research.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from lightgbm import LGBMClassifier\n",
    "from sklearn.model_selection import cross_validate, cross_val_score\n",
    "import matplotlib.pyplot as plt\n",
    "import cv2\n",
    "from tqdm import tqdm\n",
    "import json\n",
    "from scipy.ndimage import rotate\n",
    "import copy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Gold by reason "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reasons = ['0', '6', '2', '5', '13', '1', '12', '15', '11', '14']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "columns_gold = ['reason_0', 'reason_6', 'reason_2',\n",
    "                'reason_5', 'reason_13','reason_1',\n",
    "                'reason_12', 'reason_15', 'reason_11',\n",
    "                'reason_14']\n",
    "\n",
    "def get_gold_reason(lines, placeholder=-1, n=None):\n",
    "    \n",
    "    if n == None:\n",
    "        n = len(lines)\n",
    "        \n",
    "    data = []\n",
    "    for elem in lines:\n",
    "        gold_reason = elem['gold_by_reason']\n",
    "        data.append(gold_reason)\n",
    "        \n",
    "    return data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "train_paths = ['sberbank_data/final/academy2019_final_train.jsonlines']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40403/40403 [01:10<00:00, 574.01it/s]\n"
     ]
    }
   ],
   "source": [
    "buff = []\n",
    "ids = []\n",
    "\n",
    "data_reasons = [] \n",
    "\n",
    "total = 40403\n",
    "batch_size = 5000\n",
    "for path in train_paths:\n",
    "    with open(path) as inp:\n",
    "        for i, line in tqdm(enumerate(inp), total=total):\n",
    "            record = json.loads(line)\n",
    "            buff.append(record)\n",
    "            ids.append(record['id'])\n",
    "\n",
    "            if (i + 1) % batch_size == 0:\n",
    "                # extract features\n",
    "                data_reasons += get_gold_reason(buff)\n",
    "\n",
    "                # free buff\n",
    "                buff = []\n",
    "\n",
    "        # process last batch\n",
    "        data_reasons += get_gold_reason(buff)\n",
    "\n",
    "        # free buff\n",
    "        buff = []"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reasons DF"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "reasons_df = pd.DataFrame(data_reasons).fillna(-100000)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Player pos "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def scale_data(x):\n",
    "    \"\"\"\n",
    "    Приводит данные к [0, 1]\n",
    "    \"\"\"\n",
    "    max_x = x.max()\n",
    "    min_x = x.min()\n",
    "    \n",
    "    x = (x - min_x) / (max_x - min_x)\n",
    "    return x"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# читаем данные"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40403/40403 [00:56<00:00, 708.99it/s]"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "max_pos_x: 2147483647  |  max_pos_y: 2147483647\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "poisitions_x = []\n",
    "poisitions_y = []\n",
    "\n",
    "skilled = []\n",
    "record_ids = []\n",
    "\n",
    "max_pos_x = 0\n",
    "max_pos_y = 0\n",
    "team_name = []\n",
    "total = 40403\n",
    "with open('sberbank_data/final/academy2019_final_train.jsonlines') as fin:\n",
    "    for i, line in tqdm(enumerate(fin), total=total):\n",
    "        record = json.loads(line)\n",
    "        \n",
    "        # read one line\n",
    "        skilled.append(record['skilled'])\n",
    "        record_ids.append(record['id'])\n",
    "        team_name.append(record['player_team'])\n",
    "        \n",
    "        # обработка записи\n",
    "        pos_x = record['actions']['pos_x']\n",
    "        pos_y = record['actions']['pos_y']\n",
    "        \n",
    "        poisitions_x.append(pos_x)\n",
    "        poisitions_y.append(pos_y)\n",
    "        \n",
    "        max_pos_x = max(max(pos_x), max_pos_x)\n",
