spanish_joke_classifier_tweet_only_LM_QRNN-LabelSmoothing[LM and Clas]-Regression.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Use the tweets for the updated vocab, test classifer on humor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/farzin/haha_2019\r\n"
     ]
    }
   ],
   "source": [
    "!pwd"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "%reload_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"1\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.text import *\n",
    "from fastai.callbacks import SaveModelCallback\n",
    "import sentencepiece as spm #https://github.com/google/sentencepiece"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Python code Examples (testing out tokenizer)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Reading the texts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "PATH = Path('./data/rnn/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_texts_df = pd.read_csv('./data/haha_2019_train.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "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>id</th>\n",
       "      <th>text</th>\n",
       "      <th>is_humor</th>\n",
       "      <th>votes_no</th>\n",
       "      <th>votes_1</th>\n",
       "      <th>votes_2</th>\n",
       "      <th>votes_3</th>\n",
       "      <th>votes_4</th>\n",
       "      <th>votes_5</th>\n",
       "      <th>funniness_average</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>705196579758583809</td>\n",
       "      <td>Niveles de retraso mental: \\r\\n\\r\\n— Bajo.\\r\\n...</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>678040651817213952</td>\n",
       "      <td>—Vamos Luke desenfunda tu sable, demuestra tu ...</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>546750892213829633</td>\n",
       "      <td>- ¿Te ofrezco algo?, ¿Agua, café, mi corazón, ...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2.6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>965807211292364801</td>\n",
       "      <td>No se porqué me hago la cabeza deooos</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>638403841839484928</td>\n",
       "      <td>Quisiera saber que hago durante la siesta de l...</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   id                                               text  \\\n",
       "0  705196579758583809  Niveles de retraso mental: \\r\\n\\r\\n— Bajo.\\r\\n...   \n",
       "1  678040651817213952  —Vamos Luke desenfunda tu sable, demuestra tu ...   \n",
       "2  546750892213829633  - ¿Te ofrezco algo?, ¿Agua, café, mi corazón, ...   \n",
       "3  965807211292364801              No se porqué me hago la cabeza deooos   \n",
       "4  638403841839484928  Quisiera saber que hago durante la siesta de l...   \n",
       "\n",
       "   is_humor  votes_no  votes_1  votes_2  votes_3  votes_4  votes_5  \\\n",
       "0         1         1        2        2        0        0        0   \n",
       "1         1         1        3        0        1        0        0   \n",
       "2         1         0        2        1        0        1        1   \n",
       "3         0         3        0        0        0        0        0   \n",
       "4         0         4        0        1        0        0        0   \n",
       "\n",
       "   funniness_average  \n",
       "0                1.5  \n",
       "1                1.5  \n",
       "2                2.6  \n",
       "3                NaN  \n",
       "4                NaN  "
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_texts_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "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>id</th>\n",
       "      <th>text</th>\n",
       "      <th>is_humor</th>\n",
       "      <th>votes_no</th>\n",
       "      <th>votes_1</th>\n",
       "      <th>votes_2</th>\n",
       "      <th>votes_3</th>\n",
       "      <th>votes_4</th>\n",
       "      <th>votes_5</th>\n",
       "      <th>funniness_average</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>705196579758583809</td>\n",
