qnb-starter-lstm-gru.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "import time\n",
    "import numpy as np # \n",
    "import pandas as pd \n",
    "from tqdm import tqdm, tqdm_pandas, tqdm_notebook\n",
    "from sklearn.metrics import f1_score, roc_auc_score\n",
    "import gc\n",
    "from keras.preprocessing.text import Tokenizer\n",
    "from keras.preprocessing.sequence import pad_sequences\n",
    "from keras.layers import Dense, Input, CuDNNLSTM, Embedding, Dropout, Activation, CuDNNGRU, Conv1D, BatchNormalization\n",
    "from keras.layers import Bidirectional, GlobalMaxPool1D, GlobalMaxPooling1D, GlobalAveragePooling1D\n",
    "from keras.layers import Input, Conv2D, MaxPool2D, concatenate\n",
    "from keras.layers import Reshape, Flatten, Concatenate, Dropout, SpatialDropout1D\n",
    "from keras.models import Model\n",
    "from keras import backend as K\n",
    "from keras.engine.topology import Layer\n",
    "from keras import initializers, regularizers, constraints, optimizers, layers\n",
    "from keras.callbacks import *\n",
    "import keras\n",
    "import pickle\n",
    "from sklearn.utils import compute_class_weight"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['/job:localhost/replica:0/task:0/device:GPU:0']\n"
     ]
    }
   ],
   "source": [
    "import sys\n",
    "print(K.tensorflow_backend._get_available_gpus())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [],
   "source": [
    "path = \"../input/qnb_finansbank_yarisma/QNB_Finansbank_Yapay_Öğrenme_Yarışması/\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_cust = pd.read_csv(path+\"Koc_Yaz_Okulu_Data_Train_Cust.txt\", sep=\";\")\n",
    "train_ag = pd.read_csv(path+\"Koc_Yaz_Okulu_Data_Train_Agent.txt\", sep=\";\")\n",
    "y_train = pd.read_csv(path+\"Koc_Yaz_Okulu_Data_Train_Target.txt\", sep=\";\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_cust = pd.read_csv(path+\"Koc_Yaz_Okulu_Data_Test_Cust.txt\", sep=\";\")\n",
    "test_ag = pd.read_csv(path+\"Koc_Yaz_Okulu_Data_Test_Agent.txt\", sep=\";\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((12721, 2), (12721, 2))"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_cust.shape, test_ag.shape"
   ]
  },
  {
   "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>AGENT_TXT</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>29140726</td>\n",
       "      <td>X7092177 X7352233 X8382776 X5850527 X9875515 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>29472320</td>\n",
       "      <td>X3544666 X1643318 X5329880 X4336932 X9875515 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>29012836</td>\n",
       "      <td>X7092177 X5266558 X1638055 X3274335 X2029222 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>29161698</td>\n",
       "      <td>X5168898 X2958188 X5850527 X3798496 X8950102 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>28599552</td>\n",
       "      <td>X7358694 X1638055 X6932443 X6672085 X3262622 X...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         ID                                          AGENT_TXT\n",
       "0  29140726  X7092177 X7352233 X8382776 X5850527 X9875515 X...\n",
       "1  29472320  X3544666 X1643318 X5329880 X4336932 X9875515 X...\n",
       "2  29012836  X7092177 X5266558 X1638055 X3274335 X2029222 X...\n",
       "3  29161698  X5168898 X2958188 X5850527 X3798496 X8950102 X...\n",
       "4  28599552  X7358694 X1638055 X6932443 X6672085 X3262622 X..."
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_ag.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>LABEL_1</th>\n",
       "      <th>LABEL_2</th>\n",
       "      <th>LABEL_3</th>\n",
       "      <th>LABEL_4</th>\n",
       "      <th>LABEL_5</th>\n",
       "      <th>LABEL_6</th>\n",
       "      <th>LABEL_7</th>\n",
       "      <th>LABEL_8</th>\n",
       "      <th>LABEL_9</th>\n",
       "      <th>LABEL_10</th>\n",
       "      <th>LABEL_11</th>\n",
       "      <th>LABEL_12</th>\n",
       "      <th>LABEL_13</th>\n",
       "      <th>LABEL_14</th>\n",
       "      <th>LABEL_15</th>\n",
       "      <th>LABEL_16</th>\n",
       "      <th>LABEL_17</th>\n",
       "      <th>LABEL_18</th>\n",
       "      <th>LABEL_19</th>\n",
       "      <th>LABEL_20</th>\n",
       "      <th>LABEL_21</th>\n",
       "      <th>LABEL_22</th>\n",
       "      <th>LABEL_23</th>\n",
       "      <th>LABEL_24</th>\n",
       "      <th>LABEL_25</th>\n",
       "      <th>LABEL_26</th>\n",
       "      <th>LABEL_27</th>\n",
       "      <th>LABEL_28</th>\n",
       "      <th>LABEL_29</th>\n",
       "      <th>LABEL_30</th>\n",
       "      <th>LABEL_31</th>\n",
       "      <th>LABEL_32</th>\n",
       "      <th>LABEL_33</th>\n",
       "      <th>LABEL_34</th>\n",
       "      <th>LABEL_35</th>\n",
       "      <th>LABEL_36</th>\n",
       "      <th>LABEL_37</th>\n",
       "      <th>LABEL_38</th>\n",
       "      <th>LABEL_39</th>\n",
       "      <th>LABEL_40</th>\n",
       "      <th>LABEL_41</th>\n",
       "      <th>LABEL_42</th>\n",
       "      <th>LABEL_43</th>\n",
       "      <th>LABEL_44</th>\n",
       "      <th>LABEL_45</th>\n",
       "      <th>LABEL_46</th>\n",
       "      <th>LABEL_47</th>\n",
       "      <th>LABEL_48</th>\n",
       "      <th>LABEL_49</th>\n",
       "      <th>LABEL_50</th>\n",
       "      <th>LABEL_51</th>\n",
       "      <th>LABEL_52</th>\n",
       "      <th>LABEL_53</th>\n",
       "      <th>LABEL_54</th>\n",
       "      <th>LABEL_55</th>\n",
       "      <th>LABEL_56</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>28902722</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>30265364</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>28732508</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>29109715</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>30198070</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         ID  LABEL_1  LABEL_2    ...     LABEL_54  LABEL_55  LABEL_56\n",
       "0  28902722        0        1    ...            0         0         0\n",
       "1  30265364        0        0    ...            0         0         0\n",
       "2  28732508        0        0    ...            0         0         0\n",
       "3  29109715        0        0    ...            0         0         0\n",
       "4  30198070        0        0    ...            0         0         0\n",
       "\n",
       "[5 rows x 57 columns]"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_df = train_cust.merge(train_ag, on=\"ID\", how=\"left\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_df = test_cust.merge(test_ag, on=\"ID\", how=\"left\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((40000, 3), (12721, 3))"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df.shape, test_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "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>CUST_TXT</th>\n",
       "      <th>AGENT_TXT</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>29140726</td>\n",
       "      <td>X7092177 X7352233 X6482199 X3262622 X7596215 X...</td>\n",
       "      <td>X7092177 X7352233 X8382776 X5850527 X9875515 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>29472320</td>\n",
       "      <td>X1317264 X3511202 X3074897 X8657971 X9504814 X...</td>\n",
       "      <td>X3544666 X1643318 X5329880 X4336932 X9875515 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>29012836</td>\n",
       "      <td>X7092177 X3751173 X2408482 X9784649 X3074897 X...</td>\n",
       "      <td>X7092177 X5266558 X1638055 X3274335 X2029222 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>29161698</td>\n",
       "      <td>X7092177 X5266558 X9587719 X7330829 X9093733 X...</td>\n",
       "      <td>X5168898 X2958188 X5850527 X3798496 X8950102 X...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>28599552</td>\n",
       "      <td>X1236709 X3262622 X3422547 X1495491 X4692604 X...</td>\n",
       "      <td>X7358694 X1638055 X6932443 X6672085 X3262622 X...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         ID                        ...                                                                  AGENT_TXT\n",
       "0  29140726                        ...                          X7092177 X7352233 X8382776 X5850527 X9875515 X...\n",
       "1  29472320                        ...                          X3544666 X1643318 X5329880 X4336932 X9875515 X...\n",
       "2  29012836                        ...                          X7092177 X5266558 X1638055 X3274335 X2029222 X...\n",
       "3  29161698                        ...                          X5168898 X2958188 X5850527 X3798496 X8950102 X...\n",
       "4  28599552                        ...                          X7358694 X1638055 X6932443 X6672085 X3262622 X...\n",
       "\n",
       "[5 rows x 3 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "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>CUST_TXT</th>\n",
       "      <th>AGENT_TXT</th>\n",
       "      <th>LABEL_1</th>\n",
       "      <th>LABEL_2</th>\n",
       "      <th>LABEL_3</th>\n",
       "      <th>LABEL_4</th>\n",
       "      <th>LABEL_5</th>\n",
       "      <th>LABEL_6</th>\n",
       "      <th>LABEL_7</th>\n",
       "      <th>LABEL_8</th>\n",
       "      <th>LABEL_9</th>\n",
       "      <th>LABEL_10</th>\n",
       "      <th>LABEL_11</th>\n",
       "      <th>LABEL_12</th>\n",
       "      <th>LABEL_13</th>\n",
       "      <th>LABEL_14</th>\n",
       "      <th>LABEL_15</th>\n",
       "      <th>LABEL_16</th>\n",
       "      <th>LABEL_17</th>\n",
       "      <th>LABEL_18</th>\n",
       "      <th>LABEL_19</th>\n",
       "      <th>LABEL_20</th>\n",
       "      <th>LABEL_21</th>\n",
       "      <th>LABEL_22</th>\n",
       "      <th>LABEL_23</th>\n",
       "      <th>LABEL_24</th>\n",
       "      <th>LABEL_25</th>\n",
       "      <th>LABEL_26</th>\n",
       "      <th>LABEL_27</th>\n",
       "      <th>LABEL_28</th>\n",
       "      <th>LABEL_29</th>\n",
       "      <th>LABEL_30</th>\n",
       "      <th>LABEL_31</th>\n",
       "      <th>LABEL_32</th>\n",
       "      <th>LABEL_33</th>\n",
       "      <th>LABEL_34</th>\n",
       "      <th>LABEL_35</th>\n",
       "      <th>LABEL_36</th>\n",
       "      <th>LABEL_37</th>\n",
       "      <th>LABEL_38</th>\n",
       "      <th>LABEL_39</th>\n",
       "      <th>LABEL_40</th>\n",
       "      <th>LABEL_41</th>\n",
       "      <th>LABEL_42</th>\n",
       "      <th>LABEL_43</th>\n",
       "      <th>LABEL_44</th>\n",
       "      <th>LABEL_45</th>\n",
       "      <th>LABEL_46</th>\n",
       "      <th>LABEL_47</th>\n",
       "      <th>LABEL_48</th>\n",
       "      <th>LABEL_49</th>\n",
       "      <th>LABEL_50</th>\n",
       "      <th>LABEL_51</th>\n",
       "      <th>LABEL_52</th>\n",
       "      <th>LABEL_53</th>\n",
       "      <th>LABEL_54</th>\n",
       "      <th>LABEL_55</th>\n",
       "      <th>LABEL_56</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>29140726</td>\n",
       "      <td>X7092177 X7352233 X6482199 X3262622 X7596215 X...</td>\n",
       "      <td>X7092177 X7352233 X8382776 X5850527 X9875515 X...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>29472320</td>\n",
       "      <td>X1317264 X3511202 X3074897 X8657971 X9504814 X...</td>\n",
       "      <td>X3544666 X1643318 X5329880 X4336932 X9875515 X...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>29012836</td>\n",
       "      <td>X7092177 X3751173 X2408482 X9784649 X3074897 X...</td>\n",
       "      <td>X7092177 X5266558 X1638055 X3274335 X2029222 X...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>29161698</td>\n",
       "      <td>X7092177 X5266558 X9587719 X7330829 X9093733 X...</td>\n",
       "      <td>X5168898 X2958188 X5850527 X3798496 X8950102 X...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>28599552</td>\n",
       "      <td>X1236709 X3262622 X3422547 X1495491 X4692604 X...</td>\n",
       "      <td>X7358694 X1638055 X6932443 X6672085 X3262622 X...</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         ID   ...    LABEL_56\n",
       "0  29140726   ...           0\n",
       "1  29472320   ...           0\n",
       "2  29012836   ...           0\n",
       "3  29161698   ...           0\n",
       "4  28599552   ...           0\n",
       "\n",
       "[5 rows x 59 columns]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_df.merge(y_train, on=\"ID\", how=\"left\").head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "stats_df = pd.DataFrame(index=train_df.index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "stats_df['total_length'] = train_df['CUST_TXT'].apply(len)\n",
    "stats_df['num_words'] = train_df['CUST_TXT'].apply(lambda comment: len(comment.split()))\n",
    "stats_df['num_unique_words'] = train_df['CUST_TXT'].apply(lambda comment: len(set(w for w in comment.split())))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "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>total_length</th>\n",
       "      <th>num_words</th>\n",
       "      <th>num_unique_words</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>332</td>\n",
       "      <td>37</td>\n",
       "      <td>29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1043</td>\n",
       "      <td>116</td>\n",
       "      <td>46</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>1025</td>\n",
       "      <td>114</td>\n",
       "      <td>66</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1781</td>\n",
       "      <td>198</td>\n",
       "      <td>93</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>692</td>\n",
       "      <td>77</td>\n",
       "      <td>40</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   total_length  num_words  num_unique_words\n",
       "0           332         37                29\n",
       "1          1043        116                46\n",
       "2          1025        114                66\n",
       "3          1781        198                93\n",
       "4           692         77                40"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stats_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "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>count</th>\n",
       "      <th>mean</th>\n",
       "      <th>std</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>total_length</th>\n",
       "      <td>40000.0</td>\n",
       "      <td>752.2937</td>\n",
       "      <td>402.057276</td>\n",
       "      <td>98.0</td>\n",
       "      <td>440.0</td>\n",
       "      <td>683.0</td>\n",
       "      <td>989.0</td>\n",
       "      <td>3059.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>num_words</th>\n",
       "      <td>40000.0</td>\n",
       "      <td>83.6993</td>\n",
       "      <td>44.673031</td>\n",
       "      <td>11.0</td>\n",
       "      <td>49.0</td>\n",
       "      <td>76.0</td>\n",
       "      <td>110.0</td>\n",
       "      <td>340.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>num_unique_words</th>\n",
       "      <td>40000.0</td>\n",
       "      <td>49.9131</td>\n",
       "      <td>19.468796</td>\n",
       "      <td>4.0</td>\n",
       "      <td>35.0</td>\n",
       "      <td>48.0</td>\n",
       "      <td>63.0</td>\n",
       "      <td>138.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                    count      mean         std   ...      50%    75%     max\n",
       "total_length      40000.0  752.2937  402.057276   ...    683.0  989.0  3059.0\n",
       "num_words         40000.0   83.6993   44.673031   ...     76.0  110.0   340.0\n",
       "num_unique_words  40000.0   49.9131   19.468796   ...     48.0   63.0   138.0\n",
       "\n",
       "[3 rows x 8 columns]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "stats_df.describe().T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "test_id = test_df[\"ID\"]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_features = 2000\n",
    "max_len = 50\n",
    "embed_size = 40"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer = Tokenizer(num_words=max_features)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_agent = train_df[\"AGENT_TXT\"].fillna(\"_##_\").values\n",
    "test_agent = test_df[\"AGENT_TXT\"].fillna(\"_##_\").values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_cust = train_df[\"CUST_TXT\"].fillna(\"_##_\").values\n",
    "test_cust = test_df[\"CUST_TXT\"].fillna(\"_##_\").values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer.fit_on_texts(list(train_agent) + list(test_agent))\n",
    "train_agent = tokenizer.texts_to_sequences(train_agent)\n",
    "test_agent = tokenizer.texts_to_sequences(test_agent)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "tokenizer.fit_on_texts(list(train_cust) + list(test_cust))\n",
    "train_cust = tokenizer.texts_to_sequences(train_cust)\n",