    "        max_pos_y = max(max(pos_y), max_pos_y)\n",
    "        \n",
    "        i += 1\n",
    "\n",
    "print(\"max_pos_x: %d  |  max_pos_y: %d\" % (max_pos_x, max_pos_y))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40403/40403 [00:15<00:00, 2556.86it/s]\n"
     ]
    }
   ],
   "source": [
    "### Обработаем позиции \n",
    "\n",
    "# тут будут лежать хитмапы\n",
    "heatmaps_pos = np.zeros((40403, 64, 64))\n",
    "\n",
    "for i, (x, y) in tqdm(enumerate(zip(poisitions_x, poisitions_y)), total=40403):\n",
    "    hist = np.zeros((64, 64)) \n",
    "    min_x, max_x = min(x), max(x)\n",
    "    min_y, max_y = min(y), max(y)\n",
    "    \n",
    "    \n",
    "    x = (np.array(x) - min_x) / (max_x - min_x)\n",
    "    y = (np.array(y) - min_y) / (max_y - min_y)\n",
    "    \n",
    "    hist[np.ceil(x * 64).astype(int) - 1, np.ceil(y * 64).astype(int) - 1] += 1\n",
    "    heatmaps_pos[i] = hist"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# готовим для нейронки на keras\n",
    "X_pos = np.expand_dims(heatmaps_pos, -1)\n",
    "\n",
    "# нормируем все значения на [0, 1]\n",
    "X_pos = scale_data(X_pos)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Rotate Dire map "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def rotate_pos(X_pos, team_names):\n",
    "    \"\"\"\n",
    "    Чтобы для всех игроков своя база была внизу, а \n",
    "    база врага наверху - перевернем карту для dire\n",
    "    \"\"\"\n",
    "    new_pos = copy.deepcopy(X_pos)\n",
    "    \n",
    "    for ind in range(len(X_pos)):\n",
    "        if team_name[ind] == 'dire':\n",
    "            new_pos[ind] = rotate(new_pos[ind], 180)\n",
    "    \n",
    "    return new_pos"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Before rotate "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'dire'"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "team_name[16]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fab662a3668>"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(X_pos[16]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## After rotate "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "new_position = rotate_pos(X_pos, team_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fab64306630>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(new_position[16]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Player cursor pos"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40403/40403 [01:01<00:00, 656.88it/s]\n"
     ]
    }
   ],
   "source": [
    "cursors_x = []\n",
    "cursors_y = []\n",
    "\n",
    "skilled = []\n",
    "record_ids = []\n",
    "\n",
    "total = 40403\n",
    "with open('sberbank_data/final/academy2019_final_train.jsonlines') as fin:\n",
    "    for i, line in tqdm(enumerate(fin), total=total):\n",
    "        record = json.loads(line)\n",
    "        \n",
    "        # read one line\n",
    "        skilled.append(record['skilled'])\n",
    "        record_ids.append(record['id'])\n",
    "        \n",
    "        \n",
    "        # обработка записи\n",
    "        cursors_x.append(record['actions']['cursor_y'])\n",
    "        cursors_y.append(record['actions']['cursor_x'])\n",
    "            \n",
    "        \n",
    "        i += 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "heatmaps = np.zeros((40403, 64, 64))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 40403/40403 [00:25<00:00, 1594.88it/s]\n"
     ]
    }
   ],
   "source": [
    "for i, (x, y) in tqdm(enumerate(zip(cursors_x, cursors_y)), total=40403):\n",
    "    hist = np.zeros((511, 511)) \n",
    "    hist[np.array(x) - 1, np.array(y) - 1] += 1\n",
    "    res = cv2.resize(hist, dsize=(64, 64),interpolation=cv2.INTER_AREA) \n",
    "    \n",
    "    heatmaps[i] = res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# готовим для нейронки на keras\n",
    "X_cur = np.expand_dims(heatmaps, -1)\n",
    "\n",
    "# нормируем все значения на [0, 1]\n",