       "      <td>Niveles de retraso mental: \\r\\n\\r\\n— Bajo.\\r\\n...</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>678040651817213952</td>\n",
       "      <td>—Vamos Luke desenfunda tu sable, demuestra tu ...</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1.5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>546750892213829633</td>\n",
       "      <td>- ¿Te ofrezco algo?, ¿Agua, café, mi corazón, ...</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>2.6</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>965807211292364801</td>\n",
       "      <td>No se porqué me hago la cabeza deooos</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>638403841839484928</td>\n",
       "      <td>Quisiera saber que hago durante la siesta de l...</td>\n",
       "      <td>0</td>\n",
       "      <td>4</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                   id                                               text  \\\n",
       "0  705196579758583809  Niveles de retraso mental: \\r\\n\\r\\n— Bajo.\\r\\n...   \n",
       "1  678040651817213952  —Vamos Luke desenfunda tu sable, demuestra tu ...   \n",
       "2  546750892213829633  - ¿Te ofrezco algo?, ¿Agua, café, mi corazón, ...   \n",
       "3  965807211292364801              No se porqué me hago la cabeza deooos   \n",
       "4  638403841839484928  Quisiera saber que hago durante la siesta de l...   \n",
       "\n",
       "   is_humor  votes_no  votes_1  votes_2  votes_3  votes_4  votes_5  \\\n",
       "0         1         1        2        2        0        0        0   \n",
       "1         1         1        3        0        1        0        0   \n",
       "2         1         0        2        1        0        1        1   \n",
       "3         0         3        0        0        0        0        0   \n",
       "4         0         4        0        1        0        0        0   \n",
       "\n",
       "   funniness_average  \n",
       "0                1.5  \n",
       "1                1.5  \n",
       "2                2.6  \n",
       "3                0.0  \n",
       "4                0.0  "
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_texts_df.funniness_average.fillna(0,inplace=True)\n",
    "all_texts_df.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### apply pre-rules to text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "## New Line char for replacement in text\n",
    "NL = 'xxnl'\n",
    "defaults.text_spec_tok.append(NL) #add a New Line special char\n",
    "\n",
    "def sub_nl(t:str) -> str:\n",
    "    \"Replaces \\n by xxnl\"\n",
    "    return t.replace(\"\\r\\n\",\"\\n\").replace(\"\\n\",NL+\" \")\n",
    "\n",
    "# def sub_br(t:str) -> str:\n",
    "#     \"Replaces the <br /> by \\n\"\n",
    "#     re_br = re.compile(r'<\\s*br\\s*/?>', re.IGNORECASE)\n",
    "#     return re_br.sub(\"\\n\", t)\n",
    "\n",
    "def spec_add_spaces(t:str) -> str:\n",
    "    \"Add spaces between special characters\"\n",
    "    return re.sub(r'([/#?!@,])', r' \\1 ', t)\n",
    "\n",
    "def rm_useless_spaces(t:str) -> str:\n",
    "    \"Remove multiple spaces\"\n",
    "    return re.sub(' {2,}', ' ', t)\n",
    "\n",
    "def replace_rep(t:str) -> str:\n",
    "    \"Replace repetitions at the character level\"\n",
    "    def _replace_rep(m:Collection[str]) -> str:\n",
    "        c,cc = m.groups()\n",
    "        return f' {TK_REP} {len(cc)+1} {c} '\n",
    "    re_rep = re.compile(r'(\\S)(\\1{3,})')\n",
    "    return re_rep.sub(_replace_rep, t)\n",
    "    \n",
    "def replace_wrep(t:str) -> str:\n",
    "    \"Replace word repetitions\"\n",
    "    def _replace_wrep(m:Collection[str]) -> str:\n",
    "        c,cc = m.groups()\n",
    "        return f' {TK_WREP} {len(cc.split())+1} {c} '\n",
    "    re_wrep = re.compile(r'(\\b\\w+\\W+)(\\1{3,})')\n",
    "    return re_wrep.sub(_replace_wrep, t)\n",
    "\n",
    "def deal_caps(t:str) -> str:\n",
    "    \"Replace words in all caps\"\n",
    "    res = []\n",
