Vowpal Wabbit 2. Models and hyperopt optimization.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import json\n",
    "import os\n",
    "import time\n",
    "\n",
    "import pandas as pd\n",
    "from tqdm import tqdm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Простой Baseline VW"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Вариант с модификацией dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Generating 2-grams for all namespaces.\n",
      "final_regressor = ..\\\\models\\\\logistic.model\n",
      "Num weight bits = 18\n",
      "learning rate = 0.5\n",
      "initial_t = 0\n",
      "power_t = 0.5\n",
      "decay_learning_rate = 1\n",
      "creating cache_file = ..\\\\models\\\\temp.cache\n",
      "Reading datafile = ..\\\\vw\\\\dataset.vw\n",
      "num sources = 1\n",
      "average  since         example        example  current  current  current\n",
      "loss     last          counter         weight    label  predict features\n",
      "0.693147 0.693147            1            1.0  -1.0000   0.0000       48\n",
      "0.491887 0.290626            2            2.0  -1.0000  -1.0869       46\n",
      "0.319663 0.147439            4            4.0  -1.0000  -1.5767       46\n",
      "0.217190 0.114716            8            8.0  -1.0000  -1.8737       35\n",
      "0.145423 0.073657           16           16.0  -1.0000  -3.6736      138\n",
      "0.093701 0.041980           32           32.0  -1.0000  -3.1650       64\n",
      "0.057883 0.022065           64           64.0  -1.0000  -3.8537       37\n",
      "0.035184 0.012486          128          128.0  -1.0000  -4.3918       50\n",
      "0.020033 0.004882          256          256.0  -1.0000  -7.3528       72\n",
      "0.011474 0.002914          512          512.0  -1.0000  -5.5860       44\n",
      "0.017894 0.024314         1024         1024.0  -1.0000  -5.4162       66\n",
      "0.013051 0.008208         2048         2048.0  -1.0000  -8.3397       76\n",
      "0.092708 0.172365         4096         4096.0  -1.0000  -2.2356       62\n",
      "0.166621 0.240533         8192         8192.0  -1.0000  -7.0231      104\n",
      "0.157742 0.148864        16384        16384.0  -1.0000  -3.3719       70\n",
      "0.212773 0.267804        32768        32768.0  -1.0000  -5.0034      108\n",
      "0.196676 0.180580        65536        65536.0   1.0000  -7.0104      190\n",
      "0.173643 0.150609       131072       131072.0  -1.0000  -4.4259       64\n",
      "0.183297 0.192952       262144       262144.0  -1.0000  -3.0329       42\n",
      "0.192292 0.201287       524288       524288.0  -1.0000  -3.4204       72\n",
      "0.192932 0.193572      1048576      1048576.0  -1.0000  -2.9463       32\n",
      "0.189649 0.186365      2097152      2097152.0  -1.0000  -3.8825      124\n",
      "0.185840 0.182032      4194304      4194304.0  -1.0000  -2.2543       48\n",
      "0.183060 0.180281      8388608      8388608.0  -1.0000  -3.3680      302\n",
      "0.177452 0.171843     16777216     16777216.0  -1.0000  -3.5474       40\n",
      "0.172729 0.168007     33554432     33554432.0  -1.0000  -5.1567      206\n",
      "\n",
      "finished run\n",
      "number of examples per pass = 12874345\n",
      "passes used = 3\n",
      "weighted example sum = 38623035.000000\n",
      "weighted label sum = -35205627.000000\n",
      "average loss = 0.172145\n",
      "best constant = -3.072865\n",
      "best constant's loss = 0.181194\n",
      "total feature number = 4171146213\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Wall time: 7min 41s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "!vw -d ..\\\\vw\\\\dataset.vw --loss_function logistic -f ..\\\\models\\\\logistic.model --passes 3 --ngram 2 --holdout_off --cache_file ..\\\\models\\\\cache.cache --random_seed 17"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "%%time\n",
    "\n",
    "# Предсказание на трейне\n",
    "!vw -i ..\\\\models\\\\logistic.model -t -d ..\\\\vw\\\\dataset_test.vw -p ..\\\\output\\\\pred_train_logistic_p3n2.txt --quiet"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Оптимизация параметров: wv-hyperopt.py\n",
    "\n",
    "Оригинальный скрипт wv-hyperopt.py доступен на github: https://github.com/JohnLangford/vowpal_wabbit/tree/master/utl\n",
    "\n",
    "Обучение модели на stratified shuffled данных 90% от подготовленного датасета. Валидация на 10% shuffled."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1.0527669957839534\n",
      "19.951239977624464\n"
     ]
    }
   ],
   "source": [
    "# Соотношение target и величины выборки\n",
    "\n",
    "print((21452 + 406542) / 406542)\n",
    "print((21452 + 406542) / 21452)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Команда для оптимизации параметров (запуск из командной строки)\n",
    "python vw-hyperopt.py --train \"..\\\\vw\\\\dataset_train_shuffled.vw\" --holdout \"..\\\\vw\\\\dataset_valid_shuffled.vw\" --max_evals 100 --outer_loss_function roc-auc --vw_space \"--algorithms=ftrl,sgd -b=28 --ngram=c2,c3,c4 --ftrl_alpha=5e-5..8e-1~L --ftrl_beta=0.01..1 --classweight=-1:0.02,-1:0.05,-1:0.1,-1:0.2,-1:0.5~O --l2=1e-8..1e-2~LO --l1=1e-8..1e-2~LO -l=0.01..1~L --decay_learning_rate=0.95,0.97~O --power_t=0.01..1 --passes=3..10~I --loss_function=logistic\" --plot"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Загрузить результаты оптимизации"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "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>current_trial</th>\n",
       "      <th>eval_time</th>\n",
       "      <th>loss</th>\n",
       "      <th>status</th>\n",
       "      <th>train_command</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>33</td>\n",
       "      <td>266</td>\n",
       "      <td>-0.588495</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>42</td>\n",
       "      <td>323</td>\n",
       "      <td>-0.588484</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>34</td>\n",
       "      <td>296</td>\n",
       "      <td>-0.588474</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>35</td>\n",
       "      <td>327</td>\n",
       "      <td>-0.588457</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>74</td>\n",
       "      <td>295</td>\n",
       "      <td>-0.588444</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>71</td>\n",
       "      <td>325</td>\n",
       "      <td>-0.588409</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>68</td>\n",
       "      <td>315</td>\n",
       "      <td>-0.588381</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>66</td>\n",
       "      <td>304</td>\n",
       "      <td>-0.588282</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>30</td>\n",
       "      <td>234</td>\n",
       "      <td>-0.588228</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>31</td>\n",
       "      <td>262</td>\n",
       "      <td>-0.588152</td>\n",
       "      <td>ok</td>\n",
       "      <td>vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   current_trial  eval_time      loss status  \\\n",
       "0             33        266 -0.588495     ok   \n",
       "1             42        323 -0.588484     ok   \n",
       "2             34        296 -0.588474     ok   \n",
       "3             35        327 -0.588457     ok   \n",
       "4             74        295 -0.588444     ok   \n",
       "5             71        325 -0.588409     ok   \n",
       "6             68        315 -0.588381     ok   \n",
       "7             66        304 -0.588282     ok   \n",
       "8             30        234 -0.588228     ok   \n",
       "9             31        262 -0.588152     ok   \n",
       "\n",
       "                                       train_command  \n",
       "0  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "1  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "2  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "3  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "4  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "5  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "6  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "7  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "8  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  \n",
       "9  vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./d...  "
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "with open('dataset_train_shuffled_trials.json') as f:\n",
    "    params = json.load(f)\n",
    "\n",
    "params_df = pd.DataFrame(params).sort_values(by=['loss'], axis=0).reset_index(drop=True)\n",
    "params_df.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0 : -0.588495297878427 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.21846519028073763 -b 28 -l 0.02203702669188238  \n",
      "\n",
      "1 : -0.5884835125943594 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.24095213922984413 -b 28 -l 0.021387359069161088  \n",
      "\n",
      "2 : -0.5884736863324582 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.20476209324788347 -b 28 -l 0.018642027897230323  \n",
      "\n",
      "3 : -0.5884571256956075 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.1831849568384751 -b 28 -l 0.01634983545699013  \n",
      "\n",
      "4 : -0.588444051831844 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.20814146307347117 -b 28 -l 0.024091422515916092  \n",
      "\n",
      "5 : -0.5884093288168382 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.14497596988811998 -b 28 -l 0.019657191516105843  \n",
      "\n",
      "6 : -0.5883812775564159 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.20149680553430757 -b 28 -l 0.025062418570621903  \n",
      "\n",
      "7 : -0.5882815962750034 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.24177372561159216 -b 28 -l 0.01736863256959812  \n",
      "\n",
      "8 : -0.5882283838764872 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 7 --power_t 0.29180043279288975 -b 28 -l 0.023935865127994706  \n",