    "test_cust = tokenizer.texts_to_sequences(test_cust)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_agent = pad_sequences(train_agent, maxlen=max_len)\n",
    "test_agent = pad_sequences(test_agent, maxlen=max_len)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_cust = pad_sequences(train_cust, maxlen=max_len)\n",
    "test_cust = pad_sequences(test_cust, maxlen=max_len)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = np.hstack([train_agent, train_cust])\n",
    "df_test = np.hstack([test_agent, test_cust])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((40000, 100), (12721, 100))"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train.shape, df_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[   0,    0,    0,    0,    0,   21,  154,  214,    7,   11,   32,\n",
       "           3,   26,   38,   30,   69,   39,  146,   12,   55,  132,   78,\n",
       "         700,    7,   11,   55,  132,   78,  146,    1,   70,   87,  177,\n",
       "         101,  597,    1,    9,   87,   57,   55,  132,   26,   32,    3,\n",
       "          78,   59,    3,   21,   15,   53,    0,    0,    0,    0,    0,\n",
       "           0,    0,    0,    0,    0,    0,    0,    0,   16,  110,  958,\n",
       "          66,   19,    4,  206,    4,  206,  129,   24,    5,  383,    8,\n",
       "          55,  114,    5,    8,  247,  270,   53,  196,  339, 1108,  145,\n",
       "         196,  179,    1,   38,   11,  196,   24,   69,  138,   86,  207,\n",
       "          11]], dtype=int32)"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train[:1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "_kg_hide-input": true
   },
   "outputs": [],
   "source": [
    "class Attention(Layer):\n",
    "    def __init__(self, step_dim,\n",
    "                 W_regularizer=None, b_regularizer=None,\n",
    "                 W_constraint=None, b_constraint=None,\n",
    "                 bias=True, **kwargs):\n",
    "        self.supports_masking = True\n",
    "        self.init = initializers.get('glorot_normal')\n",
    "\n",
    "        self.W_regularizer = regularizers.get(W_regularizer)\n",
    "        self.b_regularizer = regularizers.get(b_regularizer)\n",
    "\n",
    "        self.W_constraint = constraints.get(W_constraint)\n",
    "        self.b_constraint = constraints.get(b_constraint)\n",
    "\n",
    "        self.bias = bias\n",
    "        self.step_dim = step_dim\n",
    "        self.features_dim = 0\n",
    "        super(Attention, self).__init__(**kwargs)\n",
    "\n",
    "    def build(self, input_shape):\n",
    "        assert len(input_shape) == 3\n",
    "\n",
    "        self.W = self.add_weight((input_shape[-1],),\n",
    "                                 initializer=self.init,\n",
    "                                 name='{}_W'.format(self.name),\n",
    "                                 regularizer=self.W_regularizer,\n",
    "                                 constraint=self.W_constraint)\n",
    "        self.features_dim = input_shape[-1]\n",
    "\n",
    "        if self.bias:\n",
    "            self.b = self.add_weight((input_shape[1],),\n",
    "                                     initializer='zero',\n",
    "                                     name='{}_b'.format(self.name),\n",
    "                                     regularizer=self.b_regularizer,\n",
    "                                     constraint=self.b_constraint)\n",
    "        else:\n",
    "            self.b = None\n",
    "\n",
    "        self.built = True\n",
    "\n",
    "    def compute_mask(self, input, input_mask=None):\n",
    "        return None\n",
    "\n",
    "    def call(self, x, mask=None):\n",
    "        features_dim = self.features_dim\n",
    "        step_dim = self.step_dim\n",
    "\n",
    "        eij = K.reshape(K.dot(K.reshape(x, (-1, features_dim)),\n",
    "                        K.reshape(self.W, (features_dim, 1))), (-1, step_dim))\n",
    "\n",
    "        if self.bias:\n",
    "            eij += self.b\n",
    "\n",
    "        eij = K.tanh(eij)\n",
    "\n",
    "        a = K.exp(eij)\n",
    "\n",
    "        if mask is not None:\n",
    "            a *= K.cast(mask, K.floatx())\n",
    "\n",
    "        a /= K.cast(K.sum(a, axis=1, keepdims=True) + K.epsilon(), K.floatx())\n",
    "\n",
    "        a = K.expand_dims(a)\n",
    "        weighted_input = x * a\n",
    "        return K.sum(weighted_input, axis=1)\n",
    "\n",
    "    def compute_output_shape(self, input_shape):\n",
    "        return input_shape[0],  self.features_dim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "labels = y_train.columns.tolist()[1:]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [],
   "source": [
    "bs = 256"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0"
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tue Jun 18 08:06:49 2019       \r\n",
      "+-----------------------------------------------------------------------------+\r\n",
      "| NVIDIA-SMI 418.67       Driver Version: 418.67       CUDA Version: 10.1     |\r\n",
      "|-------------------------------+----------------------+----------------------+\r\n",
      "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\r\n",
      "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\r\n",
      "|===============================+======================+======================|\r\n",
      "|   0  Tesla P100-PCIE...  On   | 00000000:00:04.0 Off |                    0 |\r\n",
      "| N/A   38C    P0    32W / 250W |    337MiB / 16280MiB |      0%      Default |\r\n",
      "+-------------------------------+----------------------+----------------------+\r\n",
      "                                                                               \r\n",
      "+-----------------------------------------------------------------------------+\r\n",
      "| Processes:                                                       GPU Memory |\r\n",
      "|  GPU       PID   Type   Process name                             Usage      |\r\n",
      "|=============================================================================|\r\n",
      "+-----------------------------------------------------------------------------+\r\n"
     ]
    }
   ],
   "source": [
    "!nvidia-smi"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_model_all(max_len=df_train.shape[1]):\n",
    "    inp = Input(shape=(max_len,))\n",
    "    x = Embedding(max_features, embed_size, trainable=False)(inp)\n",
    "    x = SpatialDropout1D(0.2)(x)\n",
    "    x = Bidirectional(CuDNNLSTM(64, return_sequences=True ))(x)\n",
    "    y = Bidirectional(CuDNNGRU(32, return_sequences=True))(x)\n",
    "    #x = Bidirectional(CuDNNGRU(96,return_sequences=True))(x)\n",
    "    #x = Bidirectional(CuDNNGRU(64,return_sequences=True))(x) \n",
    "    \n",
    "    \n",
    "    atten_1 = Attention(max_len)(x)\n",
    "    atten_2 = Attention(max_len)(y)\n",
    "    avg_pool = GlobalAveragePooling1D()(y)\n",
    "    max_pool = GlobalMaxPooling1D()(y)\n",
    "    \n",
    "    conc = concatenate([atten_1, atten_2, avg_pool, max_pool])\n",
    "    conc = BatchNormalization()(conc)\n",
    "    conc = Dropout(0.1)(conc) \n",
    "    outp = Dense(1, activation=\"sigmoid\")(conc)\n",
    "    adam = optimizers.Adam(lr=0.005, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.01, amsgrad=False)\n",
    "    model = Model(inputs=inp, outputs=outp)\n",
    "    model.compile(loss='binary_crossentropy', optimizer=adam, metrics=['accuracy'])\n",
    "    #model.summary()\n",
    "    return model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /opt/conda/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n",
      "WARNING:tensorflow:From /opt/conda/lib/python3.6/site-packages/keras/backend/tensorflow_backend.py:3445: calling dropout (from tensorflow.python.ops.nn_ops) with keep_prob is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Please use `rate` instead of `keep_prob`. Rate should be set to `rate = 1 - keep_prob`.\n"
     ]
    }
   ],
   "source": [
    "model_all = get_model_all()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "input_1 (InputLayer)            (None, 100)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "embedding_1 (Embedding)         (None, 100, 40)      80000       input_1[0][0]                    \n",
      "__________________________________________________________________________________________________\n",
      "spatial_dropout1d_1 (SpatialDro (None, 100, 40)      0           embedding_1[0][0]                \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_1 (Bidirectional) (None, 100, 128)     54272       spatial_dropout1d_1[0][0]        \n",
      "__________________________________________________________________________________________________\n",
      "bidirectional_2 (Bidirectional) (None, 100, 64)      31104       bidirectional_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "attention_1 (Attention)         (None, 128)          228         bidirectional_1[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "attention_2 (Attention)         (None, 64)           164         bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "global_average_pooling1d_1 (Glo (None, 64)           0           bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "global_max_pooling1d_1 (GlobalM (None, 64)           0           bidirectional_2[0][0]            \n",
      "__________________________________________________________________________________________________\n",
      "concatenate_1 (Concatenate)     (None, 320)          0           attention_1[0][0]                \n",
      "                                                                 attention_2[0][0]                \n",
      "                                                                 global_average_pooling1d_1[0][0] \n",
      "                                                                 global_max_pooling1d_1[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "batch_normalization_1 (BatchNor (None, 320)          1280        concatenate_1[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 320)          0           batch_normalization_1[0][0]      \n",
      "__________________________________________________________________________________________________\n",
      "dense_1 (Dense)                 (None, 1)            321         dropout_1[0][0]                  \n",
      "==================================================================================================\n",
      "Total params: 167,369\n",
      "Trainable params: 86,729\n",
      "Non-trainable params: 80,640\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model_all.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.utils import plot_model\n",
    "plot_model(get_model_all(), to_file='model.png',show_shapes=True)\n",
    "from IPython.display import Image\n",
    "Image(filename='model.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission_all = pd.DataFrame(index=test_df[\"ID\"], columns=labels)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Label: LABEL_1\n",
      "WARNING:tensorflow:From /opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Use tf.cast instead.\n",
      "WARNING:tensorflow:From /opt/conda/lib/python3.6/site-packages/tensorflow/python/ops/math_grad.py:102: div (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Deprecated in favor of operator or tf.math.divide.\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 10s - loss: 0.9240 - acc: 0.9065 - val_loss: 2.2542 - val_acc: 0.9123\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9174 - acc: 0.9142 - val_loss: 0.9530 - val_acc: 0.9123\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.9163 - acc: 0.9142 - val_loss: 0.9327 - val_acc: 0.9123\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.9153 - acc: 0.9142 - val_loss: 1.0693 - val_acc: 0.0910\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.9136 - acc: 0.9141 - val_loss: 1.1679 - val_acc: 0.0877\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.9127 - acc: 0.9141 - val_loss: 0.9981 - val_acc: 0.9123\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9124 - acc: 0.9140 - val_loss: 0.9396 - val_acc: 0.9123\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.9107 - acc: 0.9137 - val_loss: 0.9549 - val_acc: 0.9123\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.9088 - acc: 0.9138 - val_loss: 0.9558 - val_acc: 0.9123\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.9077 - acc: 0.9127 - val_loss: 0.9466 - val_acc: 0.8923\n",
      "Label: LABEL_2\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 6s - loss: 0.8122 - acc: 0.6817 - val_loss: 0.9497 - val_acc: 0.3105\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.8093 - acc: 0.6891 - val_loss: 1.0173 - val_acc: 0.3105\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.8087 - acc: 0.6891 - val_loss: 0.8118 - val_acc: 0.6890\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.8082 - acc: 0.6888 - val_loss: 0.8103 - val_acc: 0.6895\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.8078 - acc: 0.6891 - val_loss: 0.8473 - val_acc: 0.6895\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.8074 - acc: 0.6889 - val_loss: 0.8308 - val_acc: 0.6895\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.8079 - acc: 0.6882 - val_loss: 0.9850 - val_acc: 0.3105\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.8074 - acc: 0.6891 - val_loss: 0.8251 - val_acc: 0.5057\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.8064 - acc: 0.6880 - val_loss: 0.9422 - val_acc: 0.6895\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.8065 - acc: 0.6871 - val_loss: 0.8577 - val_acc: 0.6895\n",
      "Label: LABEL_3\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 7s - loss: 0.9147 - acc: 0.8816 - val_loss: 2.7239 - val_acc: 0.1070\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9081 - acc: 0.8924 - val_loss: 0.9269 - val_acc: 0.8930\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.9070 - acc: 0.8924 - val_loss: 0.9699 - val_acc: 0.4925\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.9064 - acc: 0.8924 - val_loss: 0.9298 - val_acc: 0.8685\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.9043 - acc: 0.8921 - val_loss: 1.0167 - val_acc: 0.8930\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.9043 - acc: 0.8919 - val_loss: 0.9157 - val_acc: 0.8930\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9017 - acc: 0.8912 - val_loss: 0.9192 - val_acc: 0.8930\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.9010 - acc: 0.8915 - val_loss: 0.9500 - val_acc: 0.7457\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.9004 - acc: 0.8900 - val_loss: 1.0648 - val_acc: 0.1365\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.8991 - acc: 0.8890 - val_loss: 0.9443 - val_acc: 0.8930\n",
      "Label: LABEL_4\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 7s - loss: 0.9035 - acc: 0.8597 - val_loss: 0.9321 - val_acc: 0.8685\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.8983 - acc: 0.8688 - val_loss: 0.9725 - val_acc: 0.8685\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.8957 - acc: 0.8689 - val_loss: 0.9086 - val_acc: 0.8685\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.8949 - acc: 0.8687 - val_loss: 0.9296 - val_acc: 0.7973\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.8936 - acc: 0.8689 - val_loss: 0.9057 - val_acc: 0.8685\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.8925 - acc: 0.8684 - val_loss: 0.9265 - val_acc: 0.8685\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.8908 - acc: 0.8661 - val_loss: 0.9176 - val_acc: 0.8685\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.8899 - acc: 0.8673 - val_loss: 0.9115 - val_acc: 0.8685\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.8876 - acc: 0.8651 - val_loss: 0.9874 - val_acc: 0.2878\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.8884 - acc: 0.8648 - val_loss: 0.9065 - val_acc: 0.8655\n",
      "Label: LABEL_5\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 8s - loss: 0.9322 - acc: 0.9218 - val_loss: 1.4558 - val_acc: 0.9310\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9251 - acc: 0.9317 - val_loss: 1.6298 - val_acc: 0.9310\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.9226 - acc: 0.9318 - val_loss: 1.0936 - val_acc: 0.0695\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9202 - acc: 0.9317 - val_loss: 0.9626 - val_acc: 0.9310\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9192 - acc: 0.9314 - val_loss: 3.9860 - val_acc: 0.0690\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.9178 - acc: 0.9294 - val_loss: 1.4114 - val_acc: 0.0690\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9136 - acc: 0.9295 - val_loss: 1.1027 - val_acc: 0.1083\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.9140 - acc: 0.9261 - val_loss: 1.3943 - val_acc: 0.0690\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.9141 - acc: 0.9270 - val_loss: 0.9880 - val_acc: 0.6515\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.9122 - acc: 0.9247 - val_loss: 1.1596 - val_acc: 0.9310\n",