    "    for s in re.findall(r'\\w+|\\W+', t):\n",
    "        res += ([f' {TK_UP} ',s.lower()] if (s.isupper() and (len(s)>2)) else [s.lower()])\n",
    "    return ''.join(res)\n",
    "\n",
    "def fixup(x:str) -> str:\n",
    "    \"List of replacements from html strings\"\n",
    "    re1 = re.compile(r'  +')\n",
    "    x = x.replace('#39;', \"'\").replace('amp;', '&').replace('#146;', \"'\").replace(\n",
    "        'nbsp;', ' ').replace('#36;', '$').replace('\\\\n', \"\\n\").replace('quot;', \"'\").replace(\n",
    "        '<br />', \"\\n\").replace('\\\\\"', '\"').replace('<unk>',UNK).replace(' @.@ ','.').replace(\n",
    "        ' @-@ ','-').replace('\\\\', ' \\\\ ')\n",
    "    return re1.sub(' ', html.unescape(x))\n",
    "\n",
    "default_rules = [fixup, replace_rep, replace_wrep, deal_caps, spec_add_spaces, \n",
    "                 rm_useless_spaces, sub_nl]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Default Rules:\n",
      " ['fixup', 'replace_rep', 'replace_wrep', 'deal_caps', 'spec_add_spaces', 'rm_useless_spaces', 'sub_nl'] \n",
      "\n",
      "\n",
      "<function fixup at 0x7f62bfc4c9d8>\n",
      "<function replace_rep at 0x7f62bfc4c840>\n",
      "<function replace_wrep at 0x7f62bfc4c8c8>\n",
      "<function deal_caps at 0x7f62bfc4c950>\n",
      "<function spec_add_spaces at 0x7f62bfc4c400>\n",
      "<function rm_useless_spaces at 0x7f62bfc4c7b8>\n",
      "<function sub_nl at 0x7f62bfc4c510>\n"
     ]
    }
   ],
   "source": [
    "## apply the rules\n",
    "raw_text = all_texts_df.loc[:,'text']\n",
    "\n",
    "print(\"Default Rules:\\n\",[x.__name__ for x in default_rules],\"\\n\\n\")\n",
    "\n",
    "for rule in default_rules:\n",
    "    print(rule)\n",
    "    raw_text = raw_text.apply(lambda x: rule(str(x)))\n",
    "    \n",
    "all_texts_df['new_text'] =  raw_text #tokenizer adds xxbos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    niveles de retraso mental: xxnl xxnl — bajo.xx...\n",
       "1    —vamos luke desenfunda tu sable , demuestra tu...\n",
       "2    - ¿te ofrezco algo ? , ¿agua , café , mi coraz...\n",
       "3                no se porqué me hago la cabeza deooos\n",
       "4    quisiera saber que hago durante la siesta de l...\n",
       "Name: new_text, dtype: object"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_texts_df['new_text'].head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(24000, 11)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_texts_df.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Clean up the imbalanced data to balance"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "print('Number of positive samples:', (all_texts_df.is_humor == 1).sum())\n",
    "print('Number of negative samples:',  (all_texts_df.is_humor == 0).sum())\n",
    "print('Total samples:', len(all_texts_df))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "rnd_seed = 20190313\n",
    "np.random.seed(rnd_seed)\n",
    "\n",
    "idx = np.random.permutation(len(all_texts_df))\n",
    "test_cut = int(0.15 * len(idx))\n",
    "valid_cut = int(0.15 * len(idx-test_cut))\n",
    "\n",
    "df_train_all = all_texts_df.iloc[idx[:-(valid_cut+test_cut)],:]\n",
    "df_valid     = all_texts_df.iloc[idx[-(valid_cut+test_cut):-test_cut],:]\n",
    "df_test      = all_texts_df.iloc[idx[-test_cut:],:]"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "print('Test prevalence(n = %d):'%len(df_test),df_test.is_humor.sum()/ len(df_test))\n",
    "print('Valid prevalence(n = %d):'%len(df_valid),df_valid.is_humor.sum()/ len(df_valid))\n",
    "print('Train all prevalence(n = %d):'%len(df_train_all), df_train_all.is_humor.sum()/ len(df_train_all))\n",
    "print('all samples (n = %d)'%len(all_texts_df))\n",
    "assert len(all_texts_df) == (len(df_test)+len(df_valid)+len(df_train_all)),'math didnt work'"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "## oversample the less-frequent occuring cases\n",
    "# split the training data into positive and negative\n",
    "rows_pos = df_train_all.is_humor == 1\n",