      "\n",
      "9 : -0.5881517187241019 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.2898575684379044 -b 28 -l 0.02924828895162649  \n",
      "\n",
      "10 : -0.5881481766270108 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.025285905757748917 -b 28 -l 0.014900097451349247  \n",
      "\n",
      "11 : -0.5880005233365634 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.08514936136793526 -b 28 -l 0.014157515277616219  \n",
      "\n",
      "12 : -0.5879469179243516 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.10860148061149276 -b 28 -l 0.012403098950277372  \n",
      "\n",
      "13 : -0.5878966034877967 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.03953835434363273 -b 28 -l 0.010592679405592758  \n",
      "\n",
      "14 : -0.5877713551250897 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.2 --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.05652234736937908 -b 28 -l 0.012527003024277061  \n",
      "\n",
      "15 : -0.5877573233035202 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --loss_function logistic --ngram c4 --passes 8 --power_t 0.24434885616854482 -b 28 -l 0.013962818425046286  \n",
      "\n",
      "16 : -0.5873436382713828 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.30739938391747074 -b 28 -l 0.034908859376280264  \n",
      "\n",
      "17 : -0.5872789111378954 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.38013760536080476 -b 28 -l 0.027436203451276252  \n",
      "\n",
      "18 : -0.5872271131364835 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c4 --passes 10 --power_t 0.26442577496346165 -b 28 -l 0.035012977001781555  \n",
      "\n",
      "19 : -0.5871611261744824 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.1 --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.18738754707984845 -b 28 -l 0.051547486180887774  \n",
      "\n",
      "20 : -0.5870453061265862 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.05 --decay_learning_rate 0.97 --ftrl_alpha 0.001872155087451052 --ftrl_beta 0.4054045419565156 --l1 0.00023930199952916487 --l2 0.009380928295247282 --loss_function logistic --ngram c4 --passes 9 --power_t 0.6810215204406462 -b 28 -l 0.045617031436817085 --ftrl \n",
      "\n",
      "21 : -0.5868783347265762 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 6 --power_t 0.405649564225542 -b 28 -l 0.02768354442744159  \n",
      "\n",
      "22 : -0.5868178146770817 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --loss_function logistic --ngram c2 --passes 6 --power_t 0.3957582867671214 -b 28 -l 0.033936618321343585  \n",
      "\n",
      "23 : -0.5866872034813932 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.2791246814939323 -b 28 -l 0.012425548441977644  \n",
      "\n",
      "24 : -0.5864919199118523 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 7 --power_t 0.3478605041737807 -b 28 -l 0.015979426544082253  \n",
      "\n",
      "25 : -0.5862560164779965 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.12384017359384285 -b 28 -l 0.04064835509392274  \n",
      "\n",
      "26 : -0.5855115526067808 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 7 --power_t 0.3159023938177484 -b 28 -l 0.011653792539396177  \n",
      "\n",
      "27 : -0.5850539094946847 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.02 --decay_learning_rate 0.97 --l2 1.0251029225077212e-08 --loss_function logistic --ngram c3 --passes 9 --power_t 0.1898101360497031 -b 28 -l 0.040075736506296894  \n",
      "\n",
      "28 : -0.5844973024267267 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 9 --power_t 0.09628001425940946 -b 28 -l 0.05652000685736093  \n",
      "\n",
      "29 : -0.5842014922685198 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 7 --power_t 0.37393048431081194 -b 28 -l 0.011172265182211451  \n",
      "\n",
      "30 : -0.5841795806434075 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.23939335878883466 -b 28 -l 0.06577192765503319  \n",
      "\n",
      "31 : -0.5831263867790567 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c4 --passes 8 --power_t 0.457161176175725 -b 28 -l 0.045167336472782864  \n",
      "\n",
      "32 : -0.582911944589632 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --loss_function logistic --ngram c2 --passes 3 --power_t 0.2546919269269427 -b 28 -l 0.1573842376213395  \n",
      "\n",
      "33 : -0.5818087187827586 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 7 --power_t 0.4930582524015735 -b 28 -l 0.04531829038751307  \n",
      "\n",
      "34 : -0.5815192966702143 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c2 --passes 9 --power_t 0.4188257664044106 -b 28 -l 0.011012523045104897  \n",
      "\n",
      "35 : -0.5811872003402655 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c2 --passes 8 --power_t 0.3494693967985095 -b 28 -l 0.08065250371674984  \n",
      "\n",
      "36 : -0.5807251428184942 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 7 --power_t 0.45782731396534365 -b 28 -l 0.06561437025094013  \n",
      "\n",
      "37 : -0.5804802435559617 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l2 1.684340440928963e-06 --loss_function logistic --ngram c3 --passes 9 --power_t 0.21939521328450046 -b 28 -l 0.12996433545507266  \n",
      "\n",
      "38 : -0.5799483867055486 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c2 --passes 9 --power_t 0.12789330631442553 -b 28 -l 0.10105285647063288  \n",
      "\n",
      "39 : -0.57970409499788 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.5176222342797485 -b 28 -l 0.038935874434232715  \n",
      "\n",
      "40 : -0.5767980633464747 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c2 --passes 9 --power_t 0.013967036223819562 -b 28 -l 0.11477418736463273  \n",
      "\n",
      "41 : -0.5764318862747595 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.43416292050510996 -b 28 -l 0.08476941509897436  \n",
      "\n",
      "42 : -0.5749700724845269 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.1 --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.34165539411950446 -b 28 -l 0.2219391493041879  \n",
      "\n",
      "43 : -0.5746286343675865 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.02 --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 8 --power_t 0.17494702622018643 -b 28 -l 0.015042003339244683  \n",
      "\n",
      "44 : -0.5728750558579572 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 0.005766601975709161 --ftrl_beta 0.034560515236045725 --loss_function logistic --ngram c2 --passes 6 --power_t 0.9923174396516481 -b 28 -l 0.24092615319954455 --ftrl \n",
      "\n",
      "45 : -0.5725905807842274 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.4429926152839122 -b 28 -l 0.11123923643275216  \n",
      "\n",
      "46 : -0.572086457529982 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.5 --ftrl_alpha 0.010770800881443426 --ftrl_beta 0.0407042942486861 --loss_function logistic --ngram c3 --passes 7 --power_t 0.31922100463371794 -b 28 -l 0.5373812540177513 --ftrl \n",
      "\n",
      "47 : -0.5716396181601957 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.5 --decay_learning_rate 0.95 --ftrl_alpha 0.015313255386311045 --ftrl_beta 0.6057921391790789 --l1 7.810513850675035e-07 --l2 2.335806506856591e-07 --loss_function logistic --ngram c3 --passes 5 --power_t 0.18586125511992435 -b 28 -l 0.47133378791671754 --ftrl \n",
      "\n",
      "48 : -0.5691245520749405 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 10 --power_t 0.5918750152539038 -b 28 -l 0.021352788658365894  \n",
      "\n",
      "49 : -0.568758282784244 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 0.003955691981639907 --ftrl_beta 0.22330546070428198 --loss_function logistic --ngram c2 --passes 3 --power_t 0.9031042107193625 -b 28 -l 0.15307185522851788 --ftrl \n",
      "\n",
      "50 : -0.5685116562407879 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --ftrl_alpha 0.022740108767017753 --ftrl_beta 0.4752944475143615 --l1 6.032113018829296e-08 --loss_function logistic --ngram c4 --passes 4 --power_t 0.05582634968402449 -b 28 -l 0.34984279135140767 --ftrl \n",
      "\n",
      "51 : -0.5658580147435484 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 9 --power_t 0.06353187699332313 -b 28 -l 0.35193295581889106  \n",
      "\n",
      "52 : -0.5648632214973293 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 0.001675173409353584 --ftrl_beta 0.03821970706571509 --loss_function logistic --ngram c2 --passes 10 --power_t 0.992906799867689 -b 28 -l 0.0801968256434008 --ftrl \n",
      "\n",
      "53 : -0.5635143844383866 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --ftrl_alpha 0.0013708196576513087 --ftrl_beta 0.37972416623527516 --l1 0.008942146382366949 --loss_function logistic --ngram c4 --passes 8 --power_t 0.28087111563214573 -b 28 -l 0.0922579362915561 --ftrl \n",
      "\n",
      "54 : -0.5614259886413109 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l1 1.056357283113821e-08 --l2 8.54589373885923e-07 --loss_function logistic --ngram c4 --passes 8 --power_t 0.40929951488308197 -b 28 -l 0.02957224772736366  \n",
      "\n",
      "55 : -0.5611141509237233 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 7 --power_t 0.6576607150610057 -b 28 -l 0.03141670959514401  \n",
      "\n",
      "56 : -0.5611114471276203 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.5 --decay_learning_rate 0.95 --l1 3.09365337847231e-08 --loss_function logistic --ngram c4 --passes 6 --power_t 0.551094007676805 -b 28 -l 0.010931018260341537  \n",
      "\n",
      "57 : -0.5610120881801343 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --ftrl_alpha 0.041639964868060264 --ftrl_beta 0.1681284994010946 --loss_function logistic --ngram c2 --passes 6 --power_t 0.8537016132279989 -b 28 -l 0.2549047879787487 --ftrl \n",
      "\n",
