analysis.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "        map  patch_version    map_crc current_score round_status  \\\n",
      "0  de_dust2          13737  702125823        [0, 0]   FreezeTime   \n",
      "1  de_dust2          13737  702125823        [0, 0]   FreezeTime   \n",
      "2  de_dust2          13737  702125823        [0, 0]       Normal   \n",
      "3  de_dust2          13737  702125823        [0, 0]       Normal   \n",
      "4  de_dust2          13737  702125823        [1, 0]   FreezeTime   \n",
      "\n",
      "   round_status_time_left                                      alive_players  \\\n",
      "0                20.00000  [{'health': 100, 'armor': 0, 'has_helmet': Fal...   \n",
      "1                 1.03125  [{'health': 100, 'armor': 100, 'has_helmet': F...   \n",
      "2                96.03125  [{'health': 100, 'armor': 100, 'has_helmet': F...   \n",
      "3                76.03125  [{'health': 100, 'armor': 100, 'has_helmet': F...   \n",
      "4                19.96875  [{'health': 100, 'armor': 100, 'has_helmet': F...   \n",
      "\n",
      "                                       active_smokes active_molotovs  \\\n",
      "0                                                 []              []   \n",
      "1                                                 []              []   \n",
      "2                                                 []              []   \n",
      "3  [{'position': {'x': 1184.4062, 'y': 2223.8438,...              []   \n",
      "4                                                 []              []   \n",
      "\n",
      "                                      previous_kills planted_bomb  \n",
      "0                                                 []         None  \n",
      "1                                                 []         None  \n",
      "2  [{'weapon': 'Glock', 'attacker_position': {'x'...         None  \n",
      "3  [{'weapon': 'Glock', 'attacker_position': {'x'...         None  \n",
      "4                                                 []         None  \n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import matplotlib.pyplot as plt\n",
    "from IPython.display import clear_output\n",
    "\n",
    "def read_dataset(template, start_idx, end_idx):\n",
    "    frames = [ pd.read_json(f) for f in [template.format(i) for i in range(start_idx, end_idx)] ]\n",
    "    return pd.concat(frames, ignore_index = True)\n",
    "\n",
    "dftrain = read_dataset(\"datasets/dataset_initial/dataset_{:02}.json\", 0, 13)\n",
    "y_train = dftrain.pop('round_winner').map(lambda s: 0 if s == \"CT\" else 1)\n",
    "\n",
    "dfeval = read_dataset(\"datasets/dataset_initial/dataset_{:02}.json\", 13, 18)\n",
    "y_eval = dfeval.pop('round_winner').map(lambda s: 0 if s == \"CT\" else 1)\n",
    "print(dftrain.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "RangeIndex: 71253 entries, 0 to 71252\n",
      "Data columns (total 17 columns):\n",
      " #   Column                  Non-Null Count  Dtype  \n",
      "---  ------                  --------------  -----  \n",
      " 0   map                     71253 non-null  object \n",