    "df_train_pos = df_train_all.loc[rows_pos]\n",
    "df_train_neg = df_train_all.loc[~rows_pos]\n",
    "df_train_pos.shape,df_train_neg.shape"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "df_resample_pos = df_train_pos.sample(n=len(df_train_neg),replace=True,\n",
    "                                      axis=0,random_state=rnd_seed).reset_index(drop=True)\n",
    "df_train = pd.concat([df_resample_pos,df_train_neg],axis=0) #randomized again in DataBunch?\n",
    "print('Train prevalence (n = %d):'%len(df_train), df_train.is_humor.sum()/ len(df_train))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### sp tokenizer from training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_prefix = './all_tweets_es_0509'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "## itos from m.vocab file: just read directly and populate the dictionary\n",
    "itos = [] #{}\n",
    "with open(f'{model_prefix}.vocab','r') as f:\n",
    "    for line_num,line in enumerate(f):\n",
    "#         itos[line_num] = line.split(\"\\t\")[0]\n",
    "        itos.append(line.split(\"\\t\")[0])\n",
    "        \n",
    "class SPTokenizer(BaseTokenizer):\n",
    "    \"Wrapper around a SentncePiece tokenizer to make it a `BaseTokenizer`.\"\n",
    "    def __init__(self, model_prefix:str):\n",
    "        self.tok = spm.SentencePieceProcessor()\n",
    "        self.tok.load(f'{model_prefix}.model')\n",
    "\n",
    "    def tokenizer(self, t:str) -> List[str]:\n",
    "#         return self.tok.EncodeAsIds(t)  ## tokenize + numericalize. We have them broken into two parts,just return tokens\n",
    "        return self.tok.EncodeAsPieces(t)\n",
    "        \n",
    "class CustomTokenizer():\n",
    "    '''Wrapper for SentencePiece toeknizer to fit into Fast.ai V1'''\n",
    "    def __init__(self,tok_func:Callable,model_prefix:str, pre_rules:ListRules=None):\n",
    "        self.tok_func,self.model_prefix = tok_func,model_prefix\n",
    "        self.pre_rules  = ifnone(pre_rules,  defaults.text_pre_rules )\n",
    "        \n",
    "    def __repr__(self) -> str:\n",
    "        res = f'Tokenizer {self.tok_func.__name__} using `{self.model_prefix}` model with the following rules:\\n'\n",
    "        for rule in self.pre_rules: res += f' - {rule.__name__}\\n'\n",
    "        return res        \n",
    "\n",
    "    def process_text(self, t:str,tok:BaseTokenizer) -> List[str]:\n",
    "        \"Processe one text `t` with tokenizer `tok`.\"\n",
    "        for rule in self.pre_rules: t = rule(t)  \n",
    "        toks = tok.tokenizer(t)\n",
    "    \n",
    "        return toks \n",
    "    \n",
    "    def _process_all_1(self,texts:Collection[str]) -> List[List[str]]:\n",
    "        'Process a list of `texts` in one process'\n",
    "        tok = self.tok_func(self.model_prefix)\n",
    "        return [self.process_text(t,tok) for t in texts]\n",
    "                                                                     \n",
    "    def process_all(self, texts:Collection[str]) -> List[List[str]]: \n",
    "        \"Process a list of `texts`.\"                                 \n",
    "        return self._process_all_1(texts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['→', '😫', '🤨', '😲', '🏟', '📖', '🎸', '🏿', '😮', '🎬']"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "itos[-10:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "mycust_tok = CustomTokenizer(SPTokenizer,model_prefix,pre_rules=default_rules)\n",
    "sp_vocab = Vocab(itos)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "data = TextClasDataBunch.from_df(PATH,df_train_all,df_valid,df_test,\n",
    "                               tokenizer=mycust_tok, vocab=sp_vocab,\n",
    "                               text_cols='new_text', label_cols='funniness_average')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>text</th>\n",