      "58 : -0.5605082449268509 \n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --ftrl_alpha 0.05317310857441863 --ftrl_beta 0.7530902424172148 --l1 0.00044735977325157365 --l2 0.00018679054532353537 --loss_function logistic --ngram c4 --passes 6 --power_t 0.43719027984742387 -b 28 -l 0.04923834162529151 --ftrl \n",
      "\n",
      "59 : -0.5590694546059615 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 8 --power_t 0.15923635009010725 -b 28 -l 0.7485630313065545  \n",
      "\n",
      "60 : -0.5590441804498597 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.1 --decay_learning_rate 0.97 --ftrl_alpha 0.08318320418603552 --ftrl_beta 0.28460336913806034 --l1 3.72632294925409e-07 --l2 2.087464513560191e-06 --loss_function logistic --ngram c3 --passes 4 --power_t 0.7925417556353701 -b 28 -l 0.7799229530729391 --ftrl \n",
      "\n",
      "61 : -0.5584616518908202 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.02 --ftrl_alpha 0.0005825460708619398 --ftrl_beta 0.6247557485582909 --loss_function logistic --ngram c3 --passes 10 --power_t 0.11996892765908096 -b 28 -l 0.49908429492819784 --ftrl \n",
      "\n",
      "62 : -0.5577817301384531 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --ftrl_alpha 0.0007065532505484619 --ftrl_beta 0.46336112409017705 --l1 0.001473553618523203 --l2 0.00012808619846125022 --loss_function logistic --ngram c4 --passes 7 --power_t 0.3862584178358551 -b 28 -l 0.017266074133523004 --ftrl \n",
      "\n",
      "63 : -0.5563437717100607 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c4 --passes 5 --power_t 0.6293320558404352 -b 28 -l 0.01759084281417116  \n",
      "\n",
      "64 : -0.5532407293822491 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l2 1.3614101155902418e-07 --loss_function logistic --ngram c4 --passes 3 --power_t 0.5189989697934595 -b 28 -l 0.01000975030037332  \n",
      "\n",
      "65 : -0.5528394008338214 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.2 --decay_learning_rate 0.97 --ftrl_alpha 0.15443268478009956 --ftrl_beta 0.8554965309231739 --l1 4.5154852292759305e-05 --l2 0.007124958503574454 --loss_function logistic --ngram c4 --passes 8 --power_t 0.21528697891099258 -b 28 -l 0.16563380442838613 --ftrl \n",
      "\n",
      "66 : -0.551609882417726 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --loss_function logistic --ngram c4 --passes 4 --power_t 0.6276451108245144 -b 28 -l 0.24053288672430323  \n",
      "\n",
      "67 : -0.5483790978412809 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.1 --ftrl_alpha 0.2876190225888076 --ftrl_beta 0.583898236116813 --l1 1.7403433224014588e-08 --l2 7.111738926184657e-05 --loss_function logistic --ngram c2 --passes 8 --power_t 0.8286994693149228 -b 28 -l 0.017572553086718932 --ftrl \n",
      "\n",
      "68 : -0.5480556807728454 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.2 --decay_learning_rate 0.95 --l2 1.4730825958632005e-08 --loss_function logistic --ngram c3 --passes 10 --power_t 0.7437644360873538 -b 28 -l 0.024614418872495724  \n",
      "\n",
      "69 : -0.5477319817466165 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.2 --ftrl_alpha 0.3171205996841974 --ftrl_beta 0.9212627909856936 --l2 5.218499879787796e-07 --loss_function logistic --ngram c3 --passes 7 --power_t 0.5843825520680324 -b 28 -l 0.027508867413001906 --ftrl \n",
      "\n",
      "70 : -0.5434331158812155 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.05 --decay_learning_rate 0.95 --ftrl_alpha 0.7754703938035725 --ftrl_beta 0.9978658045419979 --l1 0.009787491041646865 --l2 3.71478553930924e-08 --loss_function logistic --ngram c3 --passes 9 --power_t 0.06779429657848607 -b 28 -l 0.0123685090575097 --ftrl \n",
      "\n",
      "71 : -0.5419433478809638 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.02 --decay_learning_rate 0.97 --ftrl_alpha 0.0003156171142847032 --ftrl_beta 0.3164933018933227 --l2 3.4150056889547415e-06 --loss_function logistic --ngram c4 --passes 3 --power_t 0.18802215272870884 -b 28 -l 0.01997585597664341 --ftrl \n",
      "\n",
      "72 : -0.5373042935122294 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 0.731656742503798 --ftrl_beta 0.6779802417247555 --loss_function logistic --ngram c2 --passes 3 --power_t 0.6572823356480731 -b 28 -l 0.04500444510146222 --ftrl \n",
      "\n",
      "73 : -0.536810930916596 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l1 9.166041281490713e-08 --loss_function logistic --ngram c2 --passes 6 --power_t 0.35512074495164847 -b 28 -l 0.3616954716511731  \n",
      "\n",
      "74 : -0.5358033277344816 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 0.00026814837810447287 --ftrl_beta 0.014895100263117182 --loss_function logistic --ngram c2 --passes 10 --power_t 0.6803645900768598 -b 28 -l 0.03201118151087164 --ftrl \n",
      "\n",
      "75 : -0.5352840125951535 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.02 --decay_learning_rate 0.95 --ftrl_alpha 0.00018254738060266738 --ftrl_beta 0.7303441417049671 --l1 7.702067776817233e-06 --l2 3.1468485169487825e-08 --loss_function logistic --ngram c3 --passes 5 --power_t 0.3685911762022702 -b 28 -l 0.01017718111432984 --ftrl \n",
      "\n",
      "76 : -0.5349859369552759 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l2 4.6402931377979654e-06 --loss_function logistic --ngram c2 --passes 8 --power_t 0.48331135141088544 -b 28 -l 0.6511858672311172  \n",
      "\n",
      "77 : -0.5344565001428168 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l1 1.6710017998807092e-07 --l2 6.989679230659089e-08 --loss_function logistic --ngram c3 --passes 7 --power_t 0.28059868831155504 -b 28 -l 0.0560122935075826  \n",
      "\n",
      "78 : -0.5340587924869137 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l1 0.001063410556094464 --l2 0.0001009024603828359 --loss_function logistic --ngram c4 --passes 9 --power_t 0.22215570667856216 -b 28 -l 0.0284513067777258  \n",
      "\n",
      "79 : -0.5340587924869137 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.5 --decay_learning_rate 0.97 --l1 7.055282328884895e-06 --loss_function logistic --ngram c3 --passes 5 --power_t 0.07176746477740147 -b 28 -l 0.03862244948351416  \n",
      "\n",
      "80 : -0.5340587924869137 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.2 --decay_learning_rate 0.97 --l1 4.7564355452834075e-06 --loss_function logistic --ngram c2 --passes 7 --power_t 0.016383692747511502 -b 28 -l 0.0187578624064178  \n",
      "\n",
      "81 : -0.5333082847420603 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l1 2.989159918034766e-05 --l2 0.00038500496616638625 --loss_function logistic --ngram c4 --passes 4 --power_t 0.10340194184401018 -b 28 -l 0.19341064390259585  \n",
      "\n",
      "82 : -0.5299105669545872 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.1 --l2 0.009317216454642913 --loss_function logistic --ngram c2 --passes 4 --power_t 0.3158498090487026 -b 28 -l 0.030762464106896807  \n",
      "\n",
      "83 : -0.5288374927494031 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.2 --ftrl_alpha 9.941677082367569e-05 --ftrl_beta 0.13183137651266663 --l1 4.618675555711119e-06 --l2 0.0010712069514469923 --loss_function logistic --ngram c4 --passes 4 --power_t 0.020318534306259017 -b 28 -l 0.13287191701091658 --ftrl \n",
      "\n",
      "84 : -0.524372443480521 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l2 3.173625402423554e-05 --loss_function logistic --ngram c3 --passes 8 --power_t 0.5487661994982259 -b 28 -l 0.0733261255228021  \n",
      "\n",
      "85 : -0.5197281573443193 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.5 --loss_function logistic --ngram c4 --passes 9 --power_t 0.949850065543957 -b 28 -l 0.9154397596307277  \n",
      "\n",
      "86 : -0.5187610660175086 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l1 1.582590733070878e-06 --l2 0.00603818746125023 --loss_function logistic --ngram c2 --passes 5 --power_t 0.5572514910304932 -b 28 -l 0.06725392388066517  \n",
      "\n",
      "87 : -0.5181511382167636 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.05 --l1 5.0678943226167675e-05 --l2 0.0006871568110834622 --loss_function logistic --ngram c4 --passes 10 --power_t 0.6787173259063891 -b 28 -l 0.2774333089319663  \n",
      "\n",
      "88 : -0.514935861287327 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l2 0.0007693089092783241 --loss_function logistic --ngram c2 --passes 5 --power_t 0.6070914960375023 -b 28 -l 0.5708912660500872  \n",
      "\n",
      "89 : -0.5132432363664636 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.05 --decay_learning_rate 0.95 --l1 1.9376950875434463e-07 --l2 1.4575214437395283e-07 --loss_function logistic --ngram c3 --passes 5 --power_t 0.9988452674933799 -b 28 -l 0.09142194255663456  \n",
      "\n",
      "90 : -0.5117526359814305 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l2 3.349224486273001e-05 --loss_function logistic --ngram c4 --passes 7 --power_t 0.7217465125760774 -b 28 -l 0.15558763697478797  \n",
      "\n",
      "91 : -0.5091392607650533 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 5.451602753333493e-05 --ftrl_beta 0.8113774705873762 --loss_function logistic --ngram c3 --passes 9 --power_t 0.5533630713882781 -b 28 -l 0.8613181811935814 --ftrl \n",
      "\n",
      "92 : -0.5080686715723649 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --ftrl_alpha 6.176657162741166e-05 --ftrl_beta 0.9469754750385859 --loss_function logistic --ngram c2 --passes 5 --power_t 0.5072387098134955 -b 28 -l 0.07162653595490462 --ftrl \n",
      "\n",
      "93 : -0.5018431809185445 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l1 8.449499375057239e-07 --loss_function logistic --ngram c3 --passes 6 --power_t 0.8619916323737731 -b 28 -l 0.0190951916679849  \n",
      "\n",
      "94 : -0.500005181557949 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.05 --decay_learning_rate 0.97 --l1 0.00016072488332563074 --loss_function logistic --ngram c2 --passes 9 --power_t 0.8665622435720766 -b 28 -l 0.05676260418977031  \n",