      "Label: LABEL_6\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 8s - loss: 0.9212 - acc: 0.9022 - val_loss: 1.0187 - val_acc: 0.2203\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9151 - acc: 0.9087 - val_loss: 1.0538 - val_acc: 0.9057\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.9133 - acc: 0.9087 - val_loss: 0.9494 - val_acc: 0.9057\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.9115 - acc: 0.9087 - val_loss: 1.0001 - val_acc: 0.3320\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.9116 - acc: 0.9086 - val_loss: 0.9334 - val_acc: 0.9055\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.9103 - acc: 0.9085 - val_loss: 0.9835 - val_acc: 0.9057\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9083 - acc: 0.9085 - val_loss: 0.9382 - val_acc: 0.9020\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.9080 - acc: 0.9079 - val_loss: 0.9357 - val_acc: 0.9055\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.9075 - acc: 0.9083 - val_loss: 0.9369 - val_acc: 0.9050\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9059 - acc: 0.9069 - val_loss: 0.9376 - val_acc: 0.9052\n",
      "Label: LABEL_7\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 9s - loss: 0.9024 - acc: 0.8620 - val_loss: 1.1049 - val_acc: 0.1325\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.8977 - acc: 0.8677 - val_loss: 1.1073 - val_acc: 0.1325\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.8963 - acc: 0.8677 - val_loss: 0.9315 - val_acc: 0.8675\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.8957 - acc: 0.8677 - val_loss: 0.9465 - val_acc: 0.8675\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.8942 - acc: 0.8677 - val_loss: 1.0260 - val_acc: 0.8675\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.8924 - acc: 0.8675 - val_loss: 0.9898 - val_acc: 0.2070\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.8918 - acc: 0.8673 - val_loss: 0.9254 - val_acc: 0.8675\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.8911 - acc: 0.8669 - val_loss: 0.9121 - val_acc: 0.8675\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.8892 - acc: 0.8661 - val_loss: 0.9739 - val_acc: 0.8675\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.8892 - acc: 0.8657 - val_loss: 0.9460 - val_acc: 0.8675\n",
      "Label: LABEL_8\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 9s - loss: 0.9332 - acc: 0.9330 - val_loss: 2.3496 - val_acc: 0.0610\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9264 - acc: 0.9426 - val_loss: 1.2873 - val_acc: 0.0610\n",
      "Epoch 3/10\n",
      " - 4s - loss: 0.9258 - acc: 0.9426 - val_loss: 1.8616 - val_acc: 0.0610\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.9222 - acc: 0.9427 - val_loss: 1.0271 - val_acc: 0.4860\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.9217 - acc: 0.9420 - val_loss: 1.0883 - val_acc: 0.9390\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9203 - acc: 0.9420 - val_loss: 1.0160 - val_acc: 0.6135\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9215 - acc: 0.9423 - val_loss: 1.9167 - val_acc: 0.9390\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.9190 - acc: 0.9423 - val_loss: 1.6288 - val_acc: 0.9390\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.9172 - acc: 0.9409 - val_loss: 0.9855 - val_acc: 0.8650\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.9165 - acc: 0.9395 - val_loss: 1.0089 - val_acc: 0.6850\n",
      "Label: LABEL_9\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 10s - loss: 0.9292 - acc: 0.9096 - val_loss: 2.9794 - val_acc: 0.0793\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9220 - acc: 0.9156 - val_loss: 2.4843 - val_acc: 0.0793\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9215 - acc: 0.9156 - val_loss: 1.1393 - val_acc: 0.9207\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.9191 - acc: 0.9156 - val_loss: 1.5545 - val_acc: 0.0793\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9185 - acc: 0.9156 - val_loss: 0.8989 - val_acc: 0.9207\n",
      "Epoch 6/10\n",
      " - 4s - loss: 0.9170 - acc: 0.9156 - val_loss: 1.2732 - val_acc: 0.9207\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9187 - acc: 0.9153 - val_loss: 1.8247 - val_acc: 0.9207\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9163 - acc: 0.9156 - val_loss: 0.9788 - val_acc: 0.4242\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9168 - acc: 0.9151 - val_loss: 0.9203 - val_acc: 0.8985\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.9136 - acc: 0.9149 - val_loss: 0.9075 - val_acc: 0.9165\n",
      "Label: LABEL_10\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 10s - loss: 0.9050 - acc: 0.8684 - val_loss: 2.0192 - val_acc: 0.1277\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9006 - acc: 0.8741 - val_loss: 1.0354 - val_acc: 0.8722\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.8990 - acc: 0.8742 - val_loss: 1.0455 - val_acc: 0.8722\n",
      "Epoch 4/10\n",
      " - 4s - loss: 0.8979 - acc: 0.8742 - val_loss: 0.9354 - val_acc: 0.8722\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.8966 - acc: 0.8741 - val_loss: 0.9935 - val_acc: 0.1995\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.8960 - acc: 0.8741 - val_loss: 0.9145 - val_acc: 0.8722\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8948 - acc: 0.8739 - val_loss: 0.9634 - val_acc: 0.4953\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.8939 - acc: 0.8740 - val_loss: 0.9430 - val_acc: 0.7210\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.8938 - acc: 0.8737 - val_loss: 0.9989 - val_acc: 0.8722\n",
      "Epoch 10/10\n",
      " - 4s - loss: 0.8933 - acc: 0.8735 - val_loss: 0.9258 - val_acc: 0.8418\n",
      "Label: LABEL_11\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 11s - loss: 0.9452 - acc: 0.9429 - val_loss: 1.6113 - val_acc: 0.9560\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9342 - acc: 0.9569 - val_loss: 1.5469 - val_acc: 0.0440\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9318 - acc: 0.9568 - val_loss: 2.1599 - val_acc: 0.9560\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9285 - acc: 0.9562 - val_loss: 1.2036 - val_acc: 0.9560\n",
      "Epoch 5/10\n",
      " - 4s - loss: 0.9271 - acc: 0.9545 - val_loss: 1.1426 - val_acc: 0.0515\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9273 - acc: 0.9566 - val_loss: 2.1401 - val_acc: 0.0440\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9253 - acc: 0.9556 - val_loss: 1.0633 - val_acc: 0.2720\n",
      "Epoch 8/10\n",
      " - 4s - loss: 0.9234 - acc: 0.9510 - val_loss: 1.3897 - val_acc: 0.9560\n",
      "Epoch 9/10\n",
      " - 4s - loss: 0.9214 - acc: 0.9505 - val_loss: 0.9749 - val_acc: 0.9560\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9186 - acc: 0.9504 - val_loss: 1.0108 - val_acc: 0.9560\n",
      "Label: LABEL_12\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 11s - loss: 0.9422 - acc: 0.9300 - val_loss: 0.9351 - val_acc: 0.9417\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9333 - acc: 0.9392 - val_loss: 1.0397 - val_acc: 0.9437\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9315 - acc: 0.9392 - val_loss: 0.8992 - val_acc: 0.9437\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9305 - acc: 0.9393 - val_loss: 0.9471 - val_acc: 0.8518\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9279 - acc: 0.9392 - val_loss: 0.9562 - val_acc: 0.7442\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9247 - acc: 0.9376 - val_loss: 2.7105 - val_acc: 0.0563\n",
      "Epoch 7/10\n",
      " - 4s - loss: 0.9246 - acc: 0.9372 - val_loss: 1.2029 - val_acc: 0.0563\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9225 - acc: 0.9356 - val_loss: 1.3475 - val_acc: 0.0563\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9225 - acc: 0.9350 - val_loss: 1.4375 - val_acc: 0.9437\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9209 - acc: 0.9354 - val_loss: 0.9671 - val_acc: 0.9437\n",
      "Label: LABEL_13\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 12s - loss: 0.9228 - acc: 0.9058 - val_loss: 3.3561 - val_acc: 0.0893\n",
      "Epoch 2/10\n",
      " - 4s - loss: 0.9171 - acc: 0.9159 - val_loss: 3.3285 - val_acc: 0.9107\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9145 - acc: 0.9159 - val_loss: 0.9822 - val_acc: 0.9107\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9130 - acc: 0.9153 - val_loss: 0.9565 - val_acc: 0.9082\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9129 - acc: 0.9157 - val_loss: 1.0043 - val_acc: 0.4860\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9105 - acc: 0.9157 - val_loss: 1.1973 - val_acc: 0.0895\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9109 - acc: 0.9150 - val_loss: 1.0270 - val_acc: 0.9107\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9106 - acc: 0.9156 - val_loss: 0.9851 - val_acc: 0.7130\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9097 - acc: 0.9151 - val_loss: 0.9570 - val_acc: 0.9107\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9076 - acc: 0.9149 - val_loss: 0.9883 - val_acc: 0.6932\n",
      "Label: LABEL_14\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 12s - loss: 0.9211 - acc: 0.9126 - val_loss: 1.0052 - val_acc: 0.9130\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9161 - acc: 0.9195 - val_loss: 0.9655 - val_acc: 0.9130\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9146 - acc: 0.9195 - val_loss: 0.9799 - val_acc: 0.9130\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9124 - acc: 0.9194 - val_loss: 1.0759 - val_acc: 0.9130\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9112 - acc: 0.9194 - val_loss: 1.0645 - val_acc: 0.9130\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9107 - acc: 0.9193 - val_loss: 1.6865 - val_acc: 0.0870\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9090 - acc: 0.9192 - val_loss: 1.2389 - val_acc: 0.0870\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9066 - acc: 0.9185 - val_loss: 1.5937 - val_acc: 0.0870\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9056 - acc: 0.9170 - val_loss: 0.9665 - val_acc: 0.9015\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9025 - acc: 0.9167 - val_loss: 1.0139 - val_acc: 0.9130\n",
      "Label: LABEL_15\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 13s - loss: 0.9495 - acc: 0.9491 - val_loss: 2.1091 - val_acc: 0.9610\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9377 - acc: 0.9585 - val_loss: 0.9236 - val_acc: 0.9610\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9355 - acc: 0.9582 - val_loss: 1.3808 - val_acc: 0.0390\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9341 - acc: 0.9585 - val_loss: 1.3155 - val_acc: 0.0390\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9301 - acc: 0.9571 - val_loss: 1.0416 - val_acc: 0.9610\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9306 - acc: 0.9567 - val_loss: 1.0401 - val_acc: 0.9610\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9267 - acc: 0.9549 - val_loss: 1.2226 - val_acc: 0.9610\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9265 - acc: 0.9556 - val_loss: 1.0260 - val_acc: 0.9610\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9241 - acc: 0.9504 - val_loss: 0.9284 - val_acc: 0.9525\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9218 - acc: 0.9527 - val_loss: 0.9648 - val_acc: 0.7867\n",
      "Label: LABEL_16\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 13s - loss: 0.9510 - acc: 0.9568 - val_loss: 0.9426 - val_acc: 0.9700\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9421 - acc: 0.9698 - val_loss: 1.5590 - val_acc: 0.9700\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9368 - acc: 0.9693 - val_loss: 0.9471 - val_acc: 0.9690\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9351 - acc: 0.9663 - val_loss: 1.8429 - val_acc: 0.0300\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9344 - acc: 0.9689 - val_loss: 0.9477 - val_acc: 0.9688\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9351 - acc: 0.9683 - val_loss: 1.0326 - val_acc: 0.9700\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9334 - acc: 0.9665 - val_loss: 1.0318 - val_acc: 0.9700\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9323 - acc: 0.9670 - val_loss: 1.0276 - val_acc: 0.9700\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9285 - acc: 0.9642 - val_loss: 1.0897 - val_acc: 0.9700\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9277 - acc: 0.9627 - val_loss: 1.0240 - val_acc: 0.9700\n",
      "Label: LABEL_17\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 14s - loss: 0.9207 - acc: 0.8975 - val_loss: 1.8678 - val_acc: 0.9032\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9146 - acc: 0.9051 - val_loss: 1.0095 - val_acc: 0.1053\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9130 - acc: 0.9053 - val_loss: 0.9782 - val_acc: 0.9032\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9120 - acc: 0.9052 - val_loss: 0.9289 - val_acc: 0.9032\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9112 - acc: 0.9052 - val_loss: 1.5465 - val_acc: 0.9032\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9110 - acc: 0.9052 - val_loss: 0.9559 - val_acc: 0.8240\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9103 - acc: 0.9051 - val_loss: 0.9297 - val_acc: 0.9030\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9091 - acc: 0.9051 - val_loss: 0.9370 - val_acc: 0.9032\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9082 - acc: 0.9050 - val_loss: 1.0325 - val_acc: 0.9032\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9082 - acc: 0.9047 - val_loss: 0.9326 - val_acc: 0.9032\n",
      "Label: LABEL_18\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 14s - loss: 0.9328 - acc: 0.9161 - val_loss: 1.3845 - val_acc: 0.9287\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9244 - acc: 0.9251 - val_loss: 1.0754 - val_acc: 0.0723\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9239 - acc: 0.9251 - val_loss: 0.9363 - val_acc: 0.9130\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9213 - acc: 0.9244 - val_loss: 1.2878 - val_acc: 0.0713\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9194 - acc: 0.9244 - val_loss: 0.9324 - val_acc: 0.8998\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9199 - acc: 0.9247 - val_loss: 0.9116 - val_acc: 0.9282\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9184 - acc: 0.9244 - val_loss: 1.0183 - val_acc: 0.9287\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9170 - acc: 0.9237 - val_loss: 0.9321 - val_acc: 0.8885\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9168 - acc: 0.9220 - val_loss: 1.0910 - val_acc: 0.1020\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9142 - acc: 0.9216 - val_loss: 1.0016 - val_acc: 0.9287\n",
      "Label: LABEL_19\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 15s - loss: 0.9493 - acc: 0.9559 - val_loss: 0.9521 - val_acc: 0.9632\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9395 - acc: 0.9640 - val_loss: 1.1710 - val_acc: 0.0368\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9363 - acc: 0.9640 - val_loss: 1.4168 - val_acc: 0.9632\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9325 - acc: 0.9638 - val_loss: 0.9503 - val_acc: 0.9632\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9303 - acc: 0.9613 - val_loss: 1.0168 - val_acc: 0.5642\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9283 - acc: 0.9607 - val_loss: 0.9764 - val_acc: 0.8627\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9255 - acc: 0.9590 - val_loss: 1.0065 - val_acc: 0.6265\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9245 - acc: 0.9573 - val_loss: 1.4066 - val_acc: 0.9632\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9220 - acc: 0.9585 - val_loss: 0.9916 - val_acc: 0.9632\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9181 - acc: 0.9546 - val_loss: 0.9578 - val_acc: 0.9525\n",
      "Label: LABEL_20\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 16s - loss: 0.9378 - acc: 0.9362 - val_loss: 0.9651 - val_acc: 0.9445\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9313 - acc: 0.9464 - val_loss: 1.0976 - val_acc: 0.9445\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9285 - acc: 0.9464 - val_loss: 1.2991 - val_acc: 0.0555\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9269 - acc: 0.9463 - val_loss: 1.2105 - val_acc: 0.9445\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9252 - acc: 0.9462 - val_loss: 0.9964 - val_acc: 0.9445\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9237 - acc: 0.9462 - val_loss: 0.9846 - val_acc: 0.8243\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9207 - acc: 0.9442 - val_loss: 1.1699 - val_acc: 0.0623\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9180 - acc: 0.9430 - val_loss: 1.2028 - val_acc: 0.9445\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9155 - acc: 0.9406 - val_loss: 0.9766 - val_acc: 0.8398\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9129 - acc: 0.9391 - val_loss: 0.9561 - val_acc: 0.9410\n",