       "      <th>target</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁❤ .- ~ * ( ▁? ▁el ▁di vo 00 ▁d 3 ▁tw 1 tter ▁! ▁! ▁! ▁? ▁? ▁~ k 3 ▁di 4 vl 0 zz ▁ki 3 r 3 z ▁k 3 ▁p 4 rt 3 ▁d 3 ▁no 00 ▁n 0 ▁3 nt 13 nd 3 z ▁b 3 the ▁k 0 n ▁ki 3 n ▁t 3 ▁d 3 ▁l 4 ▁g 4</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁dale ▁la ▁vuelta ▁a ▁tu ▁teléfono ▁ xxnl ▁ xxnl ▁ xxunk u xxunk s ▁ xxunk o xxunk d ▁n xxunk ▁ xxunk s ▁ xxunk ▁ xxunk ▁ xxunk s xxunk ▁s xxunk n xxunk ▁ol ▁ou ▁ xxunk s ▁ xxunk ou ▁ xxunk s xxunk ▁s xxunk u xxunk nq ▁s xxunk ou ▁ xxunk u xxunk ▁ xxunk ▁s xxunk ▁o xxunk s</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁# ▁rt ▁ xxup ▁copia ▁y ▁ xxup ▁pega ▁en ▁ xxup ▁google ▁y ▁ve ▁la ▁ xxup ▁magia ▁sq rt ( cos ( x )) * cos ( 200 ▁x ) + sq rt ( abs ( x )) -. 07 ) * ( 4 - x * ▁x ) ^ 0. 01 ▁, ▁sq rt ( 9 - x ^ 2) ▁, ▁- sq rt (</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁qué ▁ricas ▁vacaciones ... xxnl ▁&lt; xxunk / xxunk - xxunk ___ xxnl ▁/ &lt; _ ▁/ ▁ xxrep ▁4 ▁_ ▁/ xxnl ▁ xxnl ▁no ▁ xxup ▁mames ▁ xxup ▁tengo ▁ xxup ▁que ▁ir ▁a ▁la ▁ xxup ▁escuela xxnl ▁ xxnl ▁ xxunk _ xxunk xxnl ▁( ・ xxunk ・ ) xxnl ▁_ | ▁ xxunk / ( ___ ▁ xxnl ▁/ ▁l ▁( __</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁ xxup ▁que ▁ xxup ▁pasó ▁ xxup ▁or ita ▁ xxup ▁boys ▁ xxup ▁los ▁ xxup ▁invitó ▁a ▁ xxup ▁que ▁ xxup ▁este ▁ xxup ▁viernes ▁le ▁ xxup ▁caigan ▁al ▁ xxup ▁suicidio ▁ xxup ▁colectivo ▁ xxup ▁banda ▁se ▁va ▁a ▁ xxup ▁poner ▁ xxup ▁bueno ▁ xxup ▁lleven ▁ xxup ▁sus ▁ xxup ▁propios ▁ xxup ▁u ten sil ios</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁—¿ cómo ▁hace ▁la ▁oveja ▁? ▁ xxnl ▁— b ▁ xxrep ▁4 ▁e ▁. xxnl ▁—¿ y ▁la ▁oveja ▁negra ▁? ▁ xxnl ▁— ▁ xxup ▁yeah ▁ xxup ▁ni gga ▁, ▁ xxup ▁this ▁is ▁ xxup ▁the ▁ xxup ▁real ▁ xxup ▁shit ▁ xxup ▁m ada fa ka ▁ xxup ▁jajaja . xxnl ▁—¿ te ▁drogas ▁? ▁ xxnl ▁— a ▁diario .</td>\n",
       "      <td>1.8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁¡ ▁ xxup ▁desde ▁el ▁ xxup ▁pasado ▁ xxup ▁jueves ▁ xxup ▁pague ▁la ▁ xxup ▁cajita ▁ xxup ▁clap ▁ xxup ▁que ▁ xxup ▁desde ▁ xxup ▁septiembre ▁no ▁ xxup ▁ven dian ▁en ▁mi ▁ xxup ▁barrio ▁ xxup ▁calle ▁ xxup ▁soledad ▁ xxup ▁sector ▁5 ▁ xxup ▁bocas ▁ xxup ▁cab im ás ▁y ▁ xxup ▁sin ▁ xxup ▁llegar ▁!</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁—¿ qué ▁te ▁pasa ▁? ▁ xxnl ▁— nada ▁ xxnl ▁—¿ segura ▁? ▁ xxnl ▁— sí . ▁ xxnl ▁— ok ▁ xxnl ▁— ... xxnl ▁— bueno ▁me ▁voy . xxnl ▁— haz ▁lo ▁que ▁quieras ▁ xxup ▁imbécil ▁, ▁ xxup ▁ignor ame ▁ xxup ▁como ▁ xxup ▁siempre . xxnl ▁—¿ qué ▁te ▁pasa ▁? ▁ xxnl ▁— nada .</td>\n",
       "      <td>1.4615384</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁— hola . xxnl ▁+ hola . ▁ xxnl ▁—¿ qué ▁tal ▁llevas ▁la ▁amnesia ▁? ▁ xxnl ▁+ hola . ▁ xxnl ▁— hola ▁, ▁¿ qué ▁tal ▁llevas ▁la ▁amnesia ▁? ▁ xxnl ▁+ bien ▁, ▁¿ y ▁tú ▁? ▁ xxnl ▁— bien . ▁ xxnl ▁+ ¿ el ▁qué ▁? ▁ xxnl ▁— la ▁amnesia . ▁ xxnl ▁+ hola .</td>\n",
       "      <td>1.1667</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>▁ xxbos ▁ xxup ▁tengo ▁un ▁ xxup ▁gran ▁ xxup ▁anuncio ▁, ▁ xxup ▁una ▁ xxup ▁noticia ▁ xxup ▁que ▁ xxup ▁cambiará ▁la ▁ xxup ▁historia ▁de ▁ xxup ▁twitter ▁, ▁ xxup ▁que ▁ xxup ▁devolver á ▁el ▁ xxup ▁sentido ▁a ▁ xxup ▁nuestra ▁ xxup ▁vida . xxnl ▁ xxnl ▁ xxup ▁pendientes ▁ xxup ▁hoy ▁en ▁la ▁ xxup ▁noche</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data.show_batch(10)  #less xxunk, but still a fair amount..."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Check `xxunk` fraction on humor data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((415,), (16800,))"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pct_xxunk = pd.Series([sum(y==0)/len(y) for y in [x.data for x in data.train_dl.x]])\n",
    "pct_xxunk[pct_xxunk>0].shape, pct_xxunk.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.024053323029366306"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "498/20704."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "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": [
    "pd.Series([sum(y==0)/len(y) for y in [x.data for x in data.train_dl.x]]).plot()\n",
    "_ = plt.title('Pct UNK in each tweet')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### build and train model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "config = awd_lstm_clas_config.copy()\n",