      "\n",
      "95 : -0.5 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l1 0.007160739465582116 --loss_function logistic --ngram c3 --passes 10 --power_t 0.5634931901814855 -b 28 -l 0.020913071720757105  \n",
      "\n",
      "96 : -0.5 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --classweight -1:0.02 --decay_learning_rate 0.95 --l1 0.0002586802278751734 --loss_function logistic --ngram c3 --passes 9 --power_t 0.12898862229790714 -b 28 -l 0.04716163067776662  \n",
      "\n",
      "97 : -0.5 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l1 0.0017574613398887572 --loss_function logistic --ngram c3 --passes 9 --power_t 0.12642074973309927 -b 28 -l 0.49428591242387127  \n",
      "\n",
      "98 : -0.5 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --l1 0.0013545437523239685 --loss_function logistic --ngram c4 --passes 4 --power_t 0.8258700427151362 -b 28 -l 0.26303465006007215  \n",
      "\n",
      "99 : -0.5 \n",
      " vw -d ..\\\\vw\\\\dataset_train_shuffled.vw -f ./dataset_train_shuffled_current.model --holdout_off -c  --decay_learning_rate 0.95 --l1 0.007534940806083406 --loss_function logistic --ngram c3 --passes 6 --power_t 0.1591119193909727 -b 28 -l 0.021833448681936412  \n",
      "\n"
     ]
    }
   ],
   "source": [
    "for row, command in params_df.iterrows():\n",
    "    print(row, ':', command['loss'], \"\\n\", command['train_command'], '\\n')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Модели с оптимизированными параметрами\n",
    "\n",
    "Выбрать некоторые модели, обучить и сохранить результаты предсказаний"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Learning command (1/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h1.model --holdout_off -c  --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.24095213922984413 -b 28 -l 0.021387359069161088 \n",
      "Learning success. Elapsed: 430.04 sec.\n",
      "Predicting command (1/10): vw -i ../models/dataset_shuffled_h1.model -t -d ../vw/dataset_test.vw -p ../output/h_1.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_1.txt. Elapsed: 59.19 sec.\n",
      "Learning command (14/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h14.model --holdout_off -c  --classweight -1:0.2 --decay_learning_rate 0.95 --loss_function logistic --ngram c3 --passes 10 --power_t 0.05652234736937908 -b 28 -l 0.012527003024277061 \n",
      "Learning success. Elapsed: 1020.83 sec.\n",
      "Predicting command (14/10): vw -i ../models/dataset_shuffled_h14.model -t -d ../vw/dataset_test.vw -p ../output/h_14.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_14.txt. Elapsed: 72.85 sec.\n",
      "Learning command (15/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h15.model --holdout_off -c  --loss_function logistic --ngram c4 --passes 8 --power_t 0.24434885616854482 -b 28 -l 0.013962818425046286 \n",
      "Learning success. Elapsed: 429.02 sec.\n",
      "Predicting command (15/10): vw -i ../models/dataset_shuffled_h15.model -t -d ../vw/dataset_test.vw -p ../output/h_15.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_15.txt. Elapsed: 61.01 sec.\n",
      "Learning command (20/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h20.model --holdout_off -c  --classweight -1:0.05 --decay_learning_rate 0.97 --ftrl_alpha 0.001872155087451052 --ftrl_beta 0.4054045419565156 --l1 0.00023930199952916487 --l2 0.009380928295247282 --loss_function logistic --ngram c4 --passes 9 --power_t 0.6810215204406462 -b 28 -l 0.045617031436817085 --ftrl\n",
      "Learning success. Elapsed: 356.08 sec.\n",
      "Predicting command (20/10): vw -i ../models/dataset_shuffled_h20.model -t -d ../vw/dataset_test.vw -p ../output/h_20.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_20.txt. Elapsed: 64.66 sec.\n",
      "Learning command (21/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h21.model --holdout_off -c  --decay_learning_rate 0.97 --loss_function logistic --ngram c3 --passes 6 --power_t 0.405649564225542 -b 28 -l 0.02768354442744159 \n",
      "Learning success. Elapsed: 409.72 sec.\n",
      "Predicting command (21/10): vw -i ../models/dataset_shuffled_h21.model -t -d ../vw/dataset_test.vw -p ../output/h_21.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_21.txt. Elapsed: 56.27 sec.\n",
      "Learning command (27/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h27.model --holdout_off -c  --classweight -1:0.02 --decay_learning_rate 0.97 --l2 1.0251029225077212e-08 --loss_function logistic --ngram c3 --passes 9 --power_t 0.1898101360497031 -b 28 -l 0.040075736506296894 \n",
      "Learning success. Elapsed: 1480.16 sec.\n",
      "Predicting command (27/10): vw -i ../models/dataset_shuffled_h27.model -t -d ../vw/dataset_test.vw -p ../output/h_27.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_27.txt. Elapsed: 56.9 sec.\n",
      "Learning command (44/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h44.model --holdout_off -c  --ftrl_alpha 0.005766601975709161 --ftrl_beta 0.034560515236045725 --loss_function logistic --ngram c2 --passes 6 --power_t 0.9923174396516481 -b 28 -l 0.24092615319954455 --ftrl\n",
      "Learning success. Elapsed: 259.37 sec.\n",
      "Predicting command (44/10): vw -i ../models/dataset_shuffled_h44.model -t -d ../vw/dataset_test.vw -p ../output/h_44.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_44.txt. Elapsed: 58.82 sec.\n",
      "Learning command (46/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h46.model --holdout_off -c  --classweight -1:0.5 --ftrl_alpha 0.010770800881443426 --ftrl_beta 0.0407042942486861 --loss_function logistic --ngram c3 --passes 7 --power_t 0.31922100463371794 -b 28 -l 0.5373812540177513 --ftrl\n",
      "Learning success. Elapsed: 276.73 sec.\n",
      "Predicting command (46/10): vw -i ../models/dataset_shuffled_h46.model -t -d ../vw/dataset_test.vw -p ../output/h_46.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_46.txt. Elapsed: 55.66 sec.\n",
      "Learning command (56/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h56.model --holdout_off -c  --classweight -1:0.5 --decay_learning_rate 0.95 --l1 3.09365337847231e-08 --loss_function logistic --ngram c4 --passes 6 --power_t 0.551094007676805 -b 28 -l 0.010931018260341537 \n",
      "Learning success. Elapsed: 995.25 sec.\n",
      "Predicting command (56/10): vw -i ../models/dataset_shuffled_h56.model -t -d ../vw/dataset_test.vw -p ../output/h_56.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_56.txt. Elapsed: 64.18 sec.\n",
      "Learning command (57/10): vw -d ../vw/dataset_shuffled.vw -f ../models/dataset_shuffled_h57.model --holdout_off -c  --decay_learning_rate 0.95 --ftrl_alpha 0.041639964868060264 --ftrl_beta 0.1681284994010946 --loss_function logistic --ngram c2 --passes 6 --power_t 0.8537016132279989 -b 28 -l 0.2549047879787487 --ftrl\n",
      "Learning success. Elapsed: 895.15 sec.\n",
      "Predicting command (57/10): vw -i ../models/dataset_shuffled_h57.model -t -d ../vw/dataset_test.vw -p ../output/h_57.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/h_57.txt. Elapsed: 64.81 sec.\n"
     ]
    }
   ],
   "source": [
    "# Выбранные модели для обучения\n",
    "# models_n = list(range(15,25))\n",
    "models_n = [1, 14, 15, 20, 21, 27, 44, 46, 56, 57]\n",
    "\n",
    "data_source = '../vw/dataset_shuffled.vw'\n",
    "\n",
    "for command, num in zip(params_df['train_command'][models_n], models_n):\n",
    "    init_time = time.time()\n",
    "    \n",
    "    model_name = '../models/dataset_shuffled_h' + str(num) + '.model'\n",
    "    \n",
    "    # --------------\n",
    "    # Обучить модель\n",
    "    # --------------\n",
    "    vw_call_string = ' '.join(['vw', '-d', data_source, '-f', model_name, command.split('model ')[1]])\n",
    "    \n",
    "    print(f'Learning command ({num}/{len(models_n)}): {vw_call_string}')\n",
    "    res = os.system(vw_call_string)\n",
    "    \n",
    "    print(f'Learning success. Elapsed: {round(time.time() - init_time, 2)} sec.' \n",
    "          if not res else 'Failed.')\n",
    "    \n",
    "    # ------------\n",
    "    # Предсказания\n",
    "    # ------------\n",
    "    time.sleep(2)\n",
    "    init_time = time.time()\n",
    "    \n",
    "    # Название файла с предсказаниями\n",
    "    pred_file = '../output/h_' + str(num) + '.txt'\n",
    "    \n",
    "    vw_pred_string = f'vw -i {model_name} -t -d ../vw/dataset_test.vw -p {pred_file} --quiet'\n",
    "    print(f'Predicting command ({num}/{len(models_n)}): {vw_pred_string}')\n",
    "    res = os.system(vw_pred_string)\n",
    "    \n",
    "    print(f'Predicting success. Wrote predictions to {pred_file}. Elapsed: {round(time.time() - init_time, 2)} sec.'\n",
    "          if not res else 'Failed.')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Оценки вероятности и submissions"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Загрузить все предсказания в один датафрейм"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "models_n = [1, 14, 15, 20, 21, 27, 44, 46, 56, 57]\n",
    "\n",
    "# Загрузить начальный df\n",
    "file_pred = '..\\\\output\\\\' + 'h_' + str(models_n[0]) + '.txt'\n",
    "pred_df = pd.read_csv(file_pred, header=None, names=['day', 'cat_feature', 'cuid'], index_col=None, sep=' |_', engine='python', usecols=[1,2,3])\n",
    "\n",
    "for nmodel in models_n:\n",
    "    file_pred = '..\\\\output\\\\' + 'h_' + str(nmodel) + '.txt'\n",
    "    temp_df = pd.read_csv(file_pred, header=None, names=['pred' + str(nmodel)], index_col=None, sep=' ', usecols=[0])\n",
    "    pred_df = pd.concat([pred_df, temp_df], axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "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>day</th>\n",
       "      <th>cat_feature</th>\n",
       "      <th>cuid</th>\n",
       "      <th>pred1</th>\n",
       "      <th>pred14</th>\n",
       "      <th>pred15</th>\n",
       "      <th>pred20</th>\n",
       "      <th>pred21</th>\n",