    "X_cur = scale_data(X_cur)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fab624e6828>"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAD8CAYAAABXXhlaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4zLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvIxREBQAAIABJREFUeJztnXtwHNd15r8zgxcBEgABECAIigJpUiIpUyQtSqLE2JFkSSvLWiubRIkf5WJSTDFVcbJWko0tb6pScW1SZa2rbO1uOc4ytmOuS5Esx/ZK0aocy7QU25FNkRJFiU+RovgACRJ8AAT4wGNmzv4xg7n3NtGNxmCmZ6D+flUo3J6+fftgZg76nHvOPVdUFYSQeJEotwCEkOih4hMSQ6j4hMQQKj4hMYSKT0gMoeITEkOo+ITEkGkpvog8ICIHReSwiDxWLKEIIaVFCk3gEZEkgLcB3AegB8AOAJ9Q1X3FE48QUgqqpnHtbQAOq+oRABCRpwE8DMBX8WukVuvQMI1bBiO1Nfm2joyW7D6EVCrDuIxRHZHJ+k1H8bsAnLCOewDcHnRBHRpwu3x4GrcMpmphd76dOnK0ZPchpFLZrttC9ZuO4k/0X+Uav0FENgPYDAB1qJ/G7QghxWI6it8D4DrreCGAU95OqroFwBYAaJSWkq4I4lOekHBMZ1Z/B4BlIrJYRGoAfBzAc8URixBSSgp+4qtqSkT+GMC/AkgC+Jaq7i2aZISQkjEdUx+q+gKAF4okCyEkIqal+GTqVC3pzrc5J0HKBVN2CYkhVHxCYghN/YiheV9abFcKmBnvdzncPz7xCYkhVHxCYggVn5AYMuN8/KruRfl26uhx51zqnltMv5++FplMMxGvL2xj+5lB73e5CPLjZ4JP76Uc7zef+ITEECo+ITFkxpn6Wu0vcrHN+yCTstAQTKVk7unFwXw7s6jTOWe7THj5jXxzKqGyUpqsM9GcD0vq+MlI7sMnPiExhIpPSAyZcaa+pDOh+oU1SwudIQ40czvn59sZy6T2Xlc1v8M5lxm6ZNqXL/uOP/Tx9fn27BPDzrnq3oF8e7i71dzL4walz18wB3Yb/l+KqZjYWlMdum85qNgMv0w6ktvwiU9IDKHiExJDqPiExJAZ5+OH9cUO/74bolr8xZ58O71hlTlxon/aMumdq53ji511+Xbjvx12zlU1zsm3M21znXNXbzEhsEvzzUcz75k9Tr/aAeMHVvVfcc75+fVTCsUVIeQoqXC+aqLB7LMgdbXOubRn7mGcUvjnlRJmjQo+8QmJIVR8QmLIjDP1w7Lkv7vmsXSbLQBSCbMXyIX1851+LVbba/LZ5uDY/CYzxkp3o5DG42P5dvrced8xMnsOOOeS7SZjrv3nfWaMoSGnX80FK4TXc9odf9/bmIigv2W0y3U5Uj/fNeEYQSSbm5zjdKtxaVLd/ounMlcsV+WqG5osJXEw54PgE5+QGELFJySGUPEJiSEzwsd3VnrZq5c86Y2BIRnLT65JJs01P3f7pVIpc25hl3NO68w23L1/Yfz45v/jphFXDZlzIw/e6srxwg74UXfUzAcce2RBvt31kjuHkNhvZPb6/4OfMOm8jU/9ysgUEAJLHu9xztkbHMotN5n7Hu11+km9kSt1wh0DO94y90YAat1Nw4UAw4YiJ+sbZyZ94ovIt0SkT0T2WK+1iMiLInIo93tu0BiEkMoijKn/bQAPeF57DMA2VV0GYFvumBAyQ5jU1FfVn4lIt+flhwHclWtvBfAygM8XUa6iY5uAJx5qz7fbX290+qVrjRuQGnVN+Hc3GrO05XmTZda42zWBbeSXR53jK795e75dPZhyO//EhLq6Hnevs0nYIUGPqd/0vZ35tm2yn77PzWSc96rJmNNd7l6njrncb60SbPK8V5aLUKiJXeyMuVJnJIaWwy5mgsqrAVno5F6HqvYCQO53+yT9CSEVRMkn90RkM4DNAFCH+kl6E0KioFDFPyMinaraKyKdAPr8OqrqFgBbAKBRWtSvXxCF1GwLMj07tjfn2zWnBpx+owv8zy3baMaQajPDn1p7o3vvE2fz7bF7XZOv4TnL5LOiCwCQDGmKDt1sDKzEinnu+PvNve2/JTHmdHPM+6BS2zbnNrjuQpvVLoXpXAzTPOwYxXYDKs2091Koqf8cgI259kYAzxZHHEJIFIQJ5z0F4JcAbhSRHhHZBOBLAO4TkUMA7ssdE0JmCGFm9T/hc+rDRZaFEBIRFZO5VwwfK6i4pJ2Fl96+z9xrbNTpp9d/wPQ76R+mU+s6O5MOAPo/arLdmne7q/Nk0UJz7wJ9zjnbzZyHXr3qXthicqmS1t/ZftSdC/AEEkPR/J1fTnsMbzak314F3nNhx8xc8BRWsVb8JepMgZT0B9x5mdQru0Pd670Cc/UJiSFUfEJiSMWY+sUOoXjNRttsV495b5N86XXTL+BeQTvzNv7QFLJIj4z4XlcHlyCz1+nXe9r3nCwzC5oS58wddI6bQ5FsNFl4mbOuOzJ66w1GjgLr9vnhlaMYpM+eM+N73m+n0IeFlNO0T7hh3KpFxlWJakdiPvEJiSFUfEJiCBWfkBhSMT5+WMKG/YqRnqkb1jjH9r50CEjJ9PqZzr2t64Y9K7hsn7/QOY/kebNab3jdsgnvC8DxM8fucf/OWUeMzx8UsrPfH+e9gafQhzWfgHOefvZ7UGCaa9D7XQhTWVkXOiW4yxRWSZ085Z5MRP/85ROfkBhCxSckhswIU98xk+yQl/X6NdTWOIdht7+2Sb7m1r1PDZssMCe0NeIJD9r39pwbvtGscPOakMMBoT5HjiD3ZNCY+plqs35u6HfXO/3m/sKEjaqOuOE8OwMyyP2oqjNhP/VkzCXbzBjevQVsanaY9zRz2yr35KtW3b4CMzvDXjf6gKmNWL/HNcWD3J2wsmSs7cAqoS4gn/iExBAqPiExZEaY+hmfXVP9XgeAzHC47Zhk7U3OsV2gIn3Lcudc9Sljztrmmb34w3tv77m6/aaOn9eEtMtr20ylnLRN/V6TrXjm9653rxs2i4XmvOXWUak5b7LdMnNm+d4rZW3XFWS+BsloZw0mj51xztnFtoPeg+TSxfm2XPJk6oXclqvmR6bseSGLj7wEZl5WQMlvPvEJiSFUfEJiCBWfkBgyM3x8H389rB8PuD6XHa6q3f2u08/2K+Xf33DOjdxtinQk3z2Wb+v7lzr9kpdNJll6/yFXkH6TuRa4rVWHVVBzzhynX2JeK/ywxxj49B359nV/84rTzy5eMbTOLY5x8rdMZc4b/9vghGMDhWVRev9mZ18Azx4BfuNfM0aPmctIe74T9hxLkLxSa/ZJSM5rc86lek5iqlSCHx8En/iExBAqPiExRFQLKnVfEI3SordLeWp0OubhmAnYpE+7oaxkq6lZlzrthpcS1u6wdoEH20wEgNEPvt+MP8stulC/Lb/3KBLz3Q2I/MzDZHOTcyzWVlaZ5tnOuczu/ROOcU1hkhZzXeId15S9codxXU59yHiDix9za+4VQjHCXPbnAACJRssVqp/ldrbCeUEFTMJSCVl3QWzXbRjUCzJZPz7xCYkhVHxCYggVn5AYMiPCeYWszAqbQnpN4U2vj2hh+/X2ykBv6nDtOVPr/uqCBuecPUa6w+O7HzFtO5yHlJtEmjphrR47loYfTkHQfT3OOWfmYa4rR+0LJn112YHufPvSQ7c5/YbnmlG8NfeLUTDFj6C5kZEHb3XO2X+LTaG+eqX59IUSZgut60TkJRHZLyJ7ReSzuddbRORFETmU+z13srEIIZVBGFM/BeDPVXUFgPUAPiMiKwE8BmCbqi4DsC13TAiZAYTZO68XQG+uPSQi+wF0AXgYwF25blsBvAzg86UQMrQZFrJ+m1OUw1NfbaDLFNFoDjJRvXXTLBIpswIPnsDKsGUuS8YNpQ785zvz7dp+/zBr27+bTLXAlXvWe3Dkr+50zi3+mikyklnkbn9tk7HCnQ2X3ZVvVTdd53ud06/IW1B7x7DN+5p+/z0TbDkO/Ml859zSPw0nV6WH88Iypck9EekGsBbAdgAduX8K4/8c2v2vJIRUEqEVX0RmA/g+gEdVdXCy/tZ1m0Vkp4jsHENxq6ESQgojlOKLSDWySv+kqv4g9/IZEenMne8E0DfRtaq6RVXXqeq6atRO1IUQEjGTpuyKiCDrw19Q1Uet178M4LyqfklEHgPQoqqfCxqrGCm7jr/Y3uie6zUVctRTbFNSJuyVqTcrtjJ73IKaiQYTfkuvWeacq+q30nSH/X1J2+/zVvg5+Ecm3XT5o/ucc1hsfObElXArD89tcP3z2SeNXLaPL+ve7/RLHD6Rb6cHLjrn7HmP4VYzDTT7hLsld/JdkwIrDW4arZ6zQpw11eZeAYU3vRR7bqDY41UiYVN2w8TxNwD4NIC3RGR8nep/BfAlAM+IyCYAxwE8UqiwhJBoCTOr/wtcMzedpzwrbggh02JGZO7ZOEUuz7hZcSc3rc63W/e4E4l2IcvEOeMSJDzhGbtIh7fuvZ0jF7TKzBlvvmsC3/CHZhvu0bvcratsGW0TvvXH77hynDHTKW7JCH8TNnHCnYKxzfvkTTc65/qWm7mYBf9i6u9n5rorAfXS5Xzba+qnB838r/1eVTW6RUVsvNt16+WJt7ieCmGLebxXTX8/mKtPSAyh4hMSQ2aEqe9nriXaWpx+iTETofCa6WFrpVcVsFijqtPNAjv/B6bWXes3/ItXBMk423I59Io7m25juyaAu+WVbSqfv/99Tr/m71imvyey07HduAE6aqIEiUueSEPHvHwzcFsvSw7bTZmMKqu2oL1oyetWFGLCv9dM+/G/W3pqgjvm4BOfkBhCxSckhlDxCYkh79lim4HhNqvYpta7e9ulDx4ONabtI+qdq51+ydcP5ttBtf9Lsb+avT31wL0m87Bpv5udd+Hm5nzbXu3nvXdi9Yp826+QJxDsW4cdI2hMew+/qYwRJVJlpsySXW5GZerYCW/3ksBim4QQX6j4hMSQspn6U8mcSs4zYaP02bPT7ufUogso2BEW79/S8zFTj6++L+Oca/ynX4UaU25dlW/rjrcC72eTbjWZcedXzfbt1/rWJd/xC3m/R1YtcmUM+b6G/SzisMCmGNDUJ4T4QsUnJIZQ8QmJITMinJdsNAU3/FZ9AW7xB7uft68Omu2YvYUhLv/W7fl20y537zybwCKX1r0u3eSWIpx12qTfHtzshhKRMq7Zyi/7p7YG3Xv4P5pinrN3mT3xLnzQLYw5tMj8z7/+e27hUFvmWacuw4/EIauYx2C4amze4qY29urEa66bYX59uVb/0ccnhPhCxSckhlTM6rygcI2feT8V88k273VhhznhMfWbdhqz19mqCgAy/ttV+TF7r2uy995vMrqS9W5W3x+veSnffvHLpkhHWLcCAC51mm2tzq6+Pt/ufMUtTDLSZFZxeVf4NfzSFP4Qq3DGNZ+Lr1T+BIXswq6gvGbMEmRAFkLYfR0CEWOlV13vumepo8e9vQuGT3xCYggVn5AYUjGz+kEmvN857wxx7e53fe995XZTiOJyh/FwGo+5ZbJT9cZUntXrzmhfWGWiC3O/bQpsnH7U3Z5qwd+bunqJBW6RjsE1HfDDNtPn/5uJUIy2N0zUHcC1pvPFT63Pt1teN25MpsF/T4MLN7l18BIp851o+dHb+bY0ueXMbcK6I1Mxve1S55lVViGRX70Z6l5TuV+x3YVkq1skJu3ZUblUcFafEOILFZ+QGELFJySGRBrOk9oaVC3sBnCt3xToR2WsFW4J4wd7M71Gbu7OtweWuj5t+8/MKrPaF/yLbdRZvt7pe91iCm3feDXfvvKfTIZfYswjrlV8I+P5u2SVyYqrHnIDYnMvmeNUs7XNV7XrstWdturZWyvkAOD8zaZv05OH8m2vD3v+DjP30HjcDfXVHjP+qKaNTNrvFvNA21yEQS+acKzUup+LXbwi0eH+LU4RzT4zhjfsF3aPA+eargXuCyNmrsebzVm1sMscWNuBeWV0SBcS7AyPd35rXBeKVmxTROpE5FUR2S0ie0Xki7nXF4vIdhE5JCLfFZFwdySElJ0wpv4IgHtUdTWANQAeEJH1AB4H8FVVXQagH8Cm0olJCCkmYfbOUwDjVRuqcz8K4B4An8y9vhXAXwP4euBYI6MFhUaGlxgTsMrKXhq9zjU1qy4Zm7tjmxs+sbekarbq6nnDLoOrTbht/k/cWnToNplUtQPmXj0fcYttLLBMzzN3u+7CvO1m+64La135W18wtfrs0Fl6iZtZZzP0a0uc46Vf3G2us2oBHvugGxKsNnU4UHPJ/RrU2KZ5i785n55t6uAFhq8yVsj45hucfv03GLmanvQvUmJ/b05+3g2fdj3+iu91Ns5CLa/b0mm+Y0nxRMMs8z7s99e7A3Gx8duTQdV/F2ebUJN7IpLM7ZTbB+BFAO8AGFDV8fv1AOjyu54QUlmEUnxVTavqGgALAdwGYMVE3Sa6VkQ2i8hOEdk5hpGJuhBCImZK4TxVHQDwMoD1AJpFZNxGXAjglM81W1R1naquq4Z/9hghJDom9fFFZB6AMVUdEJFZAO5FdmLvJQC/DeBpABsBPDvpWAHhPEcob3gm5EonfW2fOeh2VzZdXmD8ttlWKMTeEhoAZl0w/nrPw27Ip+mICSRdWG7euhVPnHPvtdz4ixfWuP7/vKfMHEX/p1z/ueGM8dd77jZ+5VirG8Ba8rTZO67/hqRzbs7Pjd89NseMsehfXBnP32LmDep73NTkw39hts2edca8b3XnXKOu+Tsmbbn3j1y/u/3vjN+d7jfzGgPLlzv9Zp/090ntkJUduu3Y4VqOiXrzfiQ823CnTk9cTEXmNrn99r09Yb/3KmHi+J0AtopIElkL4RlVfV5E9gF4WkT+BsAuAN8soZyEkCISZlb/TQBrJ3j9CLL+PiFkhlExq/PCEnYV3zVY22b13WvcgPYX3eIGA3cszLfrzrsm9qUuk6M0d58JeR192F21lnqfydyrOTDLOVffa97vjp+64UKtNv+Hryw1bsC597vZYrap23O3mzdVO2BM8yudxs1Y+pe7nH7eLZ5s3t5sztVbpv7CZ91pnLEOy1z2RMAuvs+Y3/Z71b/Sf4VfrSeTcc6bpohJ5ozJvEytXeb0q9plMhS9W34nrLBoqve0773fK3B1HiHEFyo+ITFkxpn6NlMpulDIQg5vLbra06Zu34mPtuXbbW+5q3SuzDMmu3je37PW2orl/8s1PW1TX3uMG+BdvJJuMtluPfe7s9OLvmmy/zBq5Bq61029SIwZubwltG1z3N5J185+BICmd66YWzW5Lse51VZE4R8OGNk9BSmk2lyX9C7S6TmJibCvAQBZYaIhmTcPeLubfmtvyrd1117ffjMZmvqEEF+o+ITEECo+ITGkYurqF8JUfHq778Cn78i37ewzwM0WS4644aW+O4zPL9apup4hp1/tOfO29tzrhq+W/qnJaAuqI2/Lf+ojbgZh59PGj2045Y4/+OtL8+057xi5zq90M/zmvm1CfYl3epxzbf1WKNHy68c8u24nrpq/oH7fUedc9RLjT9t+vf3eA8DcZ0xh0kz/gHPOb14mc9rdqyDRZzIDEwGfe+K4ma8Q7/ZrF8wY16yss4q/JBrqnVOZIfeznynwiU9IDKHiExJDZnQ4rxgEuQTX7MZrFagIqpNuX5dpdE3DxKAJgdlhOQBIXvTfmdaPnq+4mYFt/9Pc78JKswBpYK27GKbqnAm3JYfd6M+SJ/bn20N3mcIZVVfdBUdnbjFjLH7KzepLtxi/QFLmut4PNjv9FnzHhNW8RT/sbL3zj9ycb8/7mZvxaLsjbT/x31shKHPPLiTi3T8gddTsCpxsdP2dUhfcmCoM5xFCfKHiExJDqPiExJAZHc4rBkEhwUK3WLavk1tXOecyZ01BiXSX69PKmIkRJq6YFX5eORJrVubbnX/rhul0hyla0gITmuz8sRsqk5RVL7/KHcMuUlFz0YTshlvdr0vTu8Z3t7chBwB7RFv+rsFuV94q/6/g6O2maEfb86ZQRubKVaffcIsJd+qYmz7trZHvhz1nI5fcuZbMndZn+Is3Qo1X6fCJT0gMoeITEkMqJpyXbLTqyA8OTtin3PgVASl0lWCq3Q0bJS6ZkJtt6uvlK06/tz9nVqPd+DU3RFWoe+KMYWUv1lwwZvVoqxs6rL5oCoKMNbq1C71138fxZu55MyedMeabPQ4u37Io3679fzt8r7lmjAK36J6pMJxHCPGFik9IDCmbqW+b9gAweJ8pFNHw/e2RyTQVCjEbC3YDgsa/zcwyJw8cc05JW4u39+TjBXDlN82uwDUD7rIiP3PeS9giKIW+N3Ez54OgqU8I8YWKT0gMoeITEkPKlrln+/RA5fr1YVBrO2oAqLpoZd3tPejt7j+OFbZLNJiVe5nLnlV7r76Vb7qlQgAUORRa/wP/z6XYfndijrv9VWKeKXxSigzLOBP6iZ/bKnuXiDyfO14sIttF5JCIfFdEaiYbgxBSGUzF1P8sgP3W8eMAvqqqywD0A9hUTMEIIaUjlKkvIgsBfBTA3wL4MxERAPcA+GSuy1YAfw3g62FvPOeQW8AgEWAa2plkdghpKqGyYuA3vryy2zm+xvwOiVj13NJnTF25oHBYMcKKYd/Hc3/oZt21/W//rLtCPgupdzMDacKXjrBP/CcAfA7A+HKsVgADqjoe2O0B0FVk2QghJWJSxReRhwD0qaqdrTFRgsCEmUAisllEdorIzjGMTNSFEBIxYUz9DQA+JiIPAqgD0IisBdAsIlW5p/5CAKcmulhVtwDYAmQz94oiNSFkWkwpZVdE7gLwX1T1IRH5HoDvq+rTIvL3AN5U1b8Lur6xcaGuu/UzAICfPPkt59x/WLBmiqKXhpmW/lkM/z8syY5259ieh4iSRL2ngGmjCQOmTp+JWpyKIoqU3c8jO9F3GFmf/5vTGIsQEiFTSuBR1ZcBvJxrHwFwW/FFIoSUmmhX51mm/tW2audc82vGRJsJJvZMww6JAuFX1pWaxGqTwSmn3fp4fiFNfj/84eo8QogvVHxCYkiki3RkNIXaY9kyxiqtzrlCzLdjz7ilq7s/aTKKNRW0F238qBTT3ktm9/7JO4HmfbHhE5+QGELFJySGUPEJiSHRFuIQgc7K1l8vhs85MuTWck8sNrXXJe1u6UwfsfKJerVlnOETn5AYQsUnJIZEauoPt1Xh4KbsDrFL/2z64634irtDa/rQkekPSsoGTfvo4BOfkBhCxSckhlDxCYkh0absKpAcnXThUGgu3jTXOZ69t2hDkwqGYb9rGX9PpCdclXs+8QmJIVR8QmJI2bbJJoQUHxbiIIT4QsUnJIZQ8QmJIVR8QmIIFZ+QGELFJySGUPEJiSGhUnZF5CiAIWS3fk+p6joRaQHwXQDdAI4C+B1V7S+NmISQYjKVJ/7dqrpGVdfljh8DsE1VlwHYljsmhMwApmPqPwxga669FcBvTF8cQkgUhFV8BfBjEXlNRDbnXutQ1V4AyP1u972aEFJRhF2Wu0FVT4lIO4AXReRA2Bvk/lFsBoA61E/SmxASBaGe+Kp6Kve7D8APkd0e+4yIdAJA7nefz7VbVHWdqq6rRu1EXQghETOp4otIg4jMGW8DuB/AHgDPAdiY67YRwLOlEpIQUlzCmPodAH4oIuP9/0lVfyQiOwA8IyKbABwH8EjpxCSEFJNJFV9VjwBYPcHr5wFwcT0hMxBm7hESQ6j4hMQQKj4hMYSKT0gMoeITEkOo+ITEECo+ITGEik9IDKHiExJDqPiExBAqPiExhIpPSAyh4hMSQ6j4hMQQKj4hMYSKT0gMoeITEkOo+ITEECo+ITGEik9IDKHiExJDqPiExBAqPiExhIpPSAyh4hMSQ0Ipvog0i8g/i8gBEdkvIneISIuIvCgih3K/55ZaWEJIcQj7xP8fAH6kqsuR3U5rP4DHAGxT1WUAtuWOCSEzgDC75TYC+BCAbwKAqo6q6gCAhwFszXXbCuA3SiUkIaS4hHniLwFwFsA/isguEflGbrvsDlXtBYDc7/YSykkIKSJhFL8KwAcAfF1V1wK4jCmY9SKyWUR2isjOMYwUKCYhpJiEUfweAD2quj13/M/I/iM4IyKdAJD73TfRxaq6RVXXqeq6atQWQ2ZCyDSZVPFV9TSAEyJyY+6lDwPYB+A5ABtzr20E8GxJJCSEFJ2qkP3+BMCTIlID4AiA30f2n8YzIrIJwHEAj5RGREJIsQml+Kr6BoB1E5z6cHHFIYREATP3CIkhVHxCYggVn5AYQsUnJIZQ8QmJIVR8QmIIFZ+QGCKqGt3NRM4COAagDcC5yG48MZUgA0A5vFAOl6nKcb2qzpusU6SKn7+pyE5VnSghKFYyUA7KUS45aOoTEkOo+ITEkHIp/pYy3demEmQAKIcXyuFSEjnK4uMTQsoLTX1CYkikii8iD4jIQRE5LCKRVeUVkW+JSJ+I7LFei7w8uIhcJyIv5UqU7xWRz5ZDFhGpE5FXRWR3To4v5l5fLCLbc3J8N1d/oeSISDJXz/H5cskhIkdF5C0ReUNEduZeK8d3JJJS9pEpvogkAXwNwEcArATwCRFZGdHtvw3gAc9r5SgPngLw56q6AsB6AJ/JvQdRyzIC4B5VXQ1gDYAHRGQ9gMcBfDUnRz+ATSWWY5zPIluyfZxyyXG3qq6xwmfl+I5EU8peVSP5AXAHgH+1jr8A4AsR3r8bwB7r+CCAzly7E8DBqGSxZHgWwH3llAVAPYDXAdyObKJI1USfVwnvvzD3Zb4HwPMApExyHAXQ5nkt0s8FQCOAd5GbeyulHFGa+l0ATljHPbnXykVZy4OLSDeAtQC2l0OWnHn9BrJFUl8E8A6AAVVN5bpE9fk8AeBzADK549YyyaEAfiwir4nI5txrUX8ukZWyj1LxZYLXYhlSEJHZAL4P4FFVHSyHDKqaVtU1yD5xbwOwYqJupZRBRB4C0Keqr9kvRy1Hjg2q+gFkXdHPiMiHIrinl2mVsp8KUSp+D4DrrOOFAE5FeH8vocqDFxsRqUZW6Z9U1R+UUxYA0OyuSC8jO+fQLCLjdRij+Hw2APiYiBwF8DRnM5Q3AAABFElEQVSy5v4TZZADqnoq97sPwA+R/WcY9ecyrVL2UyFKxd8BYFluxrYGwMeRLdFdLiIvDy4iguxWZPtV9SvlkkVE5olIc649C8C9yE4ivQTgt6OSQ1W/oKoLVbUb2e/DT1X1U1HLISINIjJnvA3gfgB7EPHnolGWsi/1pIlnkuJBAG8j60/+ZYT3fQpAL4AxZP+rbkLWl9wG4FDud0sEcvwasmbrmwDeyP08GLUsAG4GsCsnxx4Af5V7fQmAVwEcBvA9ALURfkZ3AXi+HHLk7rc797N3/LtZpu/IGgA7c5/N/wUwtxRyMHOPkBjCzD1CYggVn5AYQsUnJIZQ8QmJIVR8QmIIFZ+QGELFJySGUPEJiSH/H3RCkyqDoQBgAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(X_cur[30]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Encoder-decoder\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras.layers import Conv2D, Conv2DTranspose, Flatten, Input, Dense, Reshape\n",
    "from keras.layers.advanced_activations import LeakyReLU, PReLU\n",
    "from keras.datasets import mnist\n",
    "from keras import Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Позиция игрока на карте - вектор из 128 признаков "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "input_ = Input((64, 64, 1)) \n",