      "Label: LABEL_21\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 16s - loss: 0.9482 - acc: 0.9543 - val_loss: 1.7869 - val_acc: 0.0363\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9367 - acc: 0.9645 - val_loss: 0.9602 - val_acc: 0.9638\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9339 - acc: 0.9642 - val_loss: 0.9522 - val_acc: 0.9625\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9320 - acc: 0.9644 - val_loss: 2.8472 - val_acc: 0.9638\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9299 - acc: 0.9633 - val_loss: 0.9612 - val_acc: 0.9565\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9286 - acc: 0.9630 - val_loss: 0.9855 - val_acc: 0.8332\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9235 - acc: 0.9602 - val_loss: 1.2851 - val_acc: 0.0378\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9185 - acc: 0.9539 - val_loss: 1.2425 - val_acc: 0.0478\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9170 - acc: 0.9497 - val_loss: 1.0572 - val_acc: 0.9638\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9166 - acc: 0.9474 - val_loss: 1.0144 - val_acc: 0.9638\n",
      "Label: LABEL_22\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 17s - loss: 0.9378 - acc: 0.9381 - val_loss: 1.0828 - val_acc: 0.9430\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9286 - acc: 0.9472 - val_loss: 1.1045 - val_acc: 0.9430\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9272 - acc: 0.9472 - val_loss: 0.9714 - val_acc: 0.9430\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9254 - acc: 0.9472 - val_loss: 0.9912 - val_acc: 0.9227\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9218 - acc: 0.9469 - val_loss: 1.3313 - val_acc: 0.0570\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9221 - acc: 0.9470 - val_loss: 0.9935 - val_acc: 0.9430\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9184 - acc: 0.9465 - val_loss: 0.9763 - val_acc: 0.9425\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9178 - acc: 0.9462 - val_loss: 0.9798 - val_acc: 0.9282\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9177 - acc: 0.9436 - val_loss: 0.9855 - val_acc: 0.9023\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9175 - acc: 0.9441 - val_loss: 0.9818 - val_acc: 0.9385\n",
      "Label: LABEL_23\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 18s - loss: 0.9543 - acc: 0.9584 - val_loss: 2.2651 - val_acc: 0.9703\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9408 - acc: 0.9700 - val_loss: 2.0457 - val_acc: 0.0297\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9347 - acc: 0.9702 - val_loss: 0.9715 - val_acc: 0.9065\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9315 - acc: 0.9666 - val_loss: 1.1124 - val_acc: 0.0953\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9264 - acc: 0.9630 - val_loss: 0.9868 - val_acc: 0.9703\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9230 - acc: 0.9581 - val_loss: 0.9531 - val_acc: 0.9498\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9183 - acc: 0.9594 - val_loss: 0.9611 - val_acc: 0.9012\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9154 - acc: 0.9451 - val_loss: 1.0019 - val_acc: 0.6753\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9107 - acc: 0.9422 - val_loss: 1.0525 - val_acc: 0.4357\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9112 - acc: 0.9413 - val_loss: 1.0712 - val_acc: 0.3740\n",
      "Label: LABEL_24\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 18s - loss: 0.9628 - acc: 0.9599 - val_loss: 2.1578 - val_acc: 0.9790\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9487 - acc: 0.9771 - val_loss: 0.9144 - val_acc: 0.9790\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9433 - acc: 0.9764 - val_loss: 1.3019 - val_acc: 0.0210\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9402 - acc: 0.9744 - val_loss: 1.4297 - val_acc: 0.0210\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9348 - acc: 0.9724 - val_loss: 1.7065 - val_acc: 0.9790\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9279 - acc: 0.9601 - val_loss: 1.5781 - val_acc: 0.0210\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9230 - acc: 0.9573 - val_loss: 0.9228 - val_acc: 0.9665\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9231 - acc: 0.9476 - val_loss: 0.9575 - val_acc: 0.8060\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9214 - acc: 0.9544 - val_loss: 0.9588 - val_acc: 0.9788\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9146 - acc: 0.9420 - val_loss: 1.0421 - val_acc: 0.4200\n",
      "Label: LABEL_25\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 19s - loss: 0.9502 - acc: 0.9523 - val_loss: 1.1148 - val_acc: 0.9660\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9361 - acc: 0.9659 - val_loss: 1.8056 - val_acc: 0.9660\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9336 - acc: 0.9658 - val_loss: 1.5958 - val_acc: 0.9660\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9324 - acc: 0.9652 - val_loss: 1.1007 - val_acc: 0.9660\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9291 - acc: 0.9644 - val_loss: 0.9797 - val_acc: 0.9660\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9259 - acc: 0.9609 - val_loss: 1.2549 - val_acc: 0.0400\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9260 - acc: 0.9607 - val_loss: 1.2003 - val_acc: 0.0572\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9240 - acc: 0.9597 - val_loss: 0.9730 - val_acc: 0.9555\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9242 - acc: 0.9582 - val_loss: 1.4272 - val_acc: 0.9660\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9201 - acc: 0.9556 - val_loss: 0.9897 - val_acc: 0.9660\n",
      "Label: LABEL_26\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 19s - loss: 0.9502 - acc: 0.9484 - val_loss: 2.1230 - val_acc: 0.0375\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9362 - acc: 0.9626 - val_loss: 0.9444 - val_acc: 0.9625\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9335 - acc: 0.9621 - val_loss: 1.5274 - val_acc: 0.9625\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9289 - acc: 0.9594 - val_loss: 0.9948 - val_acc: 0.7125\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9269 - acc: 0.9583 - val_loss: 0.9551 - val_acc: 0.9475\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9230 - acc: 0.9554 - val_loss: 0.9488 - val_acc: 0.9600\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9215 - acc: 0.9527 - val_loss: 1.0826 - val_acc: 0.2240\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9235 - acc: 0.9533 - val_loss: 1.0029 - val_acc: 0.9625\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9212 - acc: 0.9519 - val_loss: 1.0237 - val_acc: 0.9625\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9178 - acc: 0.9453 - val_loss: 1.2622 - val_acc: 0.9625\n",
      "Label: LABEL_27\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 20s - loss: 0.9651 - acc: 0.9731 - val_loss: 0.9829 - val_acc: 0.9833\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9470 - acc: 0.9836 - val_loss: 1.4013 - val_acc: 0.0168\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9361 - acc: 0.9827 - val_loss: 1.0051 - val_acc: 0.8285\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9323 - acc: 0.9788 - val_loss: 0.9781 - val_acc: 0.9420\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9297 - acc: 0.9752 - val_loss: 0.9874 - val_acc: 0.9830\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9300 - acc: 0.9739 - val_loss: 1.1037 - val_acc: 0.2487\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9261 - acc: 0.9746 - val_loss: 1.0191 - val_acc: 0.9833\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9189 - acc: 0.9623 - val_loss: 1.0303 - val_acc: 0.6683\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9148 - acc: 0.9563 - val_loss: 1.0169 - val_acc: 0.7308\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9119 - acc: 0.9558 - val_loss: 0.9940 - val_acc: 0.9733\n",
      "Label: LABEL_28\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 21s - loss: 0.9514 - acc: 0.9506 - val_loss: 1.5604 - val_acc: 0.9605\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9399 - acc: 0.9573 - val_loss: 0.9158 - val_acc: 0.9605\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9365 - acc: 0.9573 - val_loss: 0.9684 - val_acc: 0.9605\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9344 - acc: 0.9568 - val_loss: 0.9082 - val_acc: 0.9605\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9318 - acc: 0.9567 - val_loss: 0.9105 - val_acc: 0.9602\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9291 - acc: 0.9529 - val_loss: 0.9619 - val_acc: 0.7620\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9262 - acc: 0.9535 - val_loss: 1.2569 - val_acc: 0.9605\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9213 - acc: 0.9446 - val_loss: 0.9200 - val_acc: 0.9535\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9215 - acc: 0.9456 - val_loss: 0.9334 - val_acc: 0.9602\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9176 - acc: 0.9420 - val_loss: 0.9364 - val_acc: 0.9605\n",
      "Label: LABEL_29\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 21s - loss: 0.9483 - acc: 0.9601 - val_loss: 1.0386 - val_acc: 0.9697\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9384 - acc: 0.9709 - val_loss: 0.9662 - val_acc: 0.9697\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9343 - acc: 0.9706 - val_loss: 1.0457 - val_acc: 0.3680\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9325 - acc: 0.9696 - val_loss: 1.4163 - val_acc: 0.0303\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9287 - acc: 0.9662 - val_loss: 1.0470 - val_acc: 0.9697\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9246 - acc: 0.9664 - val_loss: 0.9680 - val_acc: 0.9695\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9179 - acc: 0.9625 - val_loss: 0.9808 - val_acc: 0.8555\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9192 - acc: 0.9539 - val_loss: 1.0294 - val_acc: 0.5445\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9228 - acc: 0.9542 - val_loss: 0.9654 - val_acc: 0.9658\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9138 - acc: 0.9520 - val_loss: 1.0526 - val_acc: 0.4432\n",
      "Label: LABEL_30\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 22s - loss: 0.9526 - acc: 0.9550 - val_loss: 1.9230 - val_acc: 0.9693\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9440 - acc: 0.9663 - val_loss: 0.9969 - val_acc: 0.6823\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9398 - acc: 0.9676 - val_loss: 1.0207 - val_acc: 0.9693\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9335 - acc: 0.9663 - val_loss: 0.9909 - val_acc: 0.9693\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9327 - acc: 0.9631 - val_loss: 0.9810 - val_acc: 0.9693\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9299 - acc: 0.9636 - val_loss: 1.5097 - val_acc: 0.0308\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9218 - acc: 0.9560 - val_loss: 1.7958 - val_acc: 0.0308\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9229 - acc: 0.9467 - val_loss: 1.2792 - val_acc: 0.9693\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9180 - acc: 0.9425 - val_loss: 1.0487 - val_acc: 0.9693\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9159 - acc: 0.9427 - val_loss: 1.0667 - val_acc: 0.9693\n",
      "Label: LABEL_31\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 22s - loss: 0.9538 - acc: 0.9655 - val_loss: 2.3993 - val_acc: 0.9783\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9453 - acc: 0.9780 - val_loss: 8.2679 - val_acc: 0.0218\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9408 - acc: 0.9781 - val_loss: 3.6562 - val_acc: 0.0218\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9388 - acc: 0.9777 - val_loss: 1.0971 - val_acc: 0.9783\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9388 - acc: 0.9767 - val_loss: 1.0010 - val_acc: 0.9783\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9338 - acc: 0.9728 - val_loss: 1.0289 - val_acc: 0.9783\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9302 - acc: 0.9718 - val_loss: 1.3218 - val_acc: 0.0253\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9282 - acc: 0.9690 - val_loss: 1.1874 - val_acc: 0.9783\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9263 - acc: 0.9625 - val_loss: 1.7117 - val_acc: 0.9783\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9255 - acc: 0.9572 - val_loss: 1.5098 - val_acc: 0.9783\n",
      "Label: LABEL_32\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 23s - loss: 0.9613 - acc: 0.9553 - val_loss: 1.7838 - val_acc: 0.9723\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9459 - acc: 0.9705 - val_loss: 1.0009 - val_acc: 0.9723\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9418 - acc: 0.9705 - val_loss: 1.0622 - val_acc: 0.0750\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9377 - acc: 0.9704 - val_loss: 1.0438 - val_acc: 0.9723\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9321 - acc: 0.9684 - val_loss: 1.9592 - val_acc: 0.0277\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9296 - acc: 0.9664 - val_loss: 1.0466 - val_acc: 0.3170\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9294 - acc: 0.9648 - val_loss: 1.3379 - val_acc: 0.9723\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9232 - acc: 0.9593 - val_loss: 1.4449 - val_acc: 0.9723\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9220 - acc: 0.9532 - val_loss: 0.9396 - val_acc: 0.9512\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9208 - acc: 0.9557 - val_loss: 0.9355 - val_acc: 0.9550\n",
      "Label: LABEL_33\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 24s - loss: 0.9670 - acc: 0.9554 - val_loss: 0.8912 - val_acc: 0.9842\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9515 - acc: 0.9803 - val_loss: 1.0686 - val_acc: 0.9842\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9408 - acc: 0.9668 - val_loss: 1.9908 - val_acc: 0.9842\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9372 - acc: 0.9685 - val_loss: 0.9123 - val_acc: 0.9570\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9333 - acc: 0.9597 - val_loss: 0.8957 - val_acc: 0.9787\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9255 - acc: 0.9632 - val_loss: 1.6353 - val_acc: 0.9842\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9177 - acc: 0.9543 - val_loss: 0.9288 - val_acc: 0.9840\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9145 - acc: 0.9398 - val_loss: 0.9134 - val_acc: 0.9835\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9065 - acc: 0.9458 - val_loss: 1.0586 - val_acc: 0.9842\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9023 - acc: 0.9209 - val_loss: 0.9702 - val_acc: 0.9842\n",
      "Label: LABEL_34\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 25s - loss: 0.9559 - acc: 0.9595 - val_loss: 1.7339 - val_acc: 0.0192\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9445 - acc: 0.9812 - val_loss: 0.9755 - val_acc: 0.9808\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9364 - acc: 0.9776 - val_loss: 1.6881 - val_acc: 0.0192\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9330 - acc: 0.9780 - val_loss: 0.9966 - val_acc: 0.9808\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9256 - acc: 0.9738 - val_loss: 1.0119 - val_acc: 0.8393\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9229 - acc: 0.9659 - val_loss: 1.2404 - val_acc: 0.0568\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9216 - acc: 0.9644 - val_loss: 1.0458 - val_acc: 0.6345\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9152 - acc: 0.9451 - val_loss: 1.0090 - val_acc: 0.8975\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9089 - acc: 0.9606 - val_loss: 1.1137 - val_acc: 0.9808\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9072 - acc: 0.9503 - val_loss: 1.0933 - val_acc: 0.9805\n",
      "Label: LABEL_35\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 25s - loss: 0.9497 - acc: 0.9519 - val_loss: 1.1735 - val_acc: 0.0288\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9380 - acc: 0.9731 - val_loss: 1.1170 - val_acc: 0.0640\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9332 - acc: 0.9680 - val_loss: 1.9323 - val_acc: 0.0288\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9324 - acc: 0.9717 - val_loss: 1.4409 - val_acc: 0.9712\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9292 - acc: 0.9692 - val_loss: 1.1276 - val_acc: 0.9712\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9258 - acc: 0.9683 - val_loss: 1.2575 - val_acc: 0.9712\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9262 - acc: 0.9664 - val_loss: 1.1049 - val_acc: 0.9712\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9248 - acc: 0.9661 - val_loss: 1.1337 - val_acc: 0.9712\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9208 - acc: 0.9607 - val_loss: 1.7757 - val_acc: 0.9712\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9233 - acc: 0.9655 - val_loss: 1.5517 - val_acc: 0.9712\n",