       "      <th>pred27</th>\n",
       "      <th>pred44</th>\n",
       "      <th>pred46</th>\n",
       "      <th>pred56</th>\n",
       "      <th>pred57</th>\n",
       "      <th>ave</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cuid</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>000014fe918d1f97a632a796f4948be8</th>\n",
       "      <td>day7</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "      <td>-2.796509</td>\n",
       "      <td>-1.196943</td>\n",
       "      <td>-2.777098</td>\n",
       "      <td>0.281049</td>\n",
       "      <td>-2.841886</td>\n",
       "      <td>1.158517</td>\n",
       "      <td>-2.485964</td>\n",
       "      <td>-1.887757</td>\n",
       "      <td>-2.337493</td>\n",
       "      <td>-3.302210</td>\n",
       "      <td>-1.818629</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>000014fe918d1f97a632a796f4948be8</th>\n",
       "      <td>day8</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "      <td>-3.207107</td>\n",
       "      <td>-1.606038</td>\n",
       "      <td>-3.157062</td>\n",
       "      <td>-0.528776</td>\n",
       "      <td>-3.178761</td>\n",
       "      <td>0.700543</td>\n",
       "      <td>-3.915512</td>\n",
       "      <td>-3.713463</td>\n",
       "      <td>-2.282472</td>\n",
       "      <td>-5.892112</td>\n",
       "      <td>-2.678076</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                   day  cat_feature  \\\n",
       "cuid                                                  \n",
       "000014fe918d1f97a632a796f4948be8  day7            2   \n",
       "000014fe918d1f97a632a796f4948be8  day8            2   \n",
       "\n",
       "                                                              cuid     pred1  \\\n",
       "cuid                                                                           \n",
       "000014fe918d1f97a632a796f4948be8  000014fe918d1f97a632a796f4948be8 -2.796509   \n",
       "000014fe918d1f97a632a796f4948be8  000014fe918d1f97a632a796f4948be8 -3.207107   \n",
       "\n",
       "                                    pred14    pred15    pred20    pred21  \\\n",
       "cuid                                                                       \n",
       "000014fe918d1f97a632a796f4948be8 -1.196943 -2.777098  0.281049 -2.841886   \n",
       "000014fe918d1f97a632a796f4948be8 -1.606038 -3.157062 -0.528776 -3.178761   \n",
       "\n",
       "                                    pred27    pred44    pred46    pred56  \\\n",
       "cuid                                                                       \n",
       "000014fe918d1f97a632a796f4948be8  1.158517 -2.485964 -1.887757 -2.337493   \n",
       "000014fe918d1f97a632a796f4948be8  0.700543 -3.915512 -3.713463 -2.282472   \n",
       "\n",
       "                                    pred57       ave  \n",
       "cuid                                                  \n",
       "000014fe918d1f97a632a796f4948be8 -3.302210 -1.818629  \n",
       "000014fe918d1f97a632a796f4948be8 -5.892112 -2.678076  "
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_df['ave'] = pred_df[['pred'+str(x) for x in models_n]].mean(axis=1)\n",
    "pred_df.index = pred_df['cuid']\n",
    "pred_df.head(2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Усредненный submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|█████████████████████████████████| 181024/181024 [03:11<00:00, 945.82it/s]\n"
     ]
    },
    {
     "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>ave</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cuid</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>888b238b4d14c03173baa375a739f6bc</th>\n",
       "      <td>-2.17446</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ac4b8244f3ae82df511b002257473c11</th>\n",
       "      <td>-2.16608</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>483d8b91e49522c8a5bbe37f3872c749</th>\n",
       "      <td>-1.72051</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4c7ec46a0e88a7e1e1cedd2d526d5d61</th>\n",
       "      <td>-2.28438</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>fdbfba9842ff0bf86d600eb334c7c42b</th>\n",
       "      <td>-2.19152</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                      ave\n",
       "cuid                                     \n",
       "888b238b4d14c03173baa375a739f6bc -2.17446\n",
       "ac4b8244f3ae82df511b002257473c11 -2.16608\n",
       "483d8b91e49522c8a5bbe37f3872c749 -1.72051\n",
       "4c7ec46a0e88a7e1e1cedd2d526d5d61 -2.28438\n",
       "fdbfba9842ff0bf86d600eb334c7c42b -2.19152"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Последовательность cuids для теста\n",
    "cuid_test = pd.read_csv('..\\\\input\\\\mlboot_test.tsv', index_col=[0])\n",
    "\n",
    "submission = pd.DataFrame(columns=['ave'], index=cuid_test.index)\n",
    "\n",
    "for cuid in tqdm(submission.index):\n",
    "    submission['ave'][cuid] = out_df['ave'][cuid].mean()\n",
    "\n",
    "submission.to_csv('..\\\\output\\\\baseline3_h_total_ave' + str(models_n) + '.txt', header=None, index=None)\n",
    "\n",
    "submission.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Предсказания vw на dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predicting command (1/10): vw -i ../models/dataset_shuffled_h1.model -t -d ../vw/dataset.vw -p ../output/dataset_h_1.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_1.txt. Elapsed: 122.6 sec.\n",
      "Predicting command (14/10): vw -i ../models/dataset_shuffled_h14.model -t -d ../vw/dataset.vw -p ../output/dataset_h_14.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_14.txt. Elapsed: 92.44 sec.\n",
      "Predicting command (15/10): vw -i ../models/dataset_shuffled_h15.model -t -d ../vw/dataset.vw -p ../output/dataset_h_15.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_15.txt. Elapsed: 92.72 sec.\n",
      "Predicting command (20/10): vw -i ../models/dataset_shuffled_h20.model -t -d ../vw/dataset.vw -p ../output/dataset_h_20.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_20.txt. Elapsed: 91.81 sec.\n",
      "Predicting command (21/10): vw -i ../models/dataset_shuffled_h21.model -t -d ../vw/dataset.vw -p ../output/dataset_h_21.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_21.txt. Elapsed: 91.38 sec.\n",
      "Predicting command (27/10): vw -i ../models/dataset_shuffled_h27.model -t -d ../vw/dataset.vw -p ../output/dataset_h_27.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_27.txt. Elapsed: 90.95 sec.\n",
      "Predicting command (44/10): vw -i ../models/dataset_shuffled_h44.model -t -d ../vw/dataset.vw -p ../output/dataset_h_44.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_44.txt. Elapsed: 91.76 sec.\n",
      "Predicting command (46/10): vw -i ../models/dataset_shuffled_h46.model -t -d ../vw/dataset.vw -p ../output/dataset_h_46.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_46.txt. Elapsed: 94.77 sec.\n",
      "Predicting command (56/10): vw -i ../models/dataset_shuffled_h56.model -t -d ../vw/dataset.vw -p ../output/dataset_h_56.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_56.txt. Elapsed: 94.65 sec.\n",
      "Predicting command (57/10): vw -i ../models/dataset_shuffled_h57.model -t -d ../vw/dataset.vw -p ../output/dataset_h_57.txt --quiet\n",
      "Predicting success. Wrote predictions to ../output/dataset_h_57.txt. Elapsed: 103.18 sec.\n"
     ]
    }
   ],
   "source": [
    "models_n = [1, 14, 15, 20, 21, 27, 44, 46, 56, 57]\n",
    "\n",
    "for command, num in zip(params_df['train_command'][models_n], models_n):\n",
    "    # ------------\n",
    "    # Предсказания\n",
    "    # ------------\n",
    "    init_time = time.time()\n",
    "    \n",
    "    model_name = '../models/dataset_shuffled_h' + str(num) + '.model'\n",
    "    \n",
    "    # Название файла с предсказаниями\n",
    "    pred_file = '../output/dataset_h_' + str(num) + '.txt'\n",
    "    \n",
    "    vw_pred_string = f'vw -i {model_name} -t -d ../vw/dataset.vw -p {pred_file} --quiet'\n",
    "    print(f'Predicting command ({num}/{len(models_n)}): {vw_pred_string}')\n",
    "    res = os.system(vw_pred_string)\n",
    "    \n",
    "    print(f'Predicting success. Wrote predictions to {pred_file}. Elapsed: {round(time.time() - init_time, 2)} sec.'\n",
    "          if not res else 'Failed.')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "models_n = [1, 14, 15, 20, 21, 27, 44, 46, 56, 57]\n",
    "\n",
    "# Загрузить начальный df\n",
    "file_pred = '..\\\\output\\\\dataset_' + 'h_' + str(models_n[0]) + '.txt'\n",
    "pred_df = pd.read_csv(file_pred, header=None, names=['day', 'cat_feature', 'cuid'], index_col=None, sep=' |_', engine='python', usecols=[1,2,3])\n",
    "\n",
    "for nmodel in models_n:\n",
    "    file_pred = '..\\\\output\\\\' + 'h_' + str(nmodel) + '.txt'\n",
    "    temp_df = pd.read_csv(file_pred, header=None, names=['pred' + str(nmodel)], index_col=None, sep=' ', usecols=[0])\n",
    "    pred_df = pd.concat([pred_df, temp_df], axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "##### Дополнительные признаки"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "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>day</th>\n",
       "      <th>cat_feature</th>\n",
       "      <th>cuid</th>\n",
       "      <th>pred1</th>\n",
       "      <th>pred14</th>\n",
       "      <th>pred15</th>\n",
       "      <th>pred20</th>\n",
       "      <th>pred21</th>\n",
       "      <th>pred27</th>\n",
       "      <th>pred44</th>\n",
       "      <th>pred46</th>\n",
       "      <th>pred56</th>\n",
       "      <th>pred57</th>\n",
       "      <th>ave</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cuid</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>00000d2994b6df9239901389031acaac</th>\n",
       "      <td>day39</td>\n",
       "      <td>5</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "      <td>-2.796509</td>\n",