    "\n",
    "x = Conv2D(filters=4, kernel_size=2, strides=2, activation='relu',padding='same')(input_) \n",
    "x = Conv2D(filters=6, kernel_size=2, strides=2, activation='relu',padding='same')(x)\n",
    "x = Conv2D(filters=8,kernel_size=2, strides=2, activation='relu',padding='same')(x) \n",
    "flat = Flatten()(x)\n",
    "latent = Dense(128,activation='relu')(flat)\n",
    "\n",
    "\n",
    "reshape = Reshape((-1,8,8)) \n",
    "dense = Dense(512,activation='relu')\n",
    "conv_2t_1 = Conv2DTranspose(filters=8, kernel_size=2, strides=2, activation='relu',padding='same') \n",
    "conv_2t_2 = Conv2DTranspose(filters=6, kernel_size=2, strides=2, activation='relu',padding='same') \n",
    "conv_2t_3 = Conv2DTranspose(filters=1, kernel_size=2, strides=2, activation='relu',padding='same') \n",
    " \n",
    "\n",
    "x = dense(latent)\n",
    "x = reshape(x)\n",
    "x = conv_2t_1(x)\n",
    "x = conv_2t_2(x)\n",
    "decoded = conv_2t_3(x) \n",
    "\n",
    "autoencoder = Model(input_, decoded)\n",
    "encoder = Model(input_, latent)\n",
    "\n",
    "decoder_input = Input((128,))\n",
    "x_ = dense(decoder_input)\n",
    "x_ = reshape(x_)\n",
    "x_ = conv_2t_1(x_)\n",
    "x_ = conv_2t_2(x_)\n",
    "decoded_ = conv_2t_3(x_) \n",
    "decoder = Model(decoder_input, decoded_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "autoencoder.compile(optimizer='adam', loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 317,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# autoencoder.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/27\n",
      "40403/40403 [==============================] - 3s 72us/step - loss: 0.4123\n",
      "Epoch 2/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.2858\n",
      "Epoch 3/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1859\n",
      "Epoch 4/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1786\n",
      "Epoch 5/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1700\n",
      "Epoch 6/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1663\n",
      "Epoch 7/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1621\n",
      "Epoch 8/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1614\n",
      "Epoch 9/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1617\n",
      "Epoch 10/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1575\n",
      "Epoch 11/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1666\n",
      "Epoch 12/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1570\n",
      "Epoch 13/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1673\n",
      "Epoch 14/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1598\n",
      "Epoch 15/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1572\n",
      "Epoch 16/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1581\n",
      "Epoch 17/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1558\n",
      "Epoch 18/27\n",
      "40403/40403 [==============================] - 2s 61us/step - loss: 0.1551\n",
      "Epoch 19/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1515\n",
      "Epoch 20/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1549\n",
      "Epoch 21/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1512\n",
      "Epoch 22/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1528\n",
      "Epoch 23/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1545\n",
      "Epoch 24/27\n",
      "40403/40403 [==============================] - 3s 62us/step - loss: 0.1593\n",
      "Epoch 25/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1514\n",
      "Epoch 26/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1517\n",
      "Epoch 27/27\n",
