}')\n",
    "      \n",
    "      # Source: https://discuss.pytorch.org/t/what-is-the-weight-values-mean-in-torch-nn-crossentropyloss/11455/9\n",
    "      weights = [1/v for k, v in label_counter.items() if v > 0]; return weights "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Occurrences of the most common class 1852\n"
     ]
    }
   ],
   "source": [
    "weights = get_weights(dls_cls_tsk2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[0.0007262164124909223,\n",
       " 0.0005399568034557236,\n",
       " 0.003115264797507788,\n",
       " 0.006329113924050633]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "class_weights = torch.FloatTensor(weights).to(dls_cls_tsk2.device)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn_cls_tsk2 = text_classifier_learner(dls_cls_tsk2, AWD_LSTM, metrics=[accuracy, F1Score(average='macro')]).to_fp16()\n",
    "learn_cls_tsk2.load_encoder('fine_tuned_enc')\n",
    "learn_cls_tsk2.loss_func = LabelSmoothingCrossEntropyFlat(weight=class_weights)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "FlattenedLoss of LabelSmoothingCrossEntropy()"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk2.loss_func"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "SuggestedLRs(lr_min=0.02089296132326126, lr_steep=0.005248074419796467)"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn_cls_tsk2.lr_find()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.473587</td>\n",
       "      <td>1.188835</td>\n",
       "      <td>0.606880</td>\n",
       "      <td>0.416698</td>\n",
       "      <td>00:05</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "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_score</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.254503</td>\n",
       "      <td>1.067564</td>\n",
       "      <td>0.681818</td>\n",
       "      <td>0.470836</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>1.173859</td>\n",
       "      <td>0.924063</td>\n",
       "      <td>0.745700</td>\n",
       "      <td>0.492076</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>1.059327</td>\n",
       "      <td>0.822301</td>\n",
       "      <td>0.785012</td>\n",
       "      <td>0.496157</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>0.965528</td>\n",
       "      <td>0.793076</td>\n",
       "      <td>0.793612</td>\n",
       "      <td>0.488469</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.892712</td>\n",
       "      <td>0.777327</td>\n",
       "      <td>0.807125</td>\n",
       "      <td>0.516852</td>\n",
       "      <td>00:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.843603</td>\n",
       "      <td>0.777120</td>\n",
       "      <td>0.803440</td>\n",
       "      <td>0.516741</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_cls_tsk2.fine_tune(6, 5e-3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "(#3) [0.7849021553993225,0.8071253299713135,0.5111347519098411]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk2.validate(ds_idx=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>input</th>\n",
       "      <th>target</th>\n",
       "      <th>predicted</th>\n",
       "      <th>probability</th>\n",
       "      <th>loss</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>xxbos xxmaj no shade , no tea about xxmaj xxunk xxmaj xxunk . xxmaj he ’s a very beautiful man . xxmaj and i hear tell that he ’s very good at xxunk s … https : / / t.co / xxunk xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>NONE</td>\n",
       "      <td>0.9070174694061279</td>\n",
       "      <td>4.375500679016113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>xxbos xxup rt xxunk : 🇺 🇲 we 're xxup coming 🇺 🇲 \\n xxmaj we 're taking our country by xxunk , \" bigger than ever before \" . xxmaj people from all xxunk are getting smart . xxup b … xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad</td>\n",
       "      <td>OFFN</td>\n",
       "      <td>NONE</td>\n",
       "      <td>0.9218553304672241</td>\n",
       "      <td>4.256222248077393</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>xxbos xxup rt xxunk : i ca n’t wait to xxunk all my friends with my xxunk ! \\n\\n “ get in loser , we ’re going to xxmaj xxunk ” \\n\\n xxmaj here ’s $ 10 , xxrep 3 0 . xxmaj start … xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>NONE</td>\n",
       "      <td>0.914258599281311</td>\n",
       "      <td>4.10779333114624</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>xxbos xxup rt xxunk : this is the cutest shit i ’ve ever seen xxunk https : / / t.co / xxunk xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad</td>\n",
       "      <td>NONE</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>0.9241439700126648</td>\n",
       "      <td>4.045537948608398</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>xxbos xxup rt xxunk : i fuck with you but you do n’t fuck with me xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad xxpad</td>\n",
       "      <td>OFFN</td>\n",
       "      <td>PRFN</td>\n",
       "      <td>0.9536961913108826</td>\n",
       "      <td>3.983731746673584</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "interp.plot_top_losses(k=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "interp = ClassificationInterpretation.from_learner(learn_cls_tsk2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "interp.plot_confusion_matrix()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[('NONE', 'PRFN', 29),\n",
       " ('OFFN', 'PRFN', 14),\n",
       " ('OFFN', 'NONE', 10),\n",
       " ('HATE', 'NONE', 9),\n",
       " ('NONE', 'OFFN', 9),\n",
       " ('PRFN', 'NONE', 8),\n",
       " ('HATE', 'PRFN', 5),\n",
       " ('NONE', 'HATE', 4),\n",
       " ('PRFN', 'OFFN', 3),\n",
       " ('HATE', 'OFFN', 1),\n",
       " ('OFFN', 'HATE', 1)]"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "interp.most_confused()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "('PRFN', tensor(3), tensor([0.0014, 0.0010, 0.0013, 0.9963]))"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk2.predict('What the fuck are you talking. go to hell')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "('OFFN', tensor(2), tensor([0.0061, 0.1064, 0.5032, 0.3843]))"
      ]
     },
     "execution_count": null,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn_cls_tsk2.predict(\"I will chop off your head and kill you\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "submission_task2_df = test_df.drop(columns=['text', 'task1', 'task2', 'is_valid'])\n",
    "\n",
    "preds_raw, _, preds_tsk2 = learn_cls_tsk2.get_preds(dl=dls_cls_tsk2.valid, with_decoded=True)\n",
    "submission_task2_df['task2'] = learn_cls_tsk2.dls.categorize.vocab[preds_tsk2]\n",
    "submission_task2_df.to_csv('submission_EN_B.csv', index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}