       "      <td>-1.196943</td>\n",
       "      <td>-2.777098</td>\n",
       "      <td>0.281049</td>\n",
       "      <td>-2.841886</td>\n",
       "      <td>1.158517</td>\n",
       "      <td>-2.485964</td>\n",
       "      <td>-1.887757</td>\n",
       "      <td>-2.337493</td>\n",
       "      <td>-3.302210</td>\n",
       "      <td>-1.818629</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>00000d2994b6df9239901389031acaac</th>\n",
       "      <td>day43</td>\n",
       "      <td>5</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "      <td>-3.207107</td>\n",
       "      <td>-1.606038</td>\n",
       "      <td>-3.157062</td>\n",
       "      <td>-0.528776</td>\n",
       "      <td>-3.178761</td>\n",
       "      <td>0.700543</td>\n",
       "      <td>-3.915512</td>\n",
       "      <td>-3.713463</td>\n",
       "      <td>-2.282472</td>\n",
       "      <td>-5.892112</td>\n",
       "      <td>-2.678076</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                    day  cat_feature  \\\n",
       "cuid                                                   \n",
       "00000d2994b6df9239901389031acaac  day39            5   \n",
       "00000d2994b6df9239901389031acaac  day43            5   \n",
       "\n",
       "                                                              cuid     pred1  \\\n",
       "cuid                                                                           \n",
       "00000d2994b6df9239901389031acaac  00000d2994b6df9239901389031acaac -2.796509   \n",
       "00000d2994b6df9239901389031acaac  00000d2994b6df9239901389031acaac -3.207107   \n",
       "\n",
       "                                    pred14    pred15    pred20    pred21  \\\n",
       "cuid                                                                       \n",
       "00000d2994b6df9239901389031acaac -1.196943 -2.777098  0.281049 -2.841886   \n",
       "00000d2994b6df9239901389031acaac -1.606038 -3.157062 -0.528776 -3.178761   \n",
       "\n",
       "                                    pred27    pred44    pred46    pred56  \\\n",
       "cuid                                                                       \n",
       "00000d2994b6df9239901389031acaac  1.158517 -2.485964 -1.887757 -2.337493   \n",
       "00000d2994b6df9239901389031acaac  0.700543 -3.915512 -3.713463 -2.282472   \n",
       "\n",
       "                                    pred57       ave  \n",
       "cuid                                                  \n",
       "00000d2994b6df9239901389031acaac -3.302210 -1.818629  \n",
       "00000d2994b6df9239901389031acaac -5.892112 -2.678076  "
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Среднее значение\n",
    "pred_df['ave'] = pred_df[['pred'+str(x) for x in models_n]].mean(axis=1)\n",
    "pred_df.index = pred_df['cuid']\n",
    "pred_df.head(2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Признак день\n",
    "\n",
    "days = pred_df['day'].unique().tolist()\n",
    "days = [f'day{int(day[3:]):02d}' for day in days]\n",
    "days.sort()\n",
    "\n",
    "days_df = pd.DataFrame(0, index=pred_df.index, columns=days)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Загрузить данные"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>pred</th>\n",
       "      <th>day</th>\n",
       "      <th>cat_feature</th>\n",
       "      <th>cuid</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.281049</td>\n",
       "      <td>day7</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-0.528776</td>\n",
       "      <td>day8</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.061917</td>\n",
       "      <td>day11</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.203633</td>\n",
       "      <td>day13</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.170326</td>\n",
       "      <td>day14</td>\n",
       "      <td>2</td>\n",
       "      <td>000014fe918d1f97a632a796f4948be8</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       pred    day  cat_feature                              cuid\n",
       "0  0.281049   day7            2  000014fe918d1f97a632a796f4948be8\n",
       "1 -0.528776   day8            2  000014fe918d1f97a632a796f4948be8\n",
       "2  0.061917  day11            2  000014fe918d1f97a632a796f4948be8\n",
       "3  0.203633  day13            2  000014fe918d1f97a632a796f4948be8\n",
       "4  0.170326  day14            2  000014fe918d1f97a632a796f4948be8"
      ]
     },
     "execution_count": 58,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "file_pred = '..\\\\output\\\\h_20.txt'\n",
    "pred_df = pd.read_csv(file_pred, header=None, names=['pred', 'day', 'cat_feature', 'cuid'], index_col=None, sep=' |_', engine='python')\n",
    "pred_df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((6654252, 4), (6654252, 1))"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "file_cuids = '..\\\\vw\\\\dataset_cuid_test.vw'\n",
    "file_test = '..\\\\input\\\\mlboot_test.tsv'\n",
    "\n",
    "cuid_df = pd.read_csv(file_cuids, index_col=None, names=['cuid'], usecols=[0], header=None, sep='\\t')\n",
    "cuid_test = pd.read_csv(file_test, index_col=[0])\n",
    "\n",
    "pred_df.shape, cuid_df.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Использование мультииндекса\n",
    "#### Создание датафрейма"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "До/после удаления дубликатов в мультииндексе: (6654252, 2) (5542237, 2)\n"
     ]
    },
    {
     "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>day01</th>\n",
       "      <th>day02</th>\n",
       "      <th>day03</th>\n",
       "      <th>day04</th>\n",
       "      <th>day05</th>\n",
       "      <th>day06</th>\n",
       "      <th>day07</th>\n",
       "      <th>day08</th>\n",
       "      <th>day09</th>\n",
       "      <th>day10</th>\n",
       "      <th>...</th>\n",
       "      <th>day51</th>\n",
       "      <th>day52</th>\n",
       "      <th>day53</th>\n",
       "      <th>day54</th>\n",
       "      <th>day55</th>\n",
       "      <th>day56</th>\n",
       "      <th>day57</th>\n",
       "      <th>day58</th>\n",
       "      <th>day59</th>\n",
       "      <th>day60</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cuid</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>888b238b4d14c03173baa375a739f6bc</th>\n",
       "      <td>-5.006658</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-2.521204</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-2.771118</td>\n",
       "      <td>-2.221654</td>\n",
       "      <td>-2.577213</td>\n",
       "      <td>-2.700551</td>\n",
       "      <td>-2.849910</td>\n",
       "      <td>-2.786111</td>\n",
       "      <td>...</td>\n",
       "      <td>-3.347989</td>\n",
       "      <td>-2.716497</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-3.245458</td>\n",
       "      <td>-2.544358</td>\n",
       "      <td>-2.828236</td>\n",
       "      <td>-2.735087</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ac4b8244f3ae82df511b002257473c11</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>483d8b91e49522c8a5bbe37f3872c749</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-2.768052</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3 rows × 60 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                     day01  day02     day03  day04     day05  \\\n",
       "cuid                                                                           \n",
       "888b238b4d14c03173baa375a739f6bc -5.006658    NaN -2.521204    NaN -2.771118   \n",
       "ac4b8244f3ae82df511b002257473c11       NaN    NaN       NaN    NaN       NaN   \n",
       "483d8b91e49522c8a5bbe37f3872c749       NaN    NaN       NaN    NaN       NaN   \n",
       "\n",
       "                                     day06     day07     day08     day09  \\\n",
       "cuid                                                                       \n",
       "888b238b4d14c03173baa375a739f6bc -2.221654 -2.577213 -2.700551 -2.849910   \n",
       "ac4b8244f3ae82df511b002257473c11       NaN       NaN       NaN       NaN   \n",
       "483d8b91e49522c8a5bbe37f3872c749       NaN       NaN       NaN -2.768052   \n",
       "\n",
       "                                     day10  ...       day51     day52  day53  \\\n",
       "cuid                                        ...                                \n",
       "888b238b4d14c03173baa375a739f6bc -2.786111  ...   -3.347989 -2.716497    NaN   \n",
       "ac4b8244f3ae82df511b002257473c11       NaN  ...         NaN       NaN    NaN   \n",
       "483d8b91e49522c8a5bbe37f3872c749       NaN  ...         NaN       NaN    NaN   \n",
       "\n",
       "                                     day54     day55     day56     day57  \\\n",
       "cuid                                                                       \n",
       "888b238b4d14c03173baa375a739f6bc -3.245458 -2.544358 -2.828236 -2.735087   \n",
       "ac4b8244f3ae82df511b002257473c11       NaN       NaN       NaN       NaN   \n",
       "483d8b91e49522c8a5bbe37f3872c749       NaN       NaN       NaN       NaN   \n",
       "\n",
       "                                  day58  day59  day60  \n",
       "cuid                                                   \n",
       "888b238b4d14c03173baa375a739f6bc    NaN    NaN    NaN  \n",
       "ac4b8244f3ae82df511b002257473c11    NaN    NaN    NaN  \n",
       "483d8b91e49522c8a5bbe37f3872c749    NaN    NaN    NaN  \n",
       "\n",
       "[3 rows x 60 columns]"
      ]
     },
     "execution_count": 86,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "multi_index = pd.MultiIndex.from_tuples([(x, y) for x, y in zip(cuid_df['cuid'], pred_df['day'])])\n",
    "\n",
    "out_df_multi = pred_df\n",
    "out_df_multi.index = multi_index\n",
    "\n",
    "# Убрать дубликаты в индексе уровня 1\n",
    "out_df_multi = out_df_multi[~out_df_multi.index.duplicated(keep='first')]\n",