      "40403/40403 [==============================] - 2s 62us/step - loss: 0.1470\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fa41b6ccc50>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "autoencoder.fit(x = new_position, y =new_position, batch_size=128, epochs=27)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred = encoder.predict(np.expand_dims(new_position[202],axis=0))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 128)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fa41aeb8b00>"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(new_position[202])) # Реальное"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fa41aef74e0>"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(decoder.predict(pred)[0])) # Восстановленное из вектора 128 признаков"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Курсор мыши "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "input_ = Input((64, 64, 1)) \n",
    "\n",
    "x = Conv2D(filters=4, kernel_size=2, strides=2, activation='relu',padding='same')(input_) \n",
    "x = Conv2D(filters=6, kernel_size=2, strides=2, activation='relu',padding='same')(x)\n",
    "x = Conv2D(filters=8,kernel_size=2, strides=2, activation='relu',padding='same')(x) \n",
    "flat = Flatten()(x)\n",
    "latent = Dense(128,activation='relu')(flat)\n",
    "\n",
    "\n",
    "reshape = Reshape((-1,8,8)) \n",
    "dense = Dense(512,activation='relu')\n",
    "conv_2t_1 = Conv2DTranspose(filters=8, kernel_size=2, strides=2, activation='relu',padding='same') \n",
    "conv_2t_2 = Conv2DTranspose(filters=6, kernel_size=2, strides=2, activation='relu',padding='same') \n",
    "conv_2t_3 = Conv2DTranspose(filters=1, kernel_size=2, strides=2, activation='relu',padding='same') \n",
    " \n",
    "\n",
    "x = dense(latent)\n",
    "x = reshape(x)\n",
    "x = conv_2t_1(x)\n",
    "x = conv_2t_2(x)\n",
    "decoded = conv_2t_3(x) \n",
    "\n",
    "autoencoder_cur = Model(input_, decoded)\n",
    "encoder_cur = Model(input_, latent)\n",
    "\n",
    "decoder_input = Input((128,))\n",
    "x_ = dense(decoder_input)\n",
    "x_ = reshape(x_)\n",
    "x_ = conv_2t_1(x_)\n",
    "x_ = conv_2t_2(x_)\n",
    "decoded_ = conv_2t_3(x_) \n",
    "decoder_cur = Model(decoder_input, decoded_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_5 (InputLayer)         (None, 64, 64, 1)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_7 (Conv2D)            (None, 32, 32, 4)         20        \n",
      "_________________________________________________________________\n",
      "conv2d_8 (Conv2D)            (None, 16, 16, 6)         102       \n",
      "_________________________________________________________________\n",
      "conv2d_9 (Conv2D)            (None, 8, 8, 8)           200       \n",
      "_________________________________________________________________\n",
      "flatten_3 (Flatten)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 128)               65664     \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 512)               66048     \n",
      "_________________________________________________________________\n",
      "reshape_3 (Reshape)          (None, 8, 8, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_transpose_7 (Conv2DTr (None, 16, 16, 8)         264       \n",
      "_________________________________________________________________\n",
      "conv2d_transpose_8 (Conv2DTr (None, 32, 32, 6)         198       \n",
      "_________________________________________________________________\n",
      "conv2d_transpose_9 (Conv2DTr (None, 64, 64, 1)         25        \n",
      "=================================================================\n",
      "Total params: 132,521\n",
      "Trainable params: 132,521\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "autoencoder_cur.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "autoencoder_cur.compile(optimizer='adam', loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/27\n",
      "40403/40403 [==============================] - 3s 70us/step - loss: 0.0602\n",
      "Epoch 2/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0526\n",
      "Epoch 3/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0510\n",
      "Epoch 4/27\n",
      "40403/40403 [==============================] - 2s 59us/step - loss: 0.0498\n",