      "Label: LABEL_36\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 26s - loss: 0.9663 - acc: 0.9597 - val_loss: 0.8614 - val_acc: 0.9785\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9536 - acc: 0.9733 - val_loss: 0.9419 - val_acc: 0.9785\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9484 - acc: 0.9732 - val_loss: 1.3624 - val_acc: 0.9785\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9481 - acc: 0.9737 - val_loss: 1.9710 - val_acc: 0.0215\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9432 - acc: 0.9727 - val_loss: 0.9078 - val_acc: 0.9490\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9392 - acc: 0.9686 - val_loss: 1.2855 - val_acc: 0.0220\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9370 - acc: 0.9644 - val_loss: 1.0832 - val_acc: 0.1202\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9323 - acc: 0.9640 - val_loss: 0.9003 - val_acc: 0.9372\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9322 - acc: 0.9582 - val_loss: 0.9870 - val_acc: 0.9785\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9268 - acc: 0.9537 - val_loss: 1.0738 - val_acc: 0.2030\n",
      "Label: LABEL_37\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 26s - loss: 0.9536 - acc: 0.9694 - val_loss: 1.1677 - val_acc: 0.9862\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9381 - acc: 0.9880 - val_loss: 1.1983 - val_acc: 0.9862\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9306 - acc: 0.9831 - val_loss: 3.5102 - val_acc: 0.9862\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9259 - acc: 0.9825 - val_loss: 1.1578 - val_acc: 0.9862\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9152 - acc: 0.9691 - val_loss: 1.0636 - val_acc: 0.9782\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9042 - acc: 0.9622 - val_loss: 1.9958 - val_acc: 0.9862\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8988 - acc: 0.9447 - val_loss: 2.0360 - val_acc: 0.0138\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8965 - acc: 0.9433 - val_loss: 1.7558 - val_acc: 0.9862\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8977 - acc: 0.9348 - val_loss: 1.0708 - val_acc: 0.9160\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8827 - acc: 0.9337 - val_loss: 1.0740 - val_acc: 0.7868\n",
      "Label: LABEL_38\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 27s - loss: 0.9528 - acc: 0.9519 - val_loss: 0.9456 - val_acc: 0.9677\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9409 - acc: 0.9667 - val_loss: 1.0900 - val_acc: 0.9677\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9377 - acc: 0.9648 - val_loss: 0.9459 - val_acc: 0.9675\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9323 - acc: 0.9663 - val_loss: 0.9589 - val_acc: 0.8955\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9321 - acc: 0.9622 - val_loss: 1.1854 - val_acc: 0.0458\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9256 - acc: 0.9541 - val_loss: 1.0770 - val_acc: 0.1830\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9272 - acc: 0.9599 - val_loss: 1.0713 - val_acc: 0.9677\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9210 - acc: 0.9573 - val_loss: 1.2029 - val_acc: 0.0738\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9189 - acc: 0.9490 - val_loss: 0.9488 - val_acc: 0.9620\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9193 - acc: 0.9435 - val_loss: 0.9775 - val_acc: 0.7690\n",
      "Label: LABEL_39\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 28s - loss: 0.9532 - acc: 0.9651 - val_loss: 5.1073 - val_acc: 0.0248\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9383 - acc: 0.9759 - val_loss: 2.2508 - val_acc: 0.0248\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9360 - acc: 0.9689 - val_loss: 1.5007 - val_acc: 0.9752\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9283 - acc: 0.9736 - val_loss: 1.1710 - val_acc: 0.0515\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9236 - acc: 0.9674 - val_loss: 0.9927 - val_acc: 0.7650\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9231 - acc: 0.9574 - val_loss: 1.1169 - val_acc: 0.1747\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9201 - acc: 0.9585 - val_loss: 0.9593 - val_acc: 0.9713\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9157 - acc: 0.9571 - val_loss: 1.0746 - val_acc: 0.9752\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9137 - acc: 0.9433 - val_loss: 1.0159 - val_acc: 0.9752\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9059 - acc: 0.9462 - val_loss: 1.0184 - val_acc: 0.6000\n",
      "Label: LABEL_40\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 28s - loss: 0.9627 - acc: 0.9733 - val_loss: 1.8454 - val_acc: 0.9865\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9485 - acc: 0.9862 - val_loss: 2.7514 - val_acc: 0.9865\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9440 - acc: 0.9856 - val_loss: 0.9804 - val_acc: 0.9495\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9401 - acc: 0.9849 - val_loss: 0.9602 - val_acc: 0.9865\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9304 - acc: 0.9816 - val_loss: 0.9982 - val_acc: 0.7855\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9242 - acc: 0.9676 - val_loss: 4.1420 - val_acc: 0.0135\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9231 - acc: 0.9722 - val_loss: 0.9663 - val_acc: 0.9780\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9193 - acc: 0.9626 - val_loss: 1.6013 - val_acc: 0.9865\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9128 - acc: 0.9540 - val_loss: 1.3106 - val_acc: 0.0425\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9071 - acc: 0.9572 - val_loss: 1.4956 - val_acc: 0.9865\n",
      "Label: LABEL_41\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 29s - loss: 0.9569 - acc: 0.9636 - val_loss: 1.5015 - val_acc: 0.9808\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9483 - acc: 0.9758 - val_loss: 1.4330 - val_acc: 0.9808\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9437 - acc: 0.9798 - val_loss: 1.0249 - val_acc: 0.9808\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9377 - acc: 0.9781 - val_loss: 0.9647 - val_acc: 0.8980\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9343 - acc: 0.9727 - val_loss: 0.9434 - val_acc: 0.9805\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9271 - acc: 0.9700 - val_loss: 1.0566 - val_acc: 0.9808\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9250 - acc: 0.9631 - val_loss: 0.9692 - val_acc: 0.9808\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9203 - acc: 0.9595 - val_loss: 1.2621 - val_acc: 0.0520\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9115 - acc: 0.9581 - val_loss: 2.7536 - val_acc: 0.0193\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.9089 - acc: 0.9393 - val_loss: 1.1410 - val_acc: 0.2245\n",
      "Label: LABEL_42\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 30s - loss: 0.9806 - acc: 0.9528 - val_loss: 1.1094 - val_acc: 0.9933\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9565 - acc: 0.9917 - val_loss: 0.8882 - val_acc: 0.9933\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9437 - acc: 0.9877 - val_loss: 1.0821 - val_acc: 0.1867\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9383 - acc: 0.9763 - val_loss: 0.9398 - val_acc: 0.9933\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9291 - acc: 0.9714 - val_loss: 1.8099 - val_acc: 0.9933\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9153 - acc: 0.9467 - val_loss: 1.4892 - val_acc: 0.9933\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8990 - acc: 0.9384 - val_loss: 0.9145 - val_acc: 0.9685\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8863 - acc: 0.9241 - val_loss: 0.9634 - val_acc: 0.9930\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8796 - acc: 0.9026 - val_loss: 1.5525 - val_acc: 0.9933\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8847 - acc: 0.9003 - val_loss: 0.9890 - val_acc: 0.9928\n",
      "Label: LABEL_43\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 32s - loss: 0.9803 - acc: 0.9515 - val_loss: 2.3614 - val_acc: 0.0053\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9517 - acc: 0.9854 - val_loss: 0.9842 - val_acc: 0.9948\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9262 - acc: 0.9587 - val_loss: 1.4005 - val_acc: 0.0057\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9205 - acc: 0.9503 - val_loss: 3.3042 - val_acc: 0.0053\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9054 - acc: 0.9436 - val_loss: 1.7083 - val_acc: 0.9948\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.8971 - acc: 0.9298 - val_loss: 0.9517 - val_acc: 0.9855\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8866 - acc: 0.9138 - val_loss: 0.9283 - val_acc: 0.8280\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8621 - acc: 0.8940 - val_loss: 1.0803 - val_acc: 0.9935\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8465 - acc: 0.8925 - val_loss: 1.1893 - val_acc: 0.2788\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8459 - acc: 0.8937 - val_loss: 1.0390 - val_acc: 0.5542\n",
      "Label: LABEL_44\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 32s - loss: 0.9608 - acc: 0.9676 - val_loss: 1.3536 - val_acc: 0.9868\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9428 - acc: 0.9877 - val_loss: 1.2554 - val_acc: 0.9868\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9369 - acc: 0.9867 - val_loss: 1.3922 - val_acc: 0.9868\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9338 - acc: 0.9828 - val_loss: 1.3886 - val_acc: 0.9868\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9227 - acc: 0.9852 - val_loss: 0.9930 - val_acc: 0.9150\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9179 - acc: 0.9739 - val_loss: 1.1655 - val_acc: 0.1335\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9165 - acc: 0.9678 - val_loss: 1.0545 - val_acc: 0.9865\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9088 - acc: 0.9591 - val_loss: 1.2266 - val_acc: 0.0955\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8955 - acc: 0.9452 - val_loss: 2.0202 - val_acc: 0.0133\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8934 - acc: 0.9396 - val_loss: 1.2279 - val_acc: 0.9865\n",
      "Label: LABEL_45\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 33s - loss: 0.9633 - acc: 0.9755 - val_loss: 7.4295 - val_acc: 0.0145\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9445 - acc: 0.9860 - val_loss: 1.1781 - val_acc: 0.0198\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9333 - acc: 0.9813 - val_loss: 1.0171 - val_acc: 0.8230\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9307 - acc: 0.9797 - val_loss: 2.4044 - val_acc: 0.9855\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9254 - acc: 0.9709 - val_loss: 1.1521 - val_acc: 0.9855\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9213 - acc: 0.9713 - val_loss: 1.0066 - val_acc: 0.9808\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9168 - acc: 0.9622 - val_loss: 1.7670 - val_acc: 0.0145\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.9084 - acc: 0.9488 - val_loss: 1.0426 - val_acc: 0.6990\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.9045 - acc: 0.9492 - val_loss: 1.0362 - val_acc: 0.7045\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8970 - acc: 0.9410 - val_loss: 1.0788 - val_acc: 0.9835\n",
      "Label: LABEL_46\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 34s - loss: 0.9807 - acc: 0.9713 - val_loss: 1.1333 - val_acc: 0.9945\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9532 - acc: 0.9926 - val_loss: 1.6297 - val_acc: 0.9945\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9403 - acc: 0.9882 - val_loss: 1.1226 - val_acc: 0.1872\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9284 - acc: 0.9594 - val_loss: 1.0192 - val_acc: 0.9938\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9231 - acc: 0.9541 - val_loss: 3.1649 - val_acc: 0.0055\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9068 - acc: 0.9523 - val_loss: 4.8730 - val_acc: 0.0055\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.9004 - acc: 0.9426 - val_loss: 1.3741 - val_acc: 0.0735\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8747 - acc: 0.9220 - val_loss: 2.7652 - val_acc: 0.9945\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8805 - acc: 0.9170 - val_loss: 1.0873 - val_acc: 0.9672\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8525 - acc: 0.8789 - val_loss: 1.4120 - val_acc: 0.9945\n",
      "Label: LABEL_47\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 34s - loss: 1.0051 - acc: 0.9827 - val_loss: 3.2559 - val_acc: 0.0028\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9617 - acc: 0.9956 - val_loss: 2.1693 - val_acc: 0.0028\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9359 - acc: 0.9792 - val_loss: 0.8458 - val_acc: 0.9592\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9121 - acc: 0.9598 - val_loss: 1.0053 - val_acc: 0.9972\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.8925 - acc: 0.9403 - val_loss: 2.9146 - val_acc: 0.9972\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.8927 - acc: 0.9121 - val_loss: 1.0874 - val_acc: 0.3015\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8537 - acc: 0.9275 - val_loss: 1.1925 - val_acc: 0.2425\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8359 - acc: 0.8792 - val_loss: 3.2964 - val_acc: 0.9972\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8487 - acc: 0.9012 - val_loss: 3.2915 - val_acc: 0.0028\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8378 - acc: 0.8900 - val_loss: 0.9545 - val_acc: 0.6967\n",
      "Label: LABEL_48\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 35s - loss: 1.0323 - acc: 0.9591 - val_loss: 2.2624 - val_acc: 0.0013\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9392 - acc: 0.9790 - val_loss: 0.9952 - val_acc: 0.9988\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.8905 - acc: 0.9248 - val_loss: 0.9828 - val_acc: 0.9988\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.8256 - acc: 0.9077 - val_loss: 2.5121 - val_acc: 0.9988\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.7298 - acc: 0.9027 - val_loss: 0.9207 - val_acc: 0.9837\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.6935 - acc: 0.8722 - val_loss: 2.3905 - val_acc: 0.9988\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.6915 - acc: 0.8884 - val_loss: 1.7961 - val_acc: 0.9988\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.6819 - acc: 0.8664 - val_loss: 1.5404 - val_acc: 0.9988\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.6533 - acc: 0.8855 - val_loss: 1.5472 - val_acc: 0.9985\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.5963 - acc: 0.8890 - val_loss: 1.0985 - val_acc: 0.6397\n",
      "Label: LABEL_49\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 36s - loss: 0.9891 - acc: 0.9431 - val_loss: 0.7954 - val_acc: 0.9957\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9577 - acc: 0.9886 - val_loss: 0.8357 - val_acc: 0.9938\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9364 - acc: 0.9663 - val_loss: 1.8065 - val_acc: 0.9957\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9352 - acc: 0.9560 - val_loss: 1.9259 - val_acc: 0.0043\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9187 - acc: 0.9608 - val_loss: 0.8612 - val_acc: 0.9950\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9058 - acc: 0.9283 - val_loss: 0.8891 - val_acc: 0.8945\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8980 - acc: 0.9123 - val_loss: 1.0102 - val_acc: 0.5105\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8811 - acc: 0.9079 - val_loss: 1.8638 - val_acc: 0.0050\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8782 - acc: 0.9008 - val_loss: 0.8831 - val_acc: 0.9225\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8722 - acc: 0.9066 - val_loss: 1.1802 - val_acc: 0.9957\n",
      "Label: LABEL_50\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 36s - loss: 0.9883 - acc: 0.9725 - val_loss: 3.1547 - val_acc: 0.0060\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9593 - acc: 0.9874 - val_loss: 1.5897 - val_acc: 0.9940\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9447 - acc: 0.9786 - val_loss: 1.4314 - val_acc: 0.9940\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9207 - acc: 0.9602 - val_loss: 2.0568 - val_acc: 0.9940\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9144 - acc: 0.9425 - val_loss: 0.9285 - val_acc: 0.6452\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9092 - acc: 0.9167 - val_loss: 0.9896 - val_acc: 0.9940\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8970 - acc: 0.9191 - val_loss: 0.8499 - val_acc: 0.9385\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8731 - acc: 0.8960 - val_loss: 1.2922 - val_acc: 0.0912\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8857 - acc: 0.9216 - val_loss: 1.0658 - val_acc: 0.3780\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8714 - acc: 0.8997 - val_loss: 0.9145 - val_acc: 0.7307\n",