      " 1   patch_version           71253 non-null  int64  \n",
      " 2   map_crc                 71253 non-null  int64  \n",
      " 3   current_score           71253 non-null  object \n",
      " 4   round_status            71253 non-null  object \n",
      " 5   round_status_time_left  71253 non-null  float64\n",
      " 6   alive_players           71253 non-null  object \n",
      " 7   active_smokes           71253 non-null  object \n",
      " 8   active_molotovs         71253 non-null  object \n",
      " 9   previous_kills          71253 non-null  object \n",
      " 10  planted_bomb            8034 non-null   object \n",
      " 11  alive_players_t         71253 non-null  int64  \n",
      " 12  alive_players_ct        71253 non-null  int64  \n",
      " 13  health_ct               71253 non-null  int64  \n",
      " 14  health_t                71253 non-null  int64  \n",
      " 15  money_ct                71253 non-null  int64  \n",
      " 16  money_t                 71253 non-null  int64  \n",
      "dtypes: float64(1), int64(8), object(8)\n",
      "memory usage: 9.2+ MB\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "CATEGORICAL_COLUMNS = ['round_status', 'map']\n",
    "NUMERIC_COLUMNS = ['round_status_time_left']\n",
    "INTEGER_COLUMNS = ['alive_players_t', 'alive_players_ct', \"health_t\", \"health_ct\", \"money_ct\", \"money_t\"]\n",
    "\n",
    "def get_attr(ds, team, attr=None):\n",
    "    team_players = map(lambda players: filter(lambda p: p[\"team\"] == team, players), ds['alive_players'])\n",
    "    if attr:\n",
    "        team_players = map(lambda players: map(lambda p: p[attr], players), team_players)\n",
    "    \n",
    "    return list(map(lambda p: list(p), team_players))\n",
    "\n",
    "for ds in [dftrain, dfeval]:\n",
    "    ds['alive_players_t']  = list(map(len ,get_attr(ds, \"Terrorist\")))\n",
    "    ds['alive_players_ct'] = list(map(len, get_attr(ds, \"CT\")))\n",
    "    ds['health_ct']        = list(map(sum, get_attr(ds, \"CT\", \"health\")))\n",
    "    ds['health_t']         = list(map(sum, get_attr(ds, \"Terrorist\", \"health\")))\n",
    "    ds['money_ct']         = list(map(sum, get_attr(ds, \"CT\", \"money\")))\n",
    "    ds['money_t']          = list(map(sum, get_attr(ds, \"Terrorist\", \"money\")))\n",
    "\n",
    "print(dftrain.info())\n",
    "\n",
    "feature_columns = []\n",
    "feature_names = CATEGORICAL_COLUMNS + NUMERIC_COLUMNS + INTEGER_COLUMNS\n",
    "for feature_name in CATEGORICAL_COLUMNS:\n",
    "    vocabulary = dftrain[feature_name].unique()\n",
    "    feature_columns.append(tf.feature_column.categorical_column_with_vocabulary_list(feature_name, vocabulary))\n",
    "\n",
    "for feature_name in NUMERIC_COLUMNS:\n",
    "    feature_columns.append(tf.feature_column.numeric_column(feature_name, dtype=tf.float64))\n",
    "\n",
    "for feature_name in INTEGER_COLUMNS:\n",
    "    feature_columns.append(tf.feature_column.numeric_column(feature_name, dtype=tf.int32))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def make_input_fn(data_df, label_df, num_epochs=10, shuffle=True, batch_size=32):\n",