      "Epoch 5/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0495\n",
      "Epoch 6/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0493\n",
      "Epoch 7/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0492\n",
      "Epoch 8/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0491\n",
      "Epoch 9/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0491\n",
      "Epoch 10/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0490\n",
      "Epoch 11/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0490\n",
      "Epoch 12/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0490\n",
      "Epoch 13/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0490\n",
      "Epoch 14/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0489\n",
      "Epoch 15/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0489\n",
      "Epoch 16/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0489\n",
      "Epoch 17/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0488\n",
      "Epoch 18/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0488\n",
      "Epoch 19/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0488\n",
      "Epoch 20/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0488\n",
      "Epoch 21/27\n",
      "40403/40403 [==============================] - 2s 59us/step - loss: 0.0487\n",
      "Epoch 22/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0488\n",
      "Epoch 23/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0487\n",
      "Epoch 24/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0487\n",
      "Epoch 25/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0487\n",
      "Epoch 26/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0487\n",
      "Epoch 27/27\n",
      "40403/40403 [==============================] - 2s 58us/step - loss: 0.0487\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fa41ae8ccf8>"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "autoencoder_cur.fit(x = X_cur, y = X_cur, batch_size=256, epochs=27)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fab6a33b940>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(X_cur[10]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "test_dims = np.expand_dims(X_cur[10],axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 64, 64, 1)"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_dims.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pred_cur = encoder_cur.predict(test_dims)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "decoded_cur = decoder_cur.predict(pred_cur)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7fab5c101438>"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.imshow(np.squeeze(decoded_cur))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Add new features to DF "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df_train = pd.read_csv('sberbank_data/data_2019/train_extracted_final_v2.csv').drop('id',axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((40403, 10), (40403, 808))"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "reasons_df.shape, df_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "df_train = pd.concat([df_train,reasons_df],axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "cursor_data = encoder_cur.predict(X_cur)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pos_data = encoder.predict(new_position)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 84,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pos_columns = ['pos_'+str(i) for i in range(1,129)]\n",
    "cur_columns = ['cur'+str(i) for i in range(1,129)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "extra_data = np.concatenate([pos_data, cursor_data],axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "extra_data_df = pd.DataFrame(data=extra_data, columns = pos_columns+cur_columns)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "full_train = pd.concat([df_train, extra_data_df], axis=1)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}