      "Label: LABEL_51\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 37s - loss: 1.0514 - acc: 0.8925 - val_loss: 2.0394 - val_acc: 0.9992\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9384 - acc: 0.9529 - val_loss: 12.8724 - val_acc: 7.5000e-04\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9508 - acc: 0.9394 - val_loss: 7.6366 - val_acc: 7.5000e-04\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.8303 - acc: 0.9003 - val_loss: 1.1225 - val_acc: 0.9992\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.8000 - acc: 0.8946 - val_loss: 2.9666 - val_acc: 0.9992\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.7506 - acc: 0.9082 - val_loss: 1.0873 - val_acc: 0.9992\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.7053 - acc: 0.8640 - val_loss: 1.2376 - val_acc: 0.9992\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.6379 - acc: 0.8789 - val_loss: 0.9654 - val_acc: 0.6482\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.6131 - acc: 0.8905 - val_loss: 1.1022 - val_acc: 0.5185\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.5525 - acc: 0.8952 - val_loss: 0.9481 - val_acc: 0.9047\n",
      "Label: LABEL_52\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 37s - loss: 1.0097 - acc: 0.9663 - val_loss: 0.7387 - val_acc: 0.9975\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9774 - acc: 0.9883 - val_loss: 1.2314 - val_acc: 0.0025\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9709 - acc: 0.9879 - val_loss: 1.0187 - val_acc: 0.1057\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9444 - acc: 0.9634 - val_loss: 0.7592 - val_acc: 0.9530\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9333 - acc: 0.9334 - val_loss: 0.8280 - val_acc: 0.9975\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.9265 - acc: 0.9314 - val_loss: 1.6252 - val_acc: 0.9975\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8977 - acc: 0.9113 - val_loss: 3.6708 - val_acc: 0.0025\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8966 - acc: 0.9190 - val_loss: 1.6418 - val_acc: 0.0188\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8867 - acc: 0.8969 - val_loss: 2.1285 - val_acc: 0.0055\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8708 - acc: 0.8708 - val_loss: 1.3855 - val_acc: 0.1085\n",
      "Label: LABEL_53\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 39s - loss: 0.9851 - acc: 0.9446 - val_loss: 1.0537 - val_acc: 0.1905\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9472 - acc: 0.9681 - val_loss: 1.5468 - val_acc: 0.9967\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9330 - acc: 0.9665 - val_loss: 1.1940 - val_acc: 0.1127\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9058 - acc: 0.9389 - val_loss: 2.7584 - val_acc: 0.0033\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9008 - acc: 0.9307 - val_loss: 0.9558 - val_acc: 0.9950\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.8732 - acc: 0.9028 - val_loss: 1.0040 - val_acc: 0.5830\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8530 - acc: 0.9030 - val_loss: 6.0290 - val_acc: 0.0033\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8621 - acc: 0.8903 - val_loss: 1.2466 - val_acc: 0.2678\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8453 - acc: 0.8896 - val_loss: 1.3474 - val_acc: 0.1630\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8368 - acc: 0.8800 - val_loss: 0.9856 - val_acc: 0.7505\n",
      "Label: LABEL_54\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 40s - loss: 0.9530 - acc: 0.9639 - val_loss: 1.4053 - val_acc: 0.0045\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9030 - acc: 0.9836 - val_loss: 1.1890 - val_acc: 0.9040\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9029 - acc: 0.9656 - val_loss: 2.5769 - val_acc: 0.9955\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.8564 - acc: 0.9406 - val_loss: 1.6594 - val_acc: 0.0085\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.8513 - acc: 0.9403 - val_loss: 2.6818 - val_acc: 0.9955\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.8316 - acc: 0.9033 - val_loss: 1.2281 - val_acc: 0.6158\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.7983 - acc: 0.9027 - val_loss: 1.7356 - val_acc: 0.0545\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.7712 - acc: 0.9009 - val_loss: 1.2557 - val_acc: 0.6715\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.7617 - acc: 0.8801 - val_loss: 5.2876 - val_acc: 0.9955\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.7523 - acc: 0.8875 - val_loss: 2.4894 - val_acc: 0.9950\n",
      "Label: LABEL_55\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 40s - loss: 1.0009 - acc: 0.9459 - val_loss: 1.3128 - val_acc: 0.9985\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9691 - acc: 0.9795 - val_loss: 3.8054 - val_acc: 0.0015\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.9532 - acc: 0.9782 - val_loss: 1.5195 - val_acc: 0.9985\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.9270 - acc: 0.9728 - val_loss: 1.6067 - val_acc: 0.0050\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.9040 - acc: 0.9291 - val_loss: 1.5469 - val_acc: 0.0037\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.8855 - acc: 0.9135 - val_loss: 3.7546 - val_acc: 0.0015\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.8359 - acc: 0.8860 - val_loss: 1.6696 - val_acc: 0.0425\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.8476 - acc: 0.8871 - val_loss: 0.8774 - val_acc: 0.8918\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.8021 - acc: 0.8779 - val_loss: 7.2528 - val_acc: 0.0015\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.8015 - acc: 0.8678 - val_loss: 4.1739 - val_acc: 0.9985\n",
      "Label: LABEL_56\n",
      "Train on 36000 samples, validate on 4000 samples\n",
      "Epoch 1/10\n",
      " - 42s - loss: 0.9792 - acc: 0.9436 - val_loss: 1.0917 - val_acc: 0.9965\n",
      "Epoch 2/10\n",
      " - 5s - loss: 0.9049 - acc: 0.9778 - val_loss: 3.1886 - val_acc: 0.9965\n",
      "Epoch 3/10\n",
      " - 5s - loss: 0.8787 - acc: 0.9366 - val_loss: 4.8419 - val_acc: 0.0035\n",
      "Epoch 4/10\n",
      " - 5s - loss: 0.8388 - acc: 0.9227 - val_loss: 2.7364 - val_acc: 0.9965\n",
      "Epoch 5/10\n",
      " - 5s - loss: 0.8103 - acc: 0.8981 - val_loss: 1.1124 - val_acc: 0.7945\n",
      "Epoch 6/10\n",
      " - 5s - loss: 0.7641 - acc: 0.8719 - val_loss: 4.6261 - val_acc: 0.9965\n",
      "Epoch 7/10\n",
      " - 5s - loss: 0.7668 - acc: 0.8821 - val_loss: 2.1413 - val_acc: 0.0365\n",
      "Epoch 8/10\n",
      " - 5s - loss: 0.7587 - acc: 0.8827 - val_loss: 4.6404 - val_acc: 0.9965\n",
      "Epoch 9/10\n",
      " - 5s - loss: 0.7081 - acc: 0.8846 - val_loss: 1.3619 - val_acc: 0.3217\n",
      "Epoch 10/10\n",
      " - 5s - loss: 0.6893 - acc: 0.8743 - val_loss: 1.2938 - val_acc: 0.8975\n",
      "CPU times: user 1h 4min 58s, sys: 11min 38s, total: 1h 16min 36s\n",
      "Wall time: 1h 11min 3s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "for label in labels:\n",
    "    print(\"Label:\", label)\n",
    "    model_all = get_model_all()\n",
    "    cw = compute_class_weight(\"balanced\", np.unique(y_train[label]), y_train[label])\n",
    "    model_all.fit(df_train, y_train[label], batch_size=bs, epochs=10, verbose=2, validation_split = 0.1, class_weight={0:1, 1:cw[1]})\n",
    "    submission_all[label] = model_all.predict(df_test)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>LABEL_1</th>\n",
       "      <th>LABEL_2</th>\n",
       "      <th>LABEL_3</th>\n",
       "      <th>LABEL_4</th>\n",
       "      <th>LABEL_5</th>\n",
       "      <th>LABEL_6</th>\n",
       "      <th>LABEL_7</th>\n",
       "      <th>LABEL_8</th>\n",
       "      <th>LABEL_9</th>\n",
       "      <th>LABEL_10</th>\n",
       "      <th>LABEL_11</th>\n",
       "      <th>LABEL_12</th>\n",
       "      <th>LABEL_13</th>\n",
       "      <th>LABEL_14</th>\n",
       "      <th>LABEL_15</th>\n",
       "      <th>LABEL_16</th>\n",
       "      <th>LABEL_17</th>\n",
       "      <th>LABEL_18</th>\n",
       "      <th>LABEL_19</th>\n",
       "      <th>LABEL_20</th>\n",
       "      <th>LABEL_21</th>\n",
       "      <th>LABEL_22</th>\n",
       "      <th>LABEL_23</th>\n",
       "      <th>LABEL_24</th>\n",
       "      <th>LABEL_25</th>\n",
       "      <th>LABEL_26</th>\n",
       "      <th>LABEL_27</th>\n",
       "      <th>LABEL_28</th>\n",
       "      <th>LABEL_29</th>\n",
       "      <th>LABEL_30</th>\n",
       "      <th>LABEL_31</th>\n",
       "      <th>LABEL_32</th>\n",
       "      <th>LABEL_33</th>\n",
       "      <th>LABEL_34</th>\n",
       "      <th>LABEL_35</th>\n",
       "      <th>LABEL_36</th>\n",
       "      <th>LABEL_37</th>\n",
       "      <th>LABEL_38</th>\n",
       "      <th>LABEL_39</th>\n",
       "      <th>LABEL_40</th>\n",
       "      <th>LABEL_41</th>\n",
       "      <th>LABEL_42</th>\n",
       "      <th>LABEL_43</th>\n",
       "      <th>LABEL_44</th>\n",
       "      <th>LABEL_45</th>\n",
       "      <th>LABEL_46</th>\n",
       "      <th>LABEL_47</th>\n",
       "      <th>LABEL_48</th>\n",
       "      <th>LABEL_49</th>\n",
       "      <th>LABEL_50</th>\n",
       "      <th>LABEL_51</th>\n",
       "      <th>LABEL_52</th>\n",
       "      <th>LABEL_53</th>\n",
       "      <th>LABEL_54</th>\n",
       "      <th>LABEL_55</th>\n",
       "      <th>LABEL_56</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ID</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>29408483</th>\n",
       "      <td>0.375211</td>\n",
       "      <td>0.297849</td>\n",
       "      <td>0.265300</td>\n",
       "      <td>0.350288</td>\n",
       "      <td>0.154023</td>\n",
       "      <td>0.345622</td>\n",
       "      <td>0.276210</td>\n",
       "      <td>0.422184</td>\n",
       "      <td>0.424674</td>\n",
       "      <td>0.400552</td>\n",
       "      <td>0.216312</td>\n",
       "      <td>0.239041</td>\n",
       "      <td>0.519539</td>\n",
       "      <td>0.197201</td>\n",
       "      <td>0.536737</td>\n",
       "      <td>0.308854</td>\n",
       "      <td>0.317217</td>\n",
       "      <td>0.203004</td>\n",
       "      <td>0.447135</td>\n",
       "      <td>0.318407</td>\n",
       "      <td>0.307669</td>\n",
       "      <td>0.440042</td>\n",
       "      <td>0.526765</td>\n",
       "      <td>0.456734</td>\n",
       "      <td>0.253003</td>\n",
       "      <td>0.116505</td>\n",
       "      <td>0.345685</td>\n",
       "      <td>0.256384</td>\n",
       "      <td>0.454443</td>\n",
       "      <td>0.136368</td>\n",
       "      <td>0.037330</td>\n",
       "      <td>0.356299</td>\n",
       "      <td>0.145584</td>\n",
       "      <td>0.158070</td>\n",
       "      <td>0.063105</td>\n",
       "      <td>0.474795</td>\n",
       "      <td>0.349344</td>\n",
       "      <td>0.371402</td>\n",
       "      <td>0.462295</td>\n",
       "      <td>0.074447</td>\n",
       "      <td>0.400013</td>\n",
       "      <td>0.144995</td>\n",
       "      <td>0.599085</td>\n",
       "      <td>0.177583</td>\n",
       "      <td>0.271306</td>\n",
       "      <td>0.078640</td>\n",
       "      <td>0.130656</td>\n",
       "      <td>0.608880</td>\n",
       "      <td>0.023444</td>\n",
       "      <td>0.216007</td>\n",
       "      <td>0.044471</td>\n",
       "      <td>0.790826</td>\n",
       "      <td>0.615428</td>\n",
       "      <td>0.060955</td>\n",
       "      <td>8.940697e-07</td>\n",
       "      <td>0.277034</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29164528</th>\n",
       "      <td>0.390474</td>\n",
       "      <td>0.273048</td>\n",
       "      <td>0.327629</td>\n",
       "      <td>0.346334</td>\n",
       "      <td>0.136660</td>\n",
       "      <td>0.293070</td>\n",
       "      <td>0.257421</td>\n",
       "      <td>0.476090</td>\n",
       "      <td>0.384906</td>\n",
       "      <td>0.450669</td>\n",
       "      <td>0.263949</td>\n",
       "      <td>0.248000</td>\n",
       "      <td>0.512216</td>\n",
       "      <td>0.254772</td>\n",
       "      <td>0.478946</td>\n",
       "      <td>0.176017</td>\n",
       "      <td>0.307289</td>\n",
       "      <td>0.191435</td>\n",
       "      <td>0.265389</td>\n",
       "      <td>0.318141</td>\n",
       "      <td>0.147023</td>\n",
       "      <td>0.366116</td>\n",
       "      <td>0.641957</td>\n",
       "      <td>0.507710</td>\n",
       "      <td>0.315365</td>\n",
       "      <td>0.081191</td>\n",
       "      <td>0.309972</td>\n",
       "      <td>0.317458</td>\n",
       "      <td>0.473557</td>\n",
       "      <td>0.113448</td>\n",
       "      <td>0.053010</td>\n",
       "      <td>0.315977</td>\n",
       "      <td>0.204194</td>\n",
       "      <td>0.133927</td>\n",
       "      <td>0.089654</td>\n",
       "      <td>0.544612</td>\n",
       "      <td>0.350240</td>\n",
       "      <td>0.394794</td>\n",
       "      <td>0.295509</td>\n",
       "      <td>0.072132</td>\n",
       "      <td>0.632276</td>\n",
       "      <td>0.258283</td>\n",
       "      <td>0.660273</td>\n",
       "      <td>0.089375</td>\n",
       "      <td>0.205327</td>\n",
       "      <td>0.049376</td>\n",
       "      <td>0.247008</td>\n",
       "      <td>0.221580</td>\n",
       "      <td>0.035707</td>\n",
       "      <td>0.557513</td>\n",
       "      <td>0.008231</td>\n",
       "      <td>0.555697</td>\n",
       "      <td>0.483235</td>\n",
       "      <td>0.018296</td>\n",
       "      <td>1.311302e-06</td>\n",
       "      <td>0.356881</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29475634</th>\n",
       "      <td>0.387763</td>\n",
       "      <td>0.325422</td>\n",
       "      <td>0.228740</td>\n",
       "      <td>0.323630</td>\n",
       "      <td>0.119543</td>\n",
       "      <td>0.377189</td>\n",
       "      <td>0.202776</td>\n",
       "      <td>0.417498</td>\n",
       "      <td>0.418968</td>\n",
       "      <td>0.479711</td>\n",
       "      <td>0.275895</td>\n",
       "      <td>0.137610</td>\n",
       "      <td>0.444862</td>\n",
       "      <td>0.244024</td>\n",
       "      <td>0.369476</td>\n",
       "      <td>0.221020</td>\n",
       "      <td>0.293768</td>\n",
       "      <td>0.170527</td>\n",
       "      <td>0.401522</td>\n",
       "      <td>0.341259</td>\n",
       "      <td>0.309981</td>\n",
       "      <td>0.366640</td>\n",
       "      <td>0.582397</td>\n",
       "      <td>0.504908</td>\n",
       "      <td>0.227994</td>\n",
       "      <td>0.090912</td>\n",
       "      <td>0.196414</td>\n",
       "      <td>0.329291</td>\n",
       "      <td>0.271887</td>\n",
       "      <td>0.118141</td>\n",
       "      <td>0.064196</td>\n",
       "      <td>0.277095</td>\n",
       "      <td>0.158241</td>\n",
       "      <td>0.199215</td>\n",
       "      <td>0.064070</td>\n",
       "      <td>0.467808</td>\n",
       "      <td>0.569490</td>\n",
       "      <td>0.485413</td>\n",
       "      <td>0.377568</td>\n",
       "      <td>0.035061</td>\n",
       "      <td>0.723149</td>\n",
       "      <td>0.087596</td>\n",
       "      <td>0.550299</td>\n",
       "      <td>0.126805</td>\n",
       "      <td>0.282151</td>\n",
       "      <td>0.073773</td>\n",
       "      <td>0.170839</td>\n",
       "      <td>0.117012</td>\n",
       "      <td>0.051851</td>\n",
       "      <td>0.261392</td>\n",
       "      <td>0.603740</td>\n",
       "      <td>0.489925</td>\n",
       "      <td>0.231491</td>\n",
       "      <td>0.019536</td>\n",
       "      <td>3.457069e-06</td>\n",
       "      <td>0.108441</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30133404</th>\n",
       "      <td>0.364922</td>\n",
       "      <td>0.276260</td>\n",
       "      <td>0.306559</td>\n",
       "      <td>0.437671</td>\n",
       "      <td>0.116419</td>\n",
       "      <td>0.422102</td>\n",
       "      <td>0.241749</td>\n",
       "      <td>0.471419</td>\n",
       "      <td>0.368978</td>\n",
       "      <td>0.385439</td>\n",
       "      <td>0.262283</td>\n",
       "      <td>0.109220</td>\n",
       "      <td>0.526896</td>\n",
       "      <td>0.236059</td>\n",
       "      <td>0.367466</td>\n",
       "      <td>0.166012</td>\n",
       "      <td>0.317508</td>\n",
       "      <td>0.211784</td>\n",
       "      <td>0.337888</td>\n",
       "      <td>0.468679</td>\n",
       "      <td>0.179275</td>\n",
       "      <td>0.366717</td>\n",
       "      <td>0.549959</td>\n",
       "      <td>0.423666</td>\n",
       "      <td>0.328701</td>\n",
       "      <td>0.125016</td>\n",
       "      <td>0.296373</td>\n",
       "      <td>0.324396</td>\n",
       "      <td>0.366027</td>\n",
       "      <td>0.153357</td>\n",
       "      <td>0.063499</td>\n",
       "      <td>0.412587</td>\n",
       "      <td>0.160946</td>\n",
       "      <td>0.245864</td>\n",
       "      <td>0.072131</td>\n",
       "      <td>0.660811</td>\n",
       "      <td>0.542503</td>\n",
       "      <td>0.437505</td>\n",
       "      <td>0.576502</td>\n",
       "      <td>0.132781</td>\n",
       "      <td>0.613927</td>\n",