    "    def input_function():\n",
    "        ds = tf.data.Dataset.from_tensor_slices((dict(data_df.filter(items=feature_names)), label_df))\n",
    "        if shuffle:\n",
    "            ds = ds.shuffle(1000)\n",
    "        ds = ds.batch(batch_size).repeat(num_epochs)\n",
    "        return ds\n",
    "    return input_function\n",
    "        \n",
    "train_input_fn = make_input_fn(dftrain, y_train)\n",
    "eval_input_fn = make_input_fn(dfeval, y_eval, num_epochs=1, shuffle=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "linear_est = tf.estimator.LinearClassifier(feature_columns=feature_columns)\n",
    "linear_est.train(train_input_fn)\n",
    "result = linear_est.evaluate(eval_input_fn)\n",
    "clear_output()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "{'accuracy': 0.5687973, 'accuracy_baseline': 0.5128742, 'auc': 0.58633125, 'auc_precision_recall': 0.5613092, 'average_loss': 12.670018, 'label/mean': 0.5128742, 'loss': 12.679905, 'precision': 0.54423594, 'prediction/mean': 0.9145763, 'recall': 0.9795843, 'global_step': 22270}\n"
     ]
    }
   ],
   "source": [
    "print(result)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Calling model_fn.\n",
      "INFO:tensorflow:Done calling model_fn.\n",
      "INFO:tensorflow:Graph was finalized.\n",
      "INFO:tensorflow:Restoring parameters from /run/user/1000/tmpau722bis/model.ckpt-22270\n",
      "INFO:tensorflow:Running local_init_op.\n",
      "INFO:tensorflow:Done running local_init_op.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f6784253bd0>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAEICAYAAAB1f3LfAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4yLjEsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+j8jraAAAc2klEQVR4nO3de7RcZZ3m8e8jEQQlgOSAmBACGFFgaSQBWaPYKNoE1AZcoKEZQUUDiKJLV4+AF2hn4kh3K4q20Cg0F5WLIBJHUEGmoR25GDRypw0QyCFpCJeGiAgmPPPHfguLkzonFXZdUpzns1at7Prt2++tc1K/877vrl2yTURExPP1on4nEBERgy2FJCIiakkhiYiIWlJIIiKilhSSiIioJYUkIiJqSSGJFzxJiyW9vSwfL+k7PTjnnpKGu32ecq4PSPrl89x3zDwlnSbp8622lXSrpD3H2PdySYc9n7xisEzodwIRvWT7S+1sJ+ksYNj257qb0brN9pFjrNupsSzpROBVtv970/p9uptdrCvSI4mBImnc/fEzHtscgyWFJPquDD0dJ+k2SY9K+ldJLynr9pQ0LOkzkv4T+FdJL5J0rKS7JD0s6UJJL2863vsl3VvWfXbEuU6U9N2m52+W9CtJ/yVpSRkmmgscAvwPSX+Q9OOy7SslXSxpuaR7JB3TdJwNJZ1V8r8N2HUNbbakYyTdLekhSf8o6UVl3Qck/T9JJ0t6BDhR0iaSzinnvlfS5xrb/+WQ+oakxyTdIWmvphUflHS7pBXlfEe0yOf4ksdiSYc0xc+S9L/G+Lm9XdJs4HjgfeX1+l1Z/2+SPty0/YdKHo9K+pmkbRqJl7Y+WPK/SdLOY71+sW5JIYl1xSHA3sD2wKuB5iGlVwAvB7YB5gLHAPsDfwW8EngU+GcASTsCpwLvL+s2B6a0OqGkqcDlwDeAIWAGsND26cD3gH+w/TLb7y5v2j8GfgdMBvYCPilp73K4E0ru25d2tDM3cAAwC9gF2A/4UNO6NwJ3A1sA80qOmwDblXYfCnywxfaTSi4/bCquDwLvAiaWfU6WtEvTvq8o+00ueZ8uaYc28gfA9k+BLwEXlNfr9SO3kbQ/VbF5D9Vr/e/AeWX1XwNvofq5bwq8D3i43fNH/6WQxLrim7aX2H6E6o3z4KZ1zwAn2H7K9pPAEcBnbQ/bfgo4ETiwDAEdCPwf29eUdZ8v+7dyCHCl7fNs/9n2w7YXjrLtrsCQ7S/aftr23cC3gTll/XuBebYfsb0EOKWNNp9Utr8P+NqINi+1/Q3bK4Gnqd5cj7O9wvZi4CtUxbLhQeBrpR0XAHcC7wSw/RPbd7lyNfBzYI8RuXy+vL5XAz8p7emkI4D/bfv20qYvATNKr+TPwMbAawCVbZZ1+PzRRSkksa5Y0rR8L1VvomG57T81Pd8GuKQMR/0XcDuwCtiy7PfssWw/weh/3W4N3NVmftsAr2ycs5z3+HJORp63tGFNxmpz87pJwPojjnkvVQ+i4X4/9w6szx5P0