    "print('До/после удаления дубликатов в мультииндексе:', pred_df.shape, out_df_multi.shape)\n",
    "\n",
    "# Развернуть датафрейм в столбцы\n",
    "out_df_multi = out_df_multi['pred'].unstack(level=1)\n",
    "\n",
    "# Отсортировать индексы как задано в test\n",
    "out_df_multi = out_df_multi.reindex(cuid_out.index)\n",
    "\n",
    "# Переименовать столбцы и упорядочить\n",
    "names = dict([(f'day{x}', f'day{x:02d}') for x in range(1, 10)])\n",
    "out_df_multi = out_df_multi.rename(columns=names)\n",
    "out_df_multi = out_df_multi.sort_index(axis=1)\n",
    "\n",
    "out_df_multi.head(3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Усредненный submission"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Усредненный ответ\n",
    "out_df_multi.mean(axis=1).to_csv('..\\\\output\\\\baseline2_h_1.txt', header=None, index=None)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Усреднение предсказания без мультииндекса (без удаления дубликатов дней)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|████████████████████████████████| 181024/181024 [02:59<00:00, 1009.33it/s]\n"
     ]
    },
    {
     "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>ave</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>cuid</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>888b238b4d14c03173baa375a739f6bc</th>\n",
       "      <td>-0.188752</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ac4b8244f3ae82df511b002257473c11</th>\n",
       "      <td>-0.140454</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>483d8b91e49522c8a5bbe37f3872c749</th>\n",
       "      <td>0.113673</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4c7ec46a0e88a7e1e1cedd2d526d5d61</th>\n",
       "      <td>-0.244325</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>fdbfba9842ff0bf86d600eb334c7c42b</th>\n",
       "      <td>-0.0791256</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                        ave\n",
       "cuid                                       \n",
       "888b238b4d14c03173baa375a739f6bc  -0.188752\n",
       "ac4b8244f3ae82df511b002257473c11  -0.140454\n",
       "483d8b91e49522c8a5bbe37f3872c749   0.113673\n",
       "4c7ec46a0e88a7e1e1cedd2d526d5d61  -0.244325\n",
       "fdbfba9842ff0bf86d600eb334c7c42b -0.0791256"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out_df = pred_df.copy()\n",
    "out_df.index = cuid_df['cuid']\n",
    "\n",
    "submission = pd.DataFrame(columns=['ave'], index=cuid_test.index)\n",
    "\n",
    "for cuid in tqdm(submission.index):\n",
    "    submission['ave'][cuid] = out_df['pred'][cuid].mean()\n",
    "\n",
    "submission.to_csv('..\\\\output\\\\baseline3_h_20_ave.txt', header=None, index=None)\n",
    "\n",
    "submission.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Организация нового датасета из предсказаний test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((12874345, 2), (12874345, 1))"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_file = '..\\\\output\\\\pred_train_logistic_p3n2.txt'\n",
    "data_cuids = '..\\\\vw\\\\dataset_cuid.vw'\n",
    "data = '..\\\\input\\\\mlboot_test.tsv'\n",
    "\n",
    "input_df = pd.read_csv(pred_file, header=None, names=['pred', 'day'], index_col=None, sep=' ')\n",
    "cuid_df = pd.read_csv(data_cuids, index_col=None, names=['cuid'], usecols=[0], header=None, sep='\\t')\n",
    "cuid_out = pd.read_csv(data, index_col=[0])\n",
    "\n",
    "input_df.shape, cuid_df.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "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>pred</th>\n",
       "      <th>day</th>\n",
       "      <th>cuid</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>-3.148156</td>\n",
       "      <td>day39</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>-3.447425</td>\n",
       "      <td>day43</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>-3.372123</td>\n",
       "      <td>day45</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-3.472649</td>\n",
       "      <td>day46</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>-4.678955</td>\n",
       "      <td>day48</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>-3.520002</td>\n",
       "      <td>day53</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>-2.640258</td>\n",
       "      <td>day55</td>\n",
       "      <td>00000d2994b6df9239901389031acaac</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>-3.146274</td>\n",
       "      <td>day14</td>\n",
       "      <td>00000d53c06c43379ed22c775ce89614</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>-3.201166</td>\n",
       "      <td>day15</td>\n",
       "      <td>00000d53c06c43379ed22c775ce89614</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>-3.265105</td>\n",
       "      <td>day43</td>\n",
       "      <td>00000e36ecd94e5c54ab9d08c7b29163</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>-2.591345</td>\n",
       "      <td>day4</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>-2.515418</td>\n",
       "      <td>day6</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>-3.244754</td>\n",
       "      <td>day7</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>-3.365814</td>\n",
       "      <td>day10</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>-2.807720</td>\n",
       "      <td>day11</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>-3.400484</td>\n",
       "      <td>day12</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>-2.934219</td>\n",
       "      <td>day12</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>-2.907989</td>\n",
       "      <td>day13</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>-2.619267</td>\n",
       "      <td>day54</td>\n",
       "      <td>0000115cdd110ffa16d31dff81cf6272</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>-3.952002</td>\n",
       "      <td>day1</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>20</th>\n",
       "      <td>-2.490668</td>\n",
       "      <td>day6</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>21</th>\n",
       "      <td>-3.492429</td>\n",
       "      <td>day11</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>22</th>\n",
       "      <td>-3.088424</td>\n",
       "      <td>day12</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>23</th>\n",
       "      <td>-3.287211</td>\n",
       "      <td>day18</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <td>-3.456377</td>\n",
       "      <td>day20</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>-3.179842</td>\n",
       "      <td>day27</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>-3.111802</td>\n",
       "      <td>day28</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>-3.762144</td>\n",
       "      <td>day37</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>28</th>\n",
       "      <td>-3.912033</td>\n",
       "      <td>day42</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>29</th>\n",
       "      <td>-3.255645</td>\n",
       "      <td>day43</td>\n",
       "      <td>00006025b25e293ed4faa6c998f86180</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874315</th>\n",
       "      <td>-1.967516</td>\n",
       "      <td>day28</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874316</th>\n",
       "      <td>-2.557135</td>\n",
       "      <td>day29</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874317</th>\n",
       "      <td>-2.765466</td>\n",
       "      <td>day30</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874318</th>\n",
       "      <td>-1.913393</td>\n",
       "      <td>day31</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874319</th>\n",
       "      <td>-2.524762</td>\n",
       "      <td>day32</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874320</th>\n",
       "      <td>-2.112404</td>\n",
       "      <td>day34</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874321</th>\n",
       "      <td>-1.912868</td>\n",
       "      <td>day35</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874322</th>\n",
       "      <td>-3.332249</td>\n",
       "      <td>day36</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874323</th>\n",
       "      <td>-2.870097</td>\n",
       "      <td>day38</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874324</th>\n",
       "      <td>-3.033377</td>\n",
       "      <td>day40</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874325</th>\n",
       "      <td>-3.093041</td>\n",
       "      <td>day41</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874326</th>\n",
       "      <td>-3.161501</td>\n",
       "      <td>day42</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874327</th>\n",
       "      <td>-3.077222</td>\n",
       "      <td>day45</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874328</th>\n",
       "      <td>-1.311057</td>\n",
       "      <td>day46</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874329</th>\n",
       "      <td>-2.817848</td>\n",
       "      <td>day48</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874330</th>\n",
       "      <td>-2.618330</td>\n",
       "      <td>day50</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874331</th>\n",
       "      <td>-2.890488</td>\n",
       "      <td>day51</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874332</th>\n",
       "      <td>-3.038561</td>\n",
       "      <td>day52</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874333</th>\n",
       "      <td>-2.585695</td>\n",
       "      <td>day53</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874334</th>\n",
       "      <td>-2.677032</td>\n",
       "      <td>day54</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874335</th>\n",
       "      <td>-2.831249</td>\n",
       "      <td>day55</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874336</th>\n",
       "      <td>-2.926475</td>\n",