    "config['qrnn'] = True\n",
    "config['n_hid'] = 2304\n",
    "\n",
    "learn = text_classifier_learner(data, AWD_LSTM, drop_mult=0.7,pretrained=False,\n",
    "                               config=config)\n",
    "learn.load_encoder('twitter_es_enc_QRNN_0517_labelsmooth')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "FlattenedLoss of MSELoss()"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn.loss_func"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "LR Finder is complete, type {learner_name}.recorder.plot() to see the graph.\n"
     ]
    }
   ],
   "source": [
    "lr_find(learn)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "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": [
    "learn.recorder.plot(skip_end=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "Total time: 05:24 <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>time</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>0</th>\n",
       "    <th>1.035360</th>\n",
       "    <th>0.679260</th>\n",
       "    <th>00:14</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>1</th>\n",
       "    <th>0.808332</th>\n",
       "    <th>0.659964</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>2</th>\n",
       "    <th>0.762016</th>\n",
       "    <th>0.638135</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>3</th>\n",
       "    <th>0.717055</th>\n",
       "    <th>0.571713</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>4</th>\n",
       "    <th>0.782862</th>\n",
       "    <th>0.595547</th>\n",
       "    <th>00:14</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>5</th>\n",
       "    <th>0.670309</th>\n",
       "    <th>0.626590</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>6</th>\n",
       "    <th>0.628485</th>\n",
       "    <th>0.563793</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>7</th>\n",
       "    <th>0.751551</th>\n",
       "    <th>0.549994</th>\n",
       "    <th>00:14</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>8</th>\n",
       "    <th>0.663983</th>\n",
       "    <th>0.572609</th>\n",
       "    <th>00:14</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>9</th>\n",
       "    <th>0.537754</th>\n",
       "    <th>0.557305</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>10</th>\n",
       "    <th>0.503892</th>\n",
       "    <th>0.586195</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>11</th>\n",
       "    <th>0.525357</th>\n",
       "    <th>0.559791</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>12</th>\n",
       "    <th>0.476124</th>\n",
       "    <th>0.556430</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>13</th>\n",
       "    <th>0.449322</th>\n",
       "    <th>0.577831</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>14</th>\n",
       "    <th>0.466270</th>\n",
       "    <th>0.579707</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>15</th>\n",
       "    <th>0.443091</th>\n",
       "    <th>0.588327</th>\n",
       "    <th>00:14</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>16</th>\n",
       "    <th>0.436182</th>\n",
       "    <th>0.562185</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>17</th>\n",
       "    <th>0.431999</th>\n",
       "    <th>0.569002</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>18</th>\n",
       "    <th>0.410590</th>\n",
       "    <th>0.575806</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "  <tr>\n",
       "    <th>19</th>\n",
       "    <th>0.401655</th>\n",
       "    <th>0.568524</th>\n",
       "    <th>00:15</th>\n",
       "  </tr>\n",
       "</table>\n"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Better model found at epoch 0 with valid_loss value: 0.6792596578598022.\n",
      "Better model found at epoch 1 with valid_loss value: 0.659964382648468.\n",
      "Better model found at epoch 2 with valid_loss value: 0.6381351947784424.\n",
      "Better model found at epoch 3 with valid_loss value: 0.5717129707336426.\n",