j6SrpP0SMl733LMhkfL6zRaLp2wDfD1ptfuEUDAZNtXAd+k6lU+IOl0SRM7fP7oohSSWFds3bQ8FVja9HzkLaqXAPvY3rTp8RLb9wPLmo8laSOq4a1WllANRbXS6pz3jDjnxrb3Leufc97ShjVpt80PUf3Vvs2I7e9vej5ZkkYeT9IGwMXAPwFb2t4UuIzqTbxhM0kvHSOXdqzpNuJLgCNGvH4b2v4VgO1TbM8EdqIa4vq7tTx/9FEKSawrjpY0pYzrHw9cMMa2pwHzmiZrhyTtV9ZdBLyrTKKvD3yR0X/Pvwe8XdJ7JU2QtLmkGWXdA1TzEQ03AI+XSf8NJa0naWdJjUn1C4HjJG0maQrw8Tba/Hdl+62BT4zWZturyvHnSdq4tPtTwHebNtsCOEbSiyUdBLyWqmCsD2wALAdWStqHak5ipL+XtL6kPajmU37QRv7NHgCmjbgAoNlpVK/PTgDl4oGDyvKukt4o6cXAE8CfqHqYMSBSSGJd8X2qsfu7y6PllULF14H5wM8lrQCuo5psxvatwNHleMuoJuJbfuCuzE3sC3yaaqhlIdCYKD4D2LEMxfyovJm/m2pC/h6qXsJ3qCbAAf6eakjontKOc9to86XAjeW8PynnHM3Hqd5k7wZ+Wdp3ZtP664HpJa95wIFlzmcF1cUJF5bX4m+pXrtm/1nWLaUqrkfavqON/Js1Cs/Dkn4zcqXtS4CTgPMlPQ7cAjQ+ZzKRar7pUarX8GGqHlQMCOWLraLfJC0GPmz7yn7n0iuSDEy3vajfuUTUlR5JRETUkkISERG1ZGgrIiJqSY8kIiJqGXc3g5s0aZKnTZvW7zQiIgbKjTfe+JDtoVbrxl0hmTZtGgsWLOh3GhERA0XSqHdryNBWRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbWkkERERC0pJBERUUsKSURE1DLuPtkeETGoph37k1r7L/7yOzuUyXOlRxIREbWkkERERC0pJBERUUvXComkMyU9KOmWptgFkhaWx2JJC0t8mqQnm9ad1rTPTEk3S1ok6RRJKvENyvEWSbpe0rRutSUiIkbXzR7JWcDs5oDt99meYXsGcDHww6bVdzXW2T6yKX4qMBeYXh6NYx4OPGr7VcDJwEndaUZERIyla4XE9jXAI63WlV7Fe4HzxjqGpK2AibavdfWdwOcA+5fV+wFnl+WLgL0avZWIiOidfs2R7AE8YPv3TbFtJf1W0tWS9iixycBw0zbDJdZYtwTA9krgMWDzVieTNFfSAkkLli9f3sl2RESMe/0qJAfz3N7IMmCq7TcAnwK+L2ki0KqH4fLvWOueG7RPtz3L9qyhoZbfFBkREc9Tzz+QKGkC8B5gZiNm+yngqbJ8o6S7gFdT9UCmNO0+BVhaloeBrYHhcsxNGGUoLSIiuqcfPZK3A3fYfnbIStKQpPXK8nZUk+p3214GrJC0e5n/OBS4tOw2HzisLB8IXFXmUSIiooe6efnvecC1wA6ShiUdXlbNYfVJ9rcAN0n6HdXE+ZG2G72Lo4DvAIuAu4DLS/wMYHNJi6iGw47tVlsiImJ0XRvasn3wKPEPtIhdTHU5cKvtFwA7t4j/CTioXpYREVFXPtkeERG1pJBEREQtKSQREVFLCklERNSSQhIREbWkkERERC0pJBERUUsKSURE1JJCEhERtaSQRERELSkkERFRSwpJRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbWkkERERC0pJBERUUvXComkMyU9KOmWptiJku6XtLA89m1ad5ykRZLulLR3U3ympJvLulMkqcQ3kHRBiV8vaVq32hIREaPrZo/kLGB2i/jJtmeUx2UAknYE5gA7lX2+JWm9sv2pwFxgenk0jnk48KjtVwEnAyd1qyERETG6rhUS29cAj7S5+X7A+bafsn0PsAjYTdJWwETb19o2cA6wf9M+Z5fli4C9Gr2ViIjonX7MkXxM0k1l6GuzEpsMLGnaZrjEJpflkfHn7GN7JfAYsHmrE0qaK2mBpAXLly/vXEsiIqLnheRUYHtgBrAM+EqJt+pJeIz4WPusHrRPtz3L9qyhoaG1yzgiIsbU00Ji+wHbq2w/A3wb2K2sGga2btp0CrC0xKe0iD9nH0kTgE1ofygtIiI6pKeFpMx5NBwANK7omg/MKVdibUs1qX6D7WXACkm7l/mPQ4FLm/Y5rCwfCFxV5lEiIqKHJnTrwJLOA/YEJkkaBk4A9pQ0g2oIajFwBIDtWyVdCNwGrASOtr2qHOooqivANgQuLw+AM4BzJS2i6onM6VZbIiJidF0rJLYPbhE+Y4zt5wHzWsQXADu3iP8JOKhOjhERUV8+2R4REbWkkERERC0pJBERUUsKSURE1JJCEhERtaSQRERELSkkERFRSwpJRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbWkkERERC0pJBERUUsKSURE1JJCEhERtaSQRERELSkkERFRSwpJRETU0rVCIulMSQ9KuqUp9o+S7pB0k6RLJG1a4tMkPSlpYXmc1rTPTEk3S1ok6RRJKvENJF1Q4tdLmtattkRExOi62SM5C5g9InYFsLPt1wH/ARzXtO4u2zPK48im+KnAXGB6eTSOeTjwqO1XAScDJ3W+CRERsSZdKyS2rwEeGRH7ue2V5el1wJSxjiFpK2Ci7WttGzgH2L+s3g84uyxfBOzV6K1ERETv9HOO5EPA5U3Pt5X0W0lXS9qjxCYDw03bDJdYY90SgFKcHgM2b3UiSXMlLZC0YPny5Z1sQ0TEuNeXQiLps8BK4HsltAyYavsNwKeA70uaCLTqYbhxmDHWPTdon257lu1ZQ0ND9ZKPiIjnmNDrE0o6DHgXsFcZrsL2U8BTZflGSXcBr6bqgTQPf00BlpblYWBrYFjSBGATRgylRURE9/W0RyJpNvAZ4G9s/7EpPiRpvbK8HdWk+t22lwErJO1e5j8OBS4tu80HDivLBwJXNQpTRET0Ttd6JJLOA/YEJkkaBk6gukprA+CKMi9+XblC6y3AFyWtBFYBR9pu9C6OoroCbEOqOZXGvMoZwLmSFlH1ROZ0qy0RETG6rhUS2we3CJ8xyrYXAxePsm4BsHOL+J+Ag+rkGBER9eWT7RERUUsKSURE1JJCEhERtaSQRERELSkkERFRS1uFRNJqV01FRERA+z2S0yTdIOmjjVu/R0REQJuFxPabgUOobkmyQNL3Jb2jq5lFRMRAaHuOxPbvgc9R3eLkr4BTypdUvadbyUVExLqv3TmS10k6GbgdeBvwbtuvLcsndzG/iIhYx7V7i5RvAt8Gjrf9ZCNoe6mkz3Uls4iIGAjtFpJ9gSdtrwKQ9CLgJbb/aPvcrmUXERHrvHbnSK6kuvtuw0YlFhER41y7heQltv/QeFKWN+pOShERMUjaLSRPSNql8UTSTODJMbaPiIhxot05kk8CP5DU+JrbrYD3dSeliIgYJG0VEtu/lvQaYAdAwB22/9zVzCIiYiCszTck7gpMK/u8QRK2z+lKVhERMTDaKiSSzgW2BxZSfac6gIEUkoiIca7dHsksYEfbbvfAks4E3gU8aHvnEns5cAFVz2Yx8F7bj5Z1xwGHUxWqY2z/rMRnAmdRXX58GfAJ25a0AVUhmwk8DLzP9uJ284uIiM5o96qtW4BXrOWxzwJmj4gdC/zC9nTgF+U5knYE5gA7lX2+JWm9ss+pwFxgenk0jnk48KjtV1HdpuWktcwvIiI6oN1CMgm4TdLPJM1vPMbawfY1wCMjwvsBZ5fls4H9m+Ln237K9j3AImA3SVsBE21fW3pD54zYp3Gsi4C9JKnN9kRERIe0O7R1YofOt6XtZQC2l0naosQnA9c1bTdcYn8uyyPjjX2WlGOtlPQYsDnwUIdyjYiINrR7+e/VkrYBptu+UtJGwHpr2m8ttOpJeIz4WPusfnBpLtXwGFOnTn0++UVExCjavY38R6iGj/6lhCYDP3oe53ugDFdR/n2wxIepvjSrYQqwtMSntIg/Zx9JE4BNWH0oDQDbp9ueZXvW0NDQ80g7IiJG0+4cydHAm4DH4dkvudpizD1amw8cVpYPAy5tis+RtIGkbakm1W8ow2ArJO1e5j8OHbFP41gHAletzVVlERHRGe3OkTxl++nGXHbpAYz5pi3pPGBPYJKkYeAE4MvAhZIOB+4DDgKwfaukC4HbgJXA0Y1b1gNH8ZfLfy8vD4AzgHMlLaLqicxpsy0REdFB7RaSqyUdD2xYvqv9o8CPx9rB9sGjrNprlO3nAfNaxBcAO7eI/4lSiCIion/aHdo6FlgO3AwcQfXBwHwzYkREtH3V1jNUX7X77e6mExERg6bde23dQ4s5EdvbdTyjiIgYKGtzr62Gl1DNTby88+lERMSgaWuOxPbDTY/7bX8NeFuXc4uIiAHQ7tDWLk1PX0TVQ9m4KxlFRMRAaXdo6ytNyyspt4DveDYRETFw2r1q663dTiQiIgZTu0Nbnxprve2vdiadiIgYNGtz1dauVPe3Ang3cA3lNu4RETF+tVtIJgG72F4BIOlE4Ae2P9ytxCIiYjC0e4uUqcDTTc+fpvre9YiIGOfa7ZGcC9wg6RKqT7gfQPW1txERMc61e9XWPEmXA3uU0Adt/7Z7aUVExKBod2gLYCPgcdtfB4bLF1BFRMQ41+5X7Z4AfAY4roReDHy3W0lFRMTgaLdHcgDwN8ATALaXklukREQE7ReSp8v3oRtA0ku7l1JERAySdgvJhZL+BdhU0keAK8mXXEVEBG1ctSVJwAXAa4DHgR2AL9i+osu5RUTEAFhjIbFtST+yPROoXTwk7UBVmBq2A74AbAp8hOq74QGOt31Z2ec44HBgFXCM7Z+V+EzgLGBDqu+R/0QZgouIiB5pd2jrOkm7duKEtu+0PcP2DGAm8EfgkrL65Ma6piKyIzAH2AmYDXxL0npl+1OBucD08pjdiRwjIqJ97RaSt1IVk7sk3STpZkk3deD8ewF32b53jG32A863/ZTte4BFwG6StgIm2r629ELOAfbvQE4REbEWxhzakjTV9n3APl06/xzgvKbnH5N0KLAA+LTtR4HJwHVN2wyX2J/L8sj4aiTNpeq5MHXq1I4lHxERa+6R/Aig9Bi+avve5kedE0tan+qzKT8ooVOB7YEZwDL+8q2MarG7x4ivHrRPtz3L9qyhoaE6aUdExAhrKiTNb9bbdfjc+wC/sf0AgO0HbK+y/QzVpcW7le2Gga2b9psCLC3xKS3iERHRQ2sqJB5luRMOpmlYq8x5NBwA3FKW5wNzJG1Q7u81HbjB9jJghaTdyyXKhwKXdjjHiIhYgzVd/vt6SY9T9Uw2LMuU57Y98fmcVNJGwDuAI5rC/yBpBlXBWtxYZ/tWSRcCtwErgaNtryr7HMVfLv+9vDwiIqKHxiwkttcba/3zZfuPwOYjYu8fY/t5wLwW8QXAzh1PMCIi2rY2t5GPiIhYTQpJRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbWkkERERC0pJBERUUsKSURE1JJCEhERtaSQRERELSkkERFRSwpJRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbX0pZBIWizpZkkLJS0osZdLukLS78u/mzVtf5ykRZLulLR3U3xmOc4iSadIUj/aExExnvWzR/JW2zNszyrPjwV+YXs68IvyHEk7AnOAnYDZwLckrVf2ORWYC0wvj9k9zD8iIli3hrb2A84uy2cD+zfFz7f9lO17gEXAbpK2Aibavta2gXOa9omIiB7pVyEx8HNJN0qaW2Jb2l4GUP7dosQnA0ua9h0usclleWQ8IiJ6aEKfzvsm20slbQFcIemOMbZtNe/hMeKrH6AqVnMBpk6dura5RkTEGPrSI7G9tPz7IHAJsBvwQBmuovz7YNl8GNi6afcpwNISn9Ii3up8p9ueZXvW0NBQJ5sSETHu9byQSHqppI0by8BfA7cA84HDymaHAZeW5fnAHEkbSNqWalL9hjL8tULS7uVqrUOb9omIiB7px9DWlsAl5UrdCcD3bf9U0q+BCyUdDtwHHARg+1ZJFwK3ASuBo22vKsc6CjgL2BC4vDwiIqKHel5IbN8NvL5F/GFgr1H2mQfMaxFfAOzc6RwjIqJ969LlvxERMYBSSCIiopYUkoiIqCWFJCIiakkhiYiIWlJIIiKilhSSiIioJYUkIiJqSSGJiIhaUkgiIqKWFJKIiKglhSQiImpJIYmIiFpSSCIiopYUkoiIqCWFJCIiakkhiYiIWlJIIiKilhSSiIioJYUkIiJq6XkhkbS1pP8r6XZJt0r6RImfKOl+SQvLY9+mfY6TtEjSnZL2borPlHRzWXeKJPW6PRER492EPpxzJfBp27+RtDFwo6QryrqTbf9T88aSdgTmADsBrwSulPRq26uAU4G5wHXAZcBs4PIetSMiIuhDj8T2Mtu/KcsrgNuByWPssh9wvu2nbN8DLAJ2k7QVMNH2tbYNnAPs3+X0IyJihL7OkUiaBrwBuL6EPibpJklnStqsxCYDS5p2Gy6xyWV5ZLzVeeZKWiBpwfLlyzvYgoiI6FshkfQy4GLgk7Yfpxqm2h6YASwDvtLYtMXuHiO+etA+3fYs27OGhoZq5x4REX/Rl0Ii6cVUReR7tn8IYPsB26tsPwN8G9itbD4MbN20+xRgaYlPaRGPiIge6sdVWwLOAG63/dWm+FZNmx0A3FKW5wNzJG0gaVtgOnCD7WXACkm7l2MeClzak0ZERMSz+nHV1puA9wM3S1pYYscDB0uaQTU8tRg4AsD2rZIuBG6juuLr6HLFFsBRwFnAhlRXa+WKrYiIHut5IbH9S1rPb1w2xj7zgHkt4guAnTuXXURErK18sj0iImpJIYmIiFpSSCIiopYUkoiIqCWFJCIiakkhiYiIWlJIIiKilhSSiIioJYUkIiJq6cctUgbWtGN/Umv/xV9+Z4cyiYhYd6RHEhERtaSQRERELSkkERFRSwpJRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbXkk+0RET1U9w4Z66KBLySSZgNfB9YDvmP7y31OKSJe4F6IxaCOgS4kktYD/hl4BzAM/FrSfNu39TeziN6r8+ZW5z5w/TpvXSkGnTPQhQTYDVhk+24ASecD+wHrZCHJL26sq/r1u5n/Ey8Mg15IJgNLmp4PA28cuZGkucDc8vQPku58nuebBDz0PPcdVGnz+JA2jwM6qVabtxltxaAXErWIebWAfTpweu2TSQtsz6p7nEGSNo8PafP40K02D/rlv8PA1k3PpwBL+5RLRMS4NOiF5NfAdEnbSlofmAPM73NOERHjykAPbdleKeljwM+oLv890/atXTxl7eGxAZQ2jw9p8/jQlTbLXm1KISIiom2DPrQVERF9lkISERG1pJC0IGm2pDslLZJ0bIv1knRKWX+TpF36kWcntdHmQ0pbb5L0K0mv70eenbSmNjdtt6ukVZIO7GV+3dBOmyXtKWmhpFslXd3rHDupjd/rTST9WNLvSns/2I88O0nSmZIelHTLKOs7//5lO4+mB9Wk/V3AdsD6wO+AHUdssy9wOdXnWHYHru933j1o838DNivL+4yHNjdtdxVwGXBgv/Puwc95U6o7Q0wtz7fod95dbu/xwElleQh4BFi/37nXbPdbgF2AW0ZZ3/H3r/RIVvfsbVdsPw00brvSbD/gHFeuAzaVtFWvE+2gNbbZ9q9sP1qeXkf1mZ1B1s7PGeDjwMXAg71MrkvaafPfAj+0fR+A7UFudzvtNbCxJAEvoyokK3ubZmfZvoaqHaPp+PtXCsnqWt12ZfLz2GaQrG17Dqf6i2aQrbHNkiYDBwCn9TCvbmrn5/xqYDNJ/ybpRkmH9iy7zmunvd8EXkv1QeabgU/YfqY36fVNx9+/BvpzJF3Szm1X2ro1ywBpuz2S3kpVSN7c1Yy6r502fw34jO1V1R+sA6+dNk8AZgJ7ARsC10q6zvZ/dDu5LminvXsDC4G3AdsDV0j6d9uPdzu5Pur4+1cKyeraue3KC+3WLG21R9LrgO8A+9h+uEe5dUs7bZ4FnF+KyCRgX0krbf+oNyl2XLu/2w/ZfgJ4QtI1wOuBQSwk7bT3g8CXXU0eLJJ0D/Aa4IbepNgXHX//ytDW6tq57cp84NBy9cPuwGO2l/U60Q5aY5slTQV+CLx/QP86HWmNbba9re1ptqcBFwEfHeAiAu39bl8K7CFpgqSNqO6mfXuP8+yUdtp7H1XvC0lbAjsAd/c0y97r+PtXeiQjeJTbrkg6sqw/jeoKnn2BRcAfqf6qGVhttvkLwObAt8pf6Cs9wHdObbPNLyjttNn27ZJ+CtwEPEP1raMtLyNd17X5M/6fwFmSbqYa8vmM7YG+tbyk84A9gUmShoETgBdD996/couUiIioJUNbERFRSwpJRETUkkISERG1pJBEREQtKSQREVFLCklERNSSQhIREbX8f8WBn01ImiMsAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "tf.keras.backend.set_floatx('float64')\n",
    "pred_dicts = list(linear_est.predict(eval_input_fn))\n",
    "probs = pd.Series([pred['probabilities'][1] for pred in pred_dicts])\n",
    "\n",
    "probs.plot(kind='hist', bins=20, title='predicted probabilities')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "linear/linear_model/alive_players_ct/weights [[-0.1397873]]\n",
      "linear/linear_model/alive_players_t/weights [[0.9546902]]\n",
      "linear/linear_model/bias_weights [1.1570525]\n",
      "linear/linear_model/health_ct/weights [[-0.08894149]]\n",
      "linear/linear_model/health_t/weights [[0.07982726]]\n",
      "linear/linear_model/map/weights [[ 1.7082771 ]\n",
      " [ 0.03408846]\n",
      " [-0.10059735]\n",
      " [ 2.476449  ]\n",
      " [ 0.0371594 ]\n",