       "      <td>day56</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874337</th>\n",
       "      <td>-2.779369</td>\n",
       "      <td>day58</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874338</th>\n",
       "      <td>-2.303522</td>\n",
       "      <td>day59</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874339</th>\n",
       "      <td>-2.228293</td>\n",
       "      <td>day60</td>\n",
       "      <td>ffffe2043daeeb8e37e37532252ee2f8</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874340</th>\n",
       "      <td>-3.328243</td>\n",
       "      <td>day8</td>\n",
       "      <td>fffff60e3e8720d7ea3431cdc336baf5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874341</th>\n",
       "      <td>-3.321253</td>\n",
       "      <td>day9</td>\n",
       "      <td>fffff60e3e8720d7ea3431cdc336baf5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874342</th>\n",
       "      <td>-2.914036</td>\n",
       "      <td>day12</td>\n",
       "      <td>fffff60e3e8720d7ea3431cdc336baf5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874343</th>\n",
       "      <td>-3.114801</td>\n",
       "      <td>day18</td>\n",
       "      <td>fffff60e3e8720d7ea3431cdc336baf5</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12874344</th>\n",
       "      <td>-3.146065</td>\n",
       "      <td>day23</td>\n",
       "      <td>fffff60e3e8720d7ea3431cdc336baf5</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>12874345 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "              pred    day                              cuid\n",
       "0        -3.148156  day39  00000d2994b6df9239901389031acaac\n",
       "1        -3.447425  day43  00000d2994b6df9239901389031acaac\n",
       "2        -3.372123  day45  00000d2994b6df9239901389031acaac\n",
       "3        -3.472649  day46  00000d2994b6df9239901389031acaac\n",
       "4        -4.678955  day48  00000d2994b6df9239901389031acaac\n",
       "5        -3.520002  day53  00000d2994b6df9239901389031acaac\n",
       "6        -2.640258  day55  00000d2994b6df9239901389031acaac\n",
       "7        -3.146274  day14  00000d53c06c43379ed22c775ce89614\n",
       "8        -3.201166  day15  00000d53c06c43379ed22c775ce89614\n",
       "9        -3.265105  day43  00000e36ecd94e5c54ab9d08c7b29163\n",
       "10       -2.591345   day4  0000115cdd110ffa16d31dff81cf6272\n",
       "11       -2.515418   day6  0000115cdd110ffa16d31dff81cf6272\n",
       "12       -3.244754   day7  0000115cdd110ffa16d31dff81cf6272\n",
       "13       -3.365814  day10  0000115cdd110ffa16d31dff81cf6272\n",
       "14       -2.807720  day11  0000115cdd110ffa16d31dff81cf6272\n",
       "15       -3.400484  day12  0000115cdd110ffa16d31dff81cf6272\n",
       "16       -2.934219  day12  0000115cdd110ffa16d31dff81cf6272\n",
       "17       -2.907989  day13  0000115cdd110ffa16d31dff81cf6272\n",
       "18       -2.619267  day54  0000115cdd110ffa16d31dff81cf6272\n",
       "19       -3.952002   day1  00006025b25e293ed4faa6c998f86180\n",
       "20       -2.490668   day6  00006025b25e293ed4faa6c998f86180\n",
       "21       -3.492429  day11  00006025b25e293ed4faa6c998f86180\n",
       "22       -3.088424  day12  00006025b25e293ed4faa6c998f86180\n",
       "23       -3.287211  day18  00006025b25e293ed4faa6c998f86180\n",
       "24       -3.456377  day20  00006025b25e293ed4faa6c998f86180\n",
       "25       -3.179842  day27  00006025b25e293ed4faa6c998f86180\n",
       "26       -3.111802  day28  00006025b25e293ed4faa6c998f86180\n",
       "27       -3.762144  day37  00006025b25e293ed4faa6c998f86180\n",
       "28       -3.912033  day42  00006025b25e293ed4faa6c998f86180\n",
       "29       -3.255645  day43  00006025b25e293ed4faa6c998f86180\n",
       "...            ...    ...                               ...\n",
       "12874315 -1.967516  day28  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874316 -2.557135  day29  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874317 -2.765466  day30  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874318 -1.913393  day31  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874319 -2.524762  day32  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874320 -2.112404  day34  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874321 -1.912868  day35  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874322 -3.332249  day36  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874323 -2.870097  day38  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874324 -3.033377  day40  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874325 -3.093041  day41  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874326 -3.161501  day42  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874327 -3.077222  day45  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874328 -1.311057  day46  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874329 -2.817848  day48  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874330 -2.618330  day50  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874331 -2.890488  day51  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874332 -3.038561  day52  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874333 -2.585695  day53  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874334 -2.677032  day54  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874335 -2.831249  day55  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874336 -2.926475  day56  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874337 -2.779369  day58  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874338 -2.303522  day59  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874339 -2.228293  day60  ffffe2043daeeb8e37e37532252ee2f8\n",
       "12874340 -3.328243   day8  fffff60e3e8720d7ea3431cdc336baf5\n",
       "12874341 -3.321253   day9  fffff60e3e8720d7ea3431cdc336baf5\n",
       "12874342 -2.914036  day12  fffff60e3e8720d7ea3431cdc336baf5\n",
       "12874343 -3.114801  day18  fffff60e3e8720d7ea3431cdc336baf5\n",
       "12874344 -3.146065  day23  fffff60e3e8720d7ea3431cdc336baf5\n",
       "\n",
       "[12874345 rows x 3 columns]"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "input_df['cuid'] = cuid_df['cuid']\n",
    "input_df"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "# input_df.set_index(['cuid', 'day'])\n",
    "input_df.index = pd.MultiIndex.from_tuples([(x, y) for x, y in zip(input_df['cuid'], input_df['day'])])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((12874345, 3), (11446181, 3))"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Убрать дубликаты в индексе уровня 1\n",
    "input_df_dr = input_df[~input_df.index.duplicated(keep='first')]\n",
    "input_df.shape, input_df_dr.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(427994, 60)\n"
     ]
    },
    {
     "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>day1</th>\n",
       "      <th>day10</th>\n",
       "      <th>day11</th>\n",
       "      <th>day12</th>\n",
       "      <th>day13</th>\n",
       "      <th>day14</th>\n",
       "      <th>day15</th>\n",
       "      <th>day16</th>\n",
       "      <th>day17</th>\n",
       "      <th>day18</th>\n",
       "      <th>...</th>\n",
       "      <th>day55</th>\n",
       "      <th>day56</th>\n",
       "      <th>day57</th>\n",
       "      <th>day58</th>\n",
       "      <th>day59</th>\n",
       "      <th>day6</th>\n",
       "      <th>day60</th>\n",
       "      <th>day7</th>\n",
       "      <th>day8</th>\n",
       "      <th>day9</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>00000d2994b6df9239901389031acaac</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>-2.640258</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>00000d53c06c43379ed22c775ce89614</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>-3.146274</td>\n",
       "      <td>-3.201166</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>00000e36ecd94e5c54ab9d08c7b29163</th>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>3 rows × 60 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                                  day1  day10  day11  day12  day13     day14  \\\n",
       "00000d2994b6df9239901389031acaac   NaN    NaN    NaN    NaN    NaN       NaN   \n",
       "00000d53c06c43379ed22c775ce89614   NaN    NaN    NaN    NaN    NaN -3.146274   \n",
       "00000e36ecd94e5c54ab9d08c7b29163   NaN    NaN    NaN    NaN    NaN       NaN   \n",
       "\n",
       "                                     day15  day16  day17  day18  ...   \\\n",
       "00000d2994b6df9239901389031acaac       NaN    NaN    NaN    NaN  ...    \n",
       "00000d53c06c43379ed22c775ce89614 -3.201166    NaN    NaN    NaN  ...    \n",
       "00000e36ecd94e5c54ab9d08c7b29163       NaN    NaN    NaN    NaN  ...    \n",
       "\n",
       "                                     day55  day56  day57  day58  day59  day6  \\\n",
       "00000d2994b6df9239901389031acaac -2.640258    NaN    NaN    NaN    NaN   NaN   \n",
       "00000d53c06c43379ed22c775ce89614       NaN    NaN    NaN    NaN    NaN   NaN   \n",
       "00000e36ecd94e5c54ab9d08c7b29163       NaN    NaN    NaN    NaN    NaN   NaN   \n",
       "\n",
       "                                  day60  day7  day8  day9  \n",
       "00000d2994b6df9239901389031acaac    NaN   NaN   NaN   NaN  \n",
       "00000d53c06c43379ed22c775ce89614    NaN   NaN   NaN   NaN  \n",
       "00000e36ecd94e5c54ab9d08c7b29163    NaN   NaN   NaN   NaN  \n",
       "\n",
       "[3 rows x 60 columns]"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out_df = input_df_dr['pred'].unstack(level=1)\n",
    "\n",
    "# Отсортировать индексы как задано в test\n",
    "# out_df = out_df.reindex(cuid_out.index)\n",
    "\n",
    "print(out_df.shape)\n",
    "out_df.head(3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "427994"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "out_df.index.nunique()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}