      "Better model found at epoch 6 with valid_loss value: 0.563792884349823.\n",
      "Better model found at epoch 7 with valid_loss value: 0.5499940514564514.\n"
     ]
    }
   ],
   "source": [
    "learn.unfreeze()\n",
    "## 0.7 drop\n",
    "learn.fit_one_cycle(20, slice(1e-2/(2.6**4),1e-2), moms=(0.8,0.6), pct_start=0.25, div_factor=30.,\n",
    "             callbacks=[SaveModelCallback(learn,every='improvement',mode='min',\n",
    "                                          monitor='valid_loss',name='best_mse_model_Q')])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "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"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.recorder.plot_losses()\n",
    "learn.recorder.plot_lr(show_moms=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.save('rnn_trained_unfreezeAll_funninessScore')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Generate test file for submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "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>id</th>\n",
       "      <th>text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>tweet1</td>\n",
       "      <td>Historia #36.\\n*Entra corriendo y gritando al ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>tweet2</td>\n",
       "      <td>estoy tomando helado en el centro y pedi tiram...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>tweet3</td>\n",
       "      <td>— ¿De 15 o 20 centímetros?\\n\\n— Ven y averígua...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>tweet4</td>\n",
       "      <td>Que fea soy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>tweet5</td>\n",
       "      <td>#OMG, VAMOS EN RETROCESO RESULTE CON 40 SEGUID...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       id                                               text\n",
       "0  tweet1  Historia #36.\\n*Entra corriendo y gritando al ...\n",
       "1  tweet2  estoy tomando helado en el centro y pedi tiram...\n",
       "2  tweet3  — ¿De 15 o 20 centímetros?\\n\\n— Ven y averígua...\n",
       "3  tweet4                                        Que fea soy\n",
       "4  tweet5  #OMG, VAMOS EN RETROCESO RESULTE CON 40 SEGUID..."
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df = pd.read_csv('./data/haha_2019_test.csv')\n",
    "test_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(6000, 2)"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Default Rules:\n",
      " ['fixup', 'replace_rep', 'replace_wrep', 'deal_caps', 'spec_add_spaces', 'rm_useless_spaces', 'sub_nl'] \n",
      "\n",
      "\n"
     ]
    }
   ],
   "source": [
    "## apply the rules\n",
    "raw_text = test_df.loc[:,'text']\n",
    "\n",
    "print(\"Default Rules:\\n\",[x.__name__ for x in default_rules],\"\\n\\n\")\n",
    "for rule in default_rules:\n",
    "    raw_text = raw_text.apply(lambda x: rule(str(x)))    \n",
    "test_df['new_text'] =  raw_text #tokenizer adds xxbos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "data_tst = TextClasDataBunch.from_df(PATH, df_train_all, df_valid, test_df,\n",
    "                               tokenizer=mycust_tok, vocab=sp_vocab,\n",
    "                               text_cols='new_text', label_cols='funniness_average')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn.data = data_tst"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_t,y_t,losses_t = learn.get_preds(DatasetType.Valid,with_loss=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f62b5815e80>"
      ]
     },
     "execution_count": 50,
     "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": [
    "pd.Series(to_np(losses_t)).hist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_tst,y_tst,losses_tst = learn.get_preds(DatasetType.Test,with_loss=True, ordered=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text ▁ xxbos ▁¿ le ▁pusiste ▁a ▁tu ▁hijo ▁bec kh am ▁hace ▁16 ▁años ▁y ▁ahora ▁odia ▁el ▁fútbol ▁, ▁a ▁la ▁humanidad ▁y ▁a ▁ti ... ▁? ▁. ¡ normal ▁, ▁ya ▁hay ▁que ▁ser ▁gilipollas ▁! ▁."