      " [ 0.7160883 ]\n",
      " [-0.7148429 ]\n",
      " [-1.5872557 ]]\n",
      "linear/linear_model/money_ct/weights [[0.00167062]]\n",
      "linear/linear_model/money_t/weights [[0.00091042]]\n",
      "linear/linear_model/round_status/weights [[-1.4987187 ]\n",
      " [ 0.79179937]\n",
      " [ 8.228458  ]]\n",
      "linear/linear_model/round_status_time_left/weights [[-0.02394005]]\n",
      "[VocabularyListCategoricalColumn(key='round_status', vocabulary_list=('FreezeTime', 'Normal', 'BombPlanted'), dtype=tf.string, default_value=-1, num_oov_buckets=0), VocabularyListCategoricalColumn(key='map', vocabulary_list=('de_dust2', 'de_mirage', 'de_nuke', 'de_inferno', 'de_overpass', 'de_vertigo', 'de_train', 'de_cache'), dtype=tf.string, default_value=-1, num_oov_buckets=0), NumericColumn(key='round_status_time_left', shape=(1,), default_value=None, dtype=tf.float64, normalizer_fn=None), NumericColumn(key='alive_players_t', shape=(1,), default_value=None, dtype=tf.int32, normalizer_fn=None), NumericColumn(key='alive_players_ct', shape=(1,), default_value=None, dtype=tf.int32, normalizer_fn=None), NumericColumn(key='health_t', shape=(1,), default_value=None, dtype=tf.int32, normalizer_fn=None), NumericColumn(key='health_ct', shape=(1,), default_value=None, dtype=tf.int32, normalizer_fn=None), NumericColumn(key='money_ct', shape=(1,), default_value=None, dtype=tf.int32, normalizer_fn=None), NumericColumn(key='money_t', shape=(1,), default_value=None, dtype=tf.int32, normalizer_fn=None)]\n",
      "round_status FreezeTime\n",
      "map de_dust2\n",
      "round_status_time_left 20.0\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 10150\n",
      "money_t 23300\n",
      "\n",
      "Win probability: CT: 1.8e-17 T: 1.0\n",
      "\n",
      "round_status FreezeTime\n",
      "map de_dust2\n",
      "round_status_time_left 20.0\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 10150\n",
      "money_t 18700\n",
      "\n",
      "Win probability: CT: 1.2e-15 T: 1.0\n",
      "\n",
      "round_status FreezeTime\n",
      "map de_dust2\n",
      "round_status_time_left 20.0\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 10150\n",
      "money_t 18700\n",
      "\n",
      "Win probability: CT: 1.2e-15 T: 1.0\n",
      "\n",
      "round_status FreezeTime\n",
      "map de_dust2\n",
      "round_status_time_left 19.921875\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 10150\n",
      "money_t 18700\n",
      "\n",
      "Win probability: CT: 1.2e-15 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 114.890625\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 9650\n",
      "money_t 18700\n",
      "\n",
      "Win probability: CT: 2.6e-15 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 94.89844\n",
      "alive_players_t 5\n",
      "alive_players_ct 4\n",
      "health_t 485\n",
      "health_ct 400\n",
      "money_ct 7650\n",
      "money_t 19000\n",
      "\n",
      "Win probability: CT: 1.4e-17 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 74.921875\n",
      "alive_players_t 4\n",
      "alive_players_ct 3\n",
      "health_t 400\n",