       "      <td>0.161925</td>\n",
       "      <td>0.454536</td>\n",
       "      <td>0.120556</td>\n",
       "      <td>0.202739</td>\n",
       "      <td>0.078532</td>\n",
       "      <td>0.453060</td>\n",
       "      <td>0.548047</td>\n",
       "      <td>0.058342</td>\n",
       "      <td>0.471202</td>\n",
       "      <td>0.050729</td>\n",
       "      <td>0.809312</td>\n",
       "      <td>0.349704</td>\n",
       "      <td>0.079419</td>\n",
       "      <td>5.543232e-06</td>\n",
       "      <td>0.501348</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29333466</th>\n",
       "      <td>0.479236</td>\n",
       "      <td>0.271470</td>\n",
       "      <td>0.291775</td>\n",
       "      <td>0.390880</td>\n",
       "      <td>0.163174</td>\n",
       "      <td>0.349389</td>\n",
       "      <td>0.196093</td>\n",
       "      <td>0.463591</td>\n",
       "      <td>0.384706</td>\n",
       "      <td>0.446520</td>\n",
       "      <td>0.166840</td>\n",
       "      <td>0.240279</td>\n",
       "      <td>0.433335</td>\n",
       "      <td>0.220110</td>\n",
       "      <td>0.427684</td>\n",
       "      <td>0.176694</td>\n",
       "      <td>0.356380</td>\n",
       "      <td>0.164768</td>\n",
       "      <td>0.269165</td>\n",
       "      <td>0.434936</td>\n",
       "      <td>0.201477</td>\n",
       "      <td>0.407721</td>\n",
       "      <td>0.405988</td>\n",
       "      <td>0.481090</td>\n",
       "      <td>0.230533</td>\n",
       "      <td>0.118604</td>\n",
       "      <td>0.344441</td>\n",
       "      <td>0.297559</td>\n",
       "      <td>0.500397</td>\n",
       "      <td>0.116533</td>\n",
       "      <td>0.043100</td>\n",
       "      <td>0.290736</td>\n",
       "      <td>0.113262</td>\n",
       "      <td>0.239416</td>\n",
       "      <td>0.064534</td>\n",
       "      <td>0.505296</td>\n",
       "      <td>0.617156</td>\n",
       "      <td>0.442336</td>\n",
       "      <td>0.346974</td>\n",
       "      <td>0.069070</td>\n",
       "      <td>0.565067</td>\n",
       "      <td>0.250068</td>\n",
       "      <td>0.539528</td>\n",
       "      <td>0.098351</td>\n",
       "      <td>0.159062</td>\n",
       "      <td>0.323081</td>\n",
       "      <td>0.425882</td>\n",
       "      <td>0.208105</td>\n",
       "      <td>0.026306</td>\n",
       "      <td>0.221252</td>\n",
       "      <td>0.082105</td>\n",
       "      <td>0.748609</td>\n",
       "      <td>0.338905</td>\n",
       "      <td>0.064247</td>\n",
       "      <td>8.642673e-07</td>\n",
       "      <td>0.314960</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           LABEL_1   LABEL_2    ...         LABEL_55  LABEL_56\n",
       "ID                              ...                           \n",
       "29408483  0.375211  0.297849    ...     8.940697e-07  0.277034\n",
       "29164528  0.390474  0.273048    ...     1.311302e-06  0.356881\n",
       "29475634  0.387763  0.325422    ...     3.457069e-06  0.108441\n",
       "30133404  0.364922  0.276260    ...     5.543232e-06  0.501348\n",
       "29333466  0.479236  0.271470    ...     8.642673e-07  0.314960\n",
       "\n",
       "[5 rows x 56 columns]"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "submission_all.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "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>count</th>\n",
       "      <th>mean</th>\n",
       "      <th>std</th>\n",
       "      <th>min</th>\n",
       "      <th>25%</th>\n",
       "      <th>50%</th>\n",
       "      <th>75%</th>\n",
       "      <th>max</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>LABEL_1</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.406400</td>\n",
       "      <td>0.045888</td>\n",
       "      <td>0.250177</td>\n",
       "      <td>0.374624</td>\n",
       "      <td>0.405924</td>\n",
       "      <td>0.437251</td>\n",
       "      <td>0.581610</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_2</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.288898</td>\n",
       "      <td>0.022291</td>\n",
       "      <td>0.205882</td>\n",
       "      <td>0.273759</td>\n",
       "      <td>0.288714</td>\n",
       "      <td>0.303382</td>\n",
       "      <td>0.392557</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_3</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.279154</td>\n",
       "      <td>0.039231</td>\n",
       "      <td>0.144858</td>\n",
       "      <td>0.251981</td>\n",
       "      <td>0.276712</td>\n",
       "      <td>0.304075</td>\n",
       "      <td>0.460647</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_4</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.376491</td>\n",
       "      <td>0.046145</td>\n",
       "      <td>0.187167</td>\n",
       "      <td>0.345100</td>\n",
       "      <td>0.375631</td>\n",
       "      <td>0.407011</td>\n",
       "      <td>0.566920</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_5</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.135843</td>\n",
       "      <td>0.026826</td>\n",
       "      <td>0.048987</td>\n",
       "      <td>0.117050</td>\n",
       "      <td>0.133374</td>\n",
       "      <td>0.152032</td>\n",
       "      <td>0.291958</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_6</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.353397</td>\n",
       "      <td>0.040204</td>\n",
       "      <td>0.220310</td>\n",
       "      <td>0.325513</td>\n",
       "      <td>0.351337</td>\n",
       "      <td>0.380322</td>\n",
       "      <td>0.511328</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_7</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.253751</td>\n",
       "      <td>0.032973</td>\n",
       "      <td>0.137056</td>\n",
       "      <td>0.230861</td>\n",
       "      <td>0.252720</td>\n",
       "      <td>0.275267</td>\n",
       "      <td>0.391313</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_8</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.471640</td>\n",
       "      <td>0.052788</td>\n",
       "      <td>0.285626</td>\n",
       "      <td>0.435651</td>\n",
       "      <td>0.470578</td>\n",
       "      <td>0.507057</td>\n",
       "      <td>0.653889</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_9</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.387476</td>\n",
       "      <td>0.043070</td>\n",
       "      <td>0.241546</td>\n",
       "      <td>0.358402</td>\n",
       "      <td>0.388073</td>\n",
       "      <td>0.416723</td>\n",
       "      <td>0.548154</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_10</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.429143</td>\n",
       "      <td>0.039341</td>\n",
       "      <td>0.289500</td>\n",
       "      <td>0.401946</td>\n",
       "      <td>0.428572</td>\n",
       "      <td>0.455896</td>\n",
       "      <td>0.587543</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_11</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.237087</td>\n",
       "      <td>0.046153</td>\n",
       "      <td>0.093663</td>\n",
       "      <td>0.204837</td>\n",
       "      <td>0.234249</td>\n",
       "      <td>0.266824</td>\n",
       "      <td>0.516595</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_12</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.195578</td>\n",
       "      <td>0.033247</td>\n",
       "      <td>0.098151</td>\n",
       "      <td>0.172133</td>\n",
       "      <td>0.193382</td>\n",
       "      <td>0.216540</td>\n",
       "      <td>0.346705</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_13</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.469295</td>\n",
       "      <td>0.046261</td>\n",
       "      <td>0.315166</td>\n",
       "      <td>0.437483</td>\n",
       "      <td>0.467774</td>\n",
       "      <td>0.499855</td>\n",
       "      <td>0.656823</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_14</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.252574</td>\n",
       "      <td>0.040265</td>\n",
       "      <td>0.110482</td>\n",
       "      <td>0.224479</td>\n",
       "      <td>0.250478</td>\n",
       "      <td>0.278183</td>\n",
       "      <td>0.442464</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_15</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.446320</td>\n",
       "      <td>0.064306</td>\n",
       "      <td>0.236703</td>\n",
       "      <td>0.401646</td>\n",
       "      <td>0.445363</td>\n",
       "      <td>0.489158</td>\n",
       "      <td>0.691411</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_16</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.198345</td>\n",
       "      <td>0.037996</td>\n",
       "      <td>0.094464</td>\n",
       "      <td>0.171585</td>\n",
       "      <td>0.195215</td>\n",
       "      <td>0.220958</td>\n",
       "      <td>0.391935</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_17</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.324929</td>\n",
       "      <td>0.033643</td>\n",
       "      <td>0.203314</td>\n",
       "      <td>0.302321</td>\n",
       "      <td>0.324134</td>\n",
       "      <td>0.347469</td>\n",
       "      <td>0.464834</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_18</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.195217</td>\n",
       "      <td>0.037356</td>\n",
       "      <td>0.087694</td>\n",
       "      <td>0.168922</td>\n",
       "      <td>0.192863</td>\n",
       "      <td>0.219399</td>\n",
       "      <td>0.388433</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_19</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.355635</td>\n",
       "      <td>0.061783</td>\n",
       "      <td>0.149026</td>\n",
       "      <td>0.312614</td>\n",
       "      <td>0.353149</td>\n",
       "      <td>0.396222</td>\n",
       "      <td>0.599379</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_20</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.343960</td>\n",
       "      <td>0.059835</td>\n",
       "      <td>0.147329</td>\n",
       "      <td>0.302454</td>\n",
       "      <td>0.342169</td>\n",
       "      <td>0.383297</td>\n",
       "      <td>0.603300</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_21</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.227679</td>\n",
       "      <td>0.051983</td>\n",
       "      <td>0.091670</td>\n",
       "      <td>0.190322</td>\n",
       "      <td>0.223503</td>\n",
       "      <td>0.260145</td>\n",
       "      <td>0.475653</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_22</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.373958</td>\n",
       "      <td>0.048965</td>\n",
       "      <td>0.205743</td>\n",
       "      <td>0.340518</td>\n",
       "      <td>0.373789</td>\n",
       "      <td>0.406595</td>\n",
       "      <td>0.552902</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_23</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.527113</td>\n",
       "      <td>0.089893</td>\n",
       "      <td>0.178346</td>\n",
       "      <td>0.466561</td>\n",
       "      <td>0.528206</td>\n",
       "      <td>0.589416</td>\n",
       "      <td>0.841924</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_24</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.517069</td>\n",
       "      <td>0.085778</td>\n",
       "      <td>0.219530</td>\n",
       "      <td>0.459425</td>\n",
       "      <td>0.517337</td>\n",
       "      <td>0.576313</td>\n",
       "      <td>0.809878</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_25</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.270538</td>\n",
       "      <td>0.050734</td>\n",
       "      <td>0.107282</td>\n",
       "      <td>0.234959</td>\n",
       "      <td>0.269593</td>\n",
       "      <td>0.304408</td>\n",
       "      <td>0.458365</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_26</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.103267</td>\n",
       "      <td>0.027797</td>\n",
       "      <td>0.033670</td>\n",
       "      <td>0.083318</td>\n",
       "      <td>0.099742</td>\n",
       "      <td>0.119500</td>\n",
       "      <td>0.240186</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_27</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.300853</td>\n",
       "      <td>0.074224</td>\n",
       "      <td>0.086961</td>\n",
       "      <td>0.247023</td>\n",
       "      <td>0.295154</td>\n",
       "      <td>0.347686</td>\n",
       "      <td>0.595907</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_28</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.269567</td>\n",
       "      <td>0.056498</td>\n",
       "      <td>0.107803</td>\n",
       "      <td>0.229521</td>\n",
       "      <td>0.265691</td>\n",
       "      <td>0.305126</td>\n",
       "      <td>0.540192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_29</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.511839</td>\n",
       "      <td>0.079319</td>\n",
       "      <td>0.188092</td>\n",
       "      <td>0.459204</td>\n",
       "      <td>0.513155</td>\n",
       "      <td>0.566743</td>\n",
       "      <td>0.782685</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_30</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.163992</td>\n",
       "      <td>0.048980</td>\n",
       "      <td>0.041331</td>\n",
       "      <td>0.129141</td>\n",
       "      <td>0.158859</td>\n",
       "      <td>0.192675</td>\n",
       "      <td>0.443727</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_31</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.052541</td>\n",
       "      <td>0.014917</td>\n",
       "      <td>0.017107</td>\n",
       "      <td>0.041711</td>\n",
       "      <td>0.050893</td>\n",
       "      <td>0.061139</td>\n",
       "      <td>0.162113</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_32</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.352102</td>\n",
       "      <td>0.066464</td>\n",
       "      <td>0.140696</td>\n",
       "      <td>0.305269</td>\n",
       "      <td>0.350098</td>\n",
       "      <td>0.396685</td>\n",
       "      <td>0.627749</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_33</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.178654</td>\n",
       "      <td>0.061704</td>\n",
       "      <td>0.036675</td>\n",
       "      <td>0.134196</td>\n",
       "      <td>0.170777</td>\n",
       "      <td>0.214655</td>\n",
       "      <td>0.540586</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_34</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.213346</td>\n",
       "      <td>0.070531</td>\n",
       "      <td>0.034188</td>\n",
       "      <td>0.162665</td>\n",
       "      <td>0.206202</td>\n",
       "      <td>0.256593</td>\n",
       "      <td>0.597704</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_35</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.066629</td>\n",
       "      <td>0.017415</td>\n",
       "      <td>0.022213</td>\n",
       "      <td>0.054470</td>\n",
       "      <td>0.064530</td>\n",
       "      <td>0.076463</td>\n",
       "      <td>0.179736</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_36</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.560930</td>\n",
       "      <td>0.070944</td>\n",
       "      <td>0.296435</td>\n",
       "      <td>0.512677</td>\n",
       "      <td>0.561881</td>\n",
       "      <td>0.610722</td>\n",
       "      <td>0.800269</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_37</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.405197</td>\n",
       "      <td>0.105982</td>\n",
       "      <td>0.093662</td>\n",
       "      <td>0.329523</td>\n",
       "      <td>0.400734</td>\n",
       "      <td>0.476474</td>\n",
       "      <td>0.784070</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_38</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.441941</td>\n",
       "      <td>0.072889</td>\n",
       "      <td>0.184249</td>\n",
       "      <td>0.391146</td>\n",
       "      <td>0.441908</td>\n",
       "      <td>0.491383</td>\n",
       "      <td>0.708779</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_39</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.478475</td>\n",
       "      <td>0.089296</td>\n",
       "      <td>0.161826</td>\n",
       "      <td>0.415308</td>\n",
       "      <td>0.477588</td>\n",
       "      <td>0.540605</td>\n",
       "      <td>0.813953</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_40</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.063651</td>\n",
       "      <td>0.025608</td>\n",
       "      <td>0.013534</td>\n",
       "      <td>0.045057</td>\n",
       "      <td>0.059106</td>\n",
       "      <td>0.076844</td>\n",