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data_tst.test_ds.get(-1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "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>id</th>\n",
       "      <th>text</th>\n",
       "      <th>new_text</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>tweet1</td>\n",
       "      <td>Historia #36.\\n*Entra corriendo y gritando al ...</td>\n",
       "      <td>historia # 36.xxnl *entra corriendo y gritando...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>tweet2</td>\n",
       "      <td>estoy tomando helado en el centro y pedi tiram...</td>\n",
       "      <td>estoy tomando helado en el centro y pedi tiram...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>tweet3</td>\n",
       "      <td>— ¿De 15 o 20 centímetros?\\n\\n— Ven y averígua...</td>\n",
       "      <td>— ¿de 15 o 20 centímetros ? xxnl xxnl — ven y ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>tweet4</td>\n",
       "      <td>Que fea soy</td>\n",
       "      <td>que fea soy</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>tweet5</td>\n",
       "      <td>#OMG, VAMOS EN RETROCESO RESULTE CON 40 SEGUID...</td>\n",
       "      <td># xxup omg , xxup vamos en xxup retroceso xxu...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       id                                               text  \\\n",
       "0  tweet1  Historia #36.\\n*Entra corriendo y gritando al ...   \n",
       "1  tweet2  estoy tomando helado en el centro y pedi tiram...   \n",
       "2  tweet3  — ¿De 15 o 20 centímetros?\\n\\n— Ven y averígua...   \n",
       "3  tweet4                                        Que fea soy   \n",
       "4  tweet5  #OMG, VAMOS EN RETROCESO RESULTE CON 40 SEGUID...   \n",
       "\n",
       "                                            new_text  \n",
       "0  historia # 36.xxnl *entra corriendo y gritando...  \n",
       "1  estoy tomando helado en el centro y pedi tiram...  \n",
       "2  — ¿de 15 o 20 centímetros ? xxnl xxnl — ven y ...  \n",
       "3                                        que fea soy  \n",
       "4   # xxup omg , xxup vamos en xxup retroceso xxu...  "
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "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>id</th>\n",
       "      <th>is_humor</th>\n",
       "      <th>funniness_average</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>tweet1</td>\n",
       "      <td>1</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>tweet2</td>\n",
       "      <td>0</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>tweet3</td>\n",
       "      <td>1</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>tweet4</td>\n",
       "      <td>0</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>tweet5</td>\n",
       "      <td>0</td>\n",
       "      <td>3.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       id  is_humor  funniness_average\n",
       "0  tweet1         1                3.0\n",
       "1  tweet2         0                3.0\n",
       "2  tweet3         1                3.0\n",
       "3  tweet4         0                3.0\n",
       "4  tweet5         0                3.0"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out_df = pd.read_csv('sub0520_1.csv')\n",
    "out_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [],
   "source": [
    "out_df['funniness_average'] = to_np(preds_tst)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "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>id</th>\n",
       "      <th>is_humor</th>\n",
       "      <th>funniness_average</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>tweet1</td>\n",
       "      <td>1</td>\n",
       "      <td>1.646994</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>tweet2</td>\n",
       "      <td>0</td>\n",
       "      <td>0.249239</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>tweet3</td>\n",
       "      <td>1</td>\n",
       "      <td>2.337155</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>tweet4</td>\n",
       "      <td>0</td>\n",
       "      <td>0.323720</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>tweet5</td>\n",
       "      <td>0</td>\n",
       "      <td>-0.277000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       id  is_humor  funniness_average\n",
       "0  tweet1         1           1.646994\n",
       "1  tweet2         0           0.249239\n",
       "2  tweet3         1           2.337155\n",
       "3  tweet4         0           0.323720\n",
       "4  tweet5         0          -0.277000"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [],
   "source": [
    "out_df.to_csv('sub0521_2_mse.csv',header=True,index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "id,is_humor,funniness_average\r\n",
      "tweet1,1,1.646994\r\n",
      "tweet2,0,0.24923912\r\n",
      "tweet3,1,2.337155\r\n",
      "tweet4,0,0.3237204\r\n",
      "tweet5,0,-0.27699962\r\n",
      "tweet6,0,0.6858403\r\n",
      "tweet7,0,0.6659835\r\n",
      "tweet8,0,0.090969265\r\n",
      "tweet9,0,1.0877298\r\n"
     ]
    }
   ],
   "source": [
    "!head sub0521_2_mse.csv"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "  adding: sub0521_2_mse.csv (deflated 66%)\r\n"
     ]
    }
   ],
   "source": [
    "!zip  sub0521_2_mse.zip sub0521_2_mse.csv "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "!cp sub0521_2_mse.zip /mnt/awcnas4_AWX/tmp/"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3.7 fasta.ai1 DEV",
   "language": "python",
   "name": "fastai1_dev"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  },
  "varInspector": {
   "cols": {
    "lenName": 16,
    "lenType": 16,
    "lenVar": 40
   },
   "kernels_config": {
    "python": {
     "delete_cmd_postfix": "",
     "delete_cmd_prefix": "del ",
     "library": "var_list.py",
     "varRefreshCmd": "print(var_dic_list())"
    },
    "r": {
     "delete_cmd_postfix": ") ",
     "delete_cmd_prefix": "rm(",
     "library": "var_list.r",
     "varRefreshCmd": "cat(var_dic_list()) "
    }
   },
   "types_to_exclude": [
    "module",
    "function",
    "builtin_function_or_method",
    "instance",
    "_Feature"
   ],
   "window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}