       "      <td>0.232319</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_41</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.576723</td>\n",
       "      <td>0.096492</td>\n",
       "      <td>0.181733</td>\n",
       "      <td>0.512656</td>\n",
       "      <td>0.580478</td>\n",
       "      <td>0.644204</td>\n",
       "      <td>0.910294</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_42</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.176979</td>\n",
       "      <td>0.079788</td>\n",
       "      <td>0.023882</td>\n",
       "      <td>0.116809</td>\n",
       "      <td>0.164546</td>\n",
       "      <td>0.224620</td>\n",
       "      <td>0.748367</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_43</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.480036</td>\n",
       "      <td>0.163283</td>\n",
       "      <td>0.053858</td>\n",
       "      <td>0.359010</td>\n",
       "      <td>0.478272</td>\n",
       "      <td>0.597332</td>\n",
       "      <td>0.936384</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_44</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.133197</td>\n",
       "      <td>0.055230</td>\n",
       "      <td>0.022153</td>\n",
       "      <td>0.093024</td>\n",
       "      <td>0.123796</td>\n",
       "      <td>0.163031</td>\n",
       "      <td>0.451928</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_45</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.219766</td>\n",
       "      <td>0.073505</td>\n",
       "      <td>0.030381</td>\n",
       "      <td>0.167591</td>\n",
       "      <td>0.209987</td>\n",
       "      <td>0.263114</td>\n",
       "      <td>0.592338</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_46</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.094131</td>\n",
       "      <td>0.057241</td>\n",
       "      <td>0.003543</td>\n",
       "      <td>0.052353</td>\n",
       "      <td>0.081810</td>\n",
       "      <td>0.122305</td>\n",
       "      <td>0.542017</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_47</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.429800</td>\n",
       "      <td>0.155159</td>\n",
       "      <td>0.042035</td>\n",
       "      <td>0.316125</td>\n",
       "      <td>0.424096</td>\n",
       "      <td>0.536947</td>\n",
       "      <td>0.947973</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_48</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.414041</td>\n",
       "      <td>0.244557</td>\n",
       "      <td>0.001770</td>\n",
       "      <td>0.206732</td>\n",
       "      <td>0.387822</td>\n",
       "      <td>0.602776</td>\n",
       "      <td>0.994158</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_49</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.074734</td>\n",
       "      <td>0.045737</td>\n",
       "      <td>0.005558</td>\n",
       "      <td>0.042720</td>\n",
       "      <td>0.063373</td>\n",
       "      <td>0.094510</td>\n",
       "      <td>0.434175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_50</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.408766</td>\n",
       "      <td>0.140478</td>\n",
       "      <td>0.064753</td>\n",
       "      <td>0.305562</td>\n",
       "      <td>0.399708</td>\n",
       "      <td>0.504518</td>\n",
       "      <td>0.896508</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_51</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.218652</td>\n",
       "      <td>0.189561</td>\n",
       "      <td>0.000484</td>\n",
       "      <td>0.071273</td>\n",
       "      <td>0.158195</td>\n",
       "      <td>0.317888</td>\n",
       "      <td>0.968977</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_52</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.688537</td>\n",
       "      <td>0.139327</td>\n",
       "      <td>0.138190</td>\n",
       "      <td>0.602069</td>\n",
       "      <td>0.710454</td>\n",
       "      <td>0.792354</td>\n",
       "      <td>0.970756</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_53</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.384698</td>\n",
       "      <td>0.166347</td>\n",
       "      <td>0.033102</td>\n",
       "      <td>0.254528</td>\n",
       "      <td>0.367175</td>\n",
       "      <td>0.500003</td>\n",
       "      <td>0.906785</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_54</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.052367</td>\n",
       "      <td>0.050832</td>\n",
       "      <td>0.000348</td>\n",
       "      <td>0.019211</td>\n",
       "      <td>0.036662</td>\n",
       "      <td>0.067159</td>\n",
       "      <td>0.466283</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_55</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.000003</td>\n",
       "      <td>0.000003</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000001</td>\n",
       "      <td>0.000002</td>\n",
       "      <td>0.000004</td>\n",
       "      <td>0.000046</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LABEL_56</th>\n",
       "      <td>12721.0</td>\n",
       "      <td>0.236984</td>\n",
       "      <td>0.180290</td>\n",
       "      <td>0.002171</td>\n",
       "      <td>0.096070</td>\n",
       "      <td>0.186881</td>\n",
       "      <td>0.338849</td>\n",
       "      <td>0.950096</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            count      mean       std    ...          50%       75%       max\n",
       "LABEL_1   12721.0  0.406400  0.045888    ...     0.405924  0.437251  0.581610\n",
       "LABEL_2   12721.0  0.288898  0.022291    ...     0.288714  0.303382  0.392557\n",
       "LABEL_3   12721.0  0.279154  0.039231    ...     0.276712  0.304075  0.460647\n",
       "LABEL_4   12721.0  0.376491  0.046145    ...     0.375631  0.407011  0.566920\n",
       "LABEL_5   12721.0  0.135843  0.026826    ...     0.133374  0.152032  0.291958\n",
       "LABEL_6   12721.0  0.353397  0.040204    ...     0.351337  0.380322  0.511328\n",
       "LABEL_7   12721.0  0.253751  0.032973    ...     0.252720  0.275267  0.391313\n",
       "LABEL_8   12721.0  0.471640  0.052788    ...     0.470578  0.507057  0.653889\n",
       "LABEL_9   12721.0  0.387476  0.043070    ...     0.388073  0.416723  0.548154\n",
       "LABEL_10  12721.0  0.429143  0.039341    ...     0.428572  0.455896  0.587543\n",
       "LABEL_11  12721.0  0.237087  0.046153    ...     0.234249  0.266824  0.516595\n",
       "LABEL_12  12721.0  0.195578  0.033247    ...     0.193382  0.216540  0.346705\n",
       "LABEL_13  12721.0  0.469295  0.046261    ...     0.467774  0.499855  0.656823\n",
       "LABEL_14  12721.0  0.252574  0.040265    ...     0.250478  0.278183  0.442464\n",
       "LABEL_15  12721.0  0.446320  0.064306    ...     0.445363  0.489158  0.691411\n",
       "LABEL_16  12721.0  0.198345  0.037996    ...     0.195215  0.220958  0.391935\n",
       "LABEL_17  12721.0  0.324929  0.033643    ...     0.324134  0.347469  0.464834\n",
       "LABEL_18  12721.0  0.195217  0.037356    ...     0.192863  0.219399  0.388433\n",
       "LABEL_19  12721.0  0.355635  0.061783    ...     0.353149  0.396222  0.599379\n",
       "LABEL_20  12721.0  0.343960  0.059835    ...     0.342169  0.383297  0.603300\n",
       "LABEL_21  12721.0  0.227679  0.051983    ...     0.223503  0.260145  0.475653\n",
       "LABEL_22  12721.0  0.373958  0.048965    ...     0.373789  0.406595  0.552902\n",
       "LABEL_23  12721.0  0.527113  0.089893    ...     0.528206  0.589416  0.841924\n",
       "LABEL_24  12721.0  0.517069  0.085778    ...     0.517337  0.576313  0.809878\n",
       "LABEL_25  12721.0  0.270538  0.050734    ...     0.269593  0.304408  0.458365\n",
       "LABEL_26  12721.0  0.103267  0.027797    ...     0.099742  0.119500  0.240186\n",
       "LABEL_27  12721.0  0.300853  0.074224    ...     0.295154  0.347686  0.595907\n",
       "LABEL_28  12721.0  0.269567  0.056498    ...     0.265691  0.305126  0.540192\n",
       "LABEL_29  12721.0  0.511839  0.079319    ...     0.513155  0.566743  0.782685\n",
       "LABEL_30  12721.0  0.163992  0.048980    ...     0.158859  0.192675  0.443727\n",
       "LABEL_31  12721.0  0.052541  0.014917    ...     0.050893  0.061139  0.162113\n",
       "LABEL_32  12721.0  0.352102  0.066464    ...     0.350098  0.396685  0.627749\n",
       "LABEL_33  12721.0  0.178654  0.061704    ...     0.170777  0.214655  0.540586\n",
       "LABEL_34  12721.0  0.213346  0.070531    ...     0.206202  0.256593  0.597704\n",
       "LABEL_35  12721.0  0.066629  0.017415    ...     0.064530  0.076463  0.179736\n",
       "LABEL_36  12721.0  0.560930  0.070944    ...     0.561881  0.610722  0.800269\n",
       "LABEL_37  12721.0  0.405197  0.105982    ...     0.400734  0.476474  0.784070\n",
       "LABEL_38  12721.0  0.441941  0.072889    ...     0.441908  0.491383  0.708779\n",
       "LABEL_39  12721.0  0.478475  0.089296    ...     0.477588  0.540605  0.813953\n",
       "LABEL_40  12721.0  0.063651  0.025608    ...     0.059106  0.076844  0.232319\n",
       "LABEL_41  12721.0  0.576723  0.096492    ...     0.580478  0.644204  0.910294\n",
       "LABEL_42  12721.0  0.176979  0.079788    ...     0.164546  0.224620  0.748367\n",
       "LABEL_43  12721.0  0.480036  0.163283    ...     0.478272  0.597332  0.936384\n",
       "LABEL_44  12721.0  0.133197  0.055230    ...     0.123796  0.163031  0.451928\n",
       "LABEL_45  12721.0  0.219766  0.073505    ...     0.209987  0.263114  0.592338\n",
       "LABEL_46  12721.0  0.094131  0.057241    ...     0.081810  0.122305  0.542017\n",
       "LABEL_47  12721.0  0.429800  0.155159    ...     0.424096  0.536947  0.947973\n",
       "LABEL_48  12721.0  0.414041  0.244557    ...     0.387822  0.602776  0.994158\n",
       "LABEL_49  12721.0  0.074734  0.045737    ...     0.063373  0.094510  0.434175\n",
       "LABEL_50  12721.0  0.408766  0.140478    ...     0.399708  0.504518  0.896508\n",
       "LABEL_51  12721.0  0.218652  0.189561    ...     0.158195  0.317888  0.968977\n",
       "LABEL_52  12721.0  0.688537  0.139327    ...     0.710454  0.792354  0.970756\n",
       "LABEL_53  12721.0  0.384698  0.166347    ...     0.367175  0.500003  0.906785\n",
       "LABEL_54  12721.0  0.052367  0.050832    ...     0.036662  0.067159  0.466283\n",
       "LABEL_55  12721.0  0.000003  0.000003    ...     0.000002  0.000004  0.000046\n",
       "LABEL_56  12721.0  0.236984  0.180290    ...     0.186881  0.338849  0.950096\n",
       "\n",
       "[56 rows x 8 columns]"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "submission_all.describe().T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.0    12436\n",
      "1.0      285\n",
      "Name: LABEL_1, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_2, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_3, dtype: int64\n",
      "0.0    12652\n",
      "1.0       69\n",
      "Name: LABEL_4, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_5, dtype: int64\n",
      "0.0    12714\n",
      "1.0        7\n",
      "Name: LABEL_6, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_7, dtype: int64\n",
      "0.0    8968\n",
      "1.0    3753\n",
      "Name: LABEL_8, dtype: int64\n",
      "0.0    12663\n",
      "1.0       58\n",
      "Name: LABEL_9, dtype: int64\n",
      "0.0    12238\n",
      "1.0      483\n",
      "Name: LABEL_10, dtype: int64\n",
      "0.0    12720\n",
      "1.0        1\n",
      "Name: LABEL_11, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_12, dtype: int64\n",
      "0.0    9556\n",
      "1.0    3165\n",
      "Name: LABEL_13, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_14, dtype: int64\n",
      "0.0    10102\n",
      "1.0     2619\n",
      "Name: LABEL_15, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_16, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_17, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_18, dtype: int64\n",
      "0.0    12547\n",
      "1.0      174\n",
      "Name: LABEL_19, dtype: int64\n",
      "0.0    12635\n",
      "1.0       86\n",
      "Name: LABEL_20, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_21, dtype: int64\n",
      "0.0    12650\n",
      "1.0       71\n",
      "Name: LABEL_22, dtype: int64\n",
      "1.0    7868\n",
      "0.0    4853\n",
      "Name: LABEL_23, dtype: int64\n",
      "1.0    7385\n",
      "0.0    5336\n",
      "Name: LABEL_24, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_25, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_26, dtype: int64\n",
      "0.0    12621\n",
      "1.0      100\n",
      "Name: LABEL_27, dtype: int64\n",
      "0.0    12718\n",
      "1.0        3\n",
      "Name: LABEL_28, dtype: int64\n",
      "1.0    7152\n",
      "0.0    5569\n",
      "Name: LABEL_29, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_30, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_31, dtype: int64\n",
      "0.0    12524\n",
      "1.0      197\n",
      "Name: LABEL_32, dtype: int64\n",
      "0.0    12719\n",
      "1.0        2\n",
      "Name: LABEL_33, dtype: int64\n",
      "0.0    12710\n",
      "1.0       11\n",
      "Name: LABEL_34, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_35, dtype: int64\n",
      "1.0    10260\n",
      "0.0     2461\n",
      "Name: LABEL_36, dtype: int64\n",
      "0.0    10280\n",
      "1.0     2441\n",
      "Name: LABEL_37, dtype: int64\n",
      "0.0    9996\n",
      "1.0    2725\n",
      "Name: LABEL_38, dtype: int64\n",
      "0.0    7596\n",
      "1.0    5125\n",
      "Name: LABEL_39, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_40, dtype: int64\n",
      "1.0    9992\n",
      "0.0    2729\n",
      "Name: LABEL_41, dtype: int64\n",
      "0.0    12708\n",
      "1.0       13\n",
      "Name: LABEL_42, dtype: int64\n",
      "0.0    6965\n",
      "1.0    5756\n",
      "Name: LABEL_43, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_44, dtype: int64\n",
      "0.0    12695\n",
      "1.0       26\n",
      "Name: LABEL_45, dtype: int64\n",
      "0.0    12720\n",
      "1.0        1\n",
      "Name: LABEL_46, dtype: int64\n",
      "0.0    8528\n",
      "1.0    4193\n",
      "Name: LABEL_47, dtype: int64\n",
      "0.0    8111\n",
      "1.0    4610\n",
      "Name: LABEL_48, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_49, dtype: int64\n",
      "0.0    9429\n",
      "1.0    3292\n",
      "Name: LABEL_50, dtype: int64\n",
      "0.0    11418\n",
      "1.0     1303\n",
      "Name: LABEL_51, dtype: int64\n",
      "1.0    11365\n",
      "0.0     1356\n",
      "Name: LABEL_52, dtype: int64\n",
      "0.0    9539\n",
      "1.0    3182\n",
      "Name: LABEL_53, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_54, dtype: int64\n",
      "0.0    12721\n",
      "Name: LABEL_55, dtype: int64\n",
      "0.0    11408\n",
      "1.0     1313\n",
      "Name: LABEL_56, dtype: int64\n"
     ]
    }
   ],
   "source": [
    "for label in labels:\n",
    "    print(np.round(submission_all[label]).value_counts())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "submission_all.to_csv(\"submission_all.csv\")\n",
    "submission_all.to_csv(\"submission_all.txt\", sep=\";\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "230"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "K.clear_session()\n",
    "gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Tue Jun 18 09:18:03 2019       \r\n",
      "+-----------------------------------------------------------------------------+\r\n",
      "| NVIDIA-SMI 418.67       Driver Version: 418.67       CUDA Version: 10.1     |\r\n",
      "|-------------------------------+----------------------+----------------------+\r\n",
      "| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |\r\n",
      "| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |\r\n",
      "|===============================+======================+======================|\r\n",
      "|   0  Tesla P100-PCIE...  On   | 00000000:00:04.0 Off |                    0 |\r\n",
      "| N/A   43C    P0    38W / 250W |  15663MiB / 16280MiB |      0%      Default |\r\n",
      "+-------------------------------+----------------------+----------------------+\r\n",
      "                                                                               \r\n",
      "+-----------------------------------------------------------------------------+\r\n",
      "| Processes:                                                       GPU Memory |\r\n",
      "|  GPU       PID   Type   Process name                             Usage      |\r\n",
      "|=============================================================================|\r\n",
      "+-----------------------------------------------------------------------------+\r\n"
     ]
    }
   ],
   "source": [
    "!nvidia-smi"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  },
  "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": 1
}