      "health_ct 196\n",
      "money_ct 5600\n",
      "money_t 14650\n",
      "\n",
      "Win probability: CT: 3.7e-19 T: 1.0\n",
      "\n",
      "round_status BombPlanted\n",
      "map de_dust2\n",
      "round_status_time_left 32.617188\n",
      "alive_players_t 4\n",
      "alive_players_ct 2\n",
      "health_t 337\n",
      "health_ct 168\n",
      "money_ct 3400\n",
      "money_t 15250\n",
      "\n",
      "Win probability: CT: 2e-20 T: 1.0\n",
      "\n",
      "round_status BombPlanted\n",
      "map de_dust2\n",
      "round_status_time_left 12.6171875\n",
      "alive_players_t 4\n",
      "alive_players_ct 2\n",
      "health_t 337\n",
      "health_ct 168\n",
      "money_ct 3400\n",
      "money_t 15250\n",
      "\n",
      "Win probability: CT: 1.2e-20 T: 1.0\n",
      "\n",
      "round_status FreezeTime\n",
      "map de_dust2\n",
      "round_status_time_left 19.90625\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 21950\n",
      "money_t 37400\n",
      "\n",
      "Win probability: CT: 1.3e-31 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 114.91406\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 1700\n",
      "money_t 29900\n",
      "\n",
      "Win probability: CT: 5.8e-14 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 94.91406\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 1700\n",
      "money_t 29900\n",
      "\n",
      "Win probability: CT: 3.6e-14 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 74.91406\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 1700\n",
      "money_t 29900\n",
      "\n",
      "Win probability: CT: 2.2e-14 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 54.945312\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 1700\n",
      "money_t 29900\n",
      "\n",
      "Win probability: CT: 1.4e-14 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 34.945312\n",
      "alive_players_t 2\n",
      "alive_players_ct 3\n",
      "health_t 200\n",
      "health_ct 246\n",
      "money_ct 1000\n",
      "money_t 12100\n",
      "\n",
      "Win probability: CT: 1.5e-05 T: 1.0\n",
      "\n",
      "round_status FreezeTime\n",
      "map de_dust2\n",
      "round_status_time_left 19.929688\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 19250\n",
      "money_t 37300\n",
      "\n",
      "Win probability: CT: 1.3e-29 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 114.9375\n",
      "alive_players_t 5\n",
      "alive_players_ct 5\n",
      "health_t 500\n",
      "health_ct 500\n",
      "money_ct 3550\n",
      "money_t 10750\n",
      "\n",
      "Win probability: CT: 9.8e-08 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 94.890625\n",
      "alive_players_t 3\n",
      "alive_players_ct 3\n",
      "health_t 300\n",
      "health_ct 300\n",
      "money_ct 2400\n",
      "money_t 8150\n",
      "\n",
      "Win probability: CT: 3.6e-06 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 74.890625\n",
      "alive_players_t 3\n",
      "alive_players_ct 3\n",
      "health_t 277\n",
      "health_ct 300\n",
      "money_ct 2400\n",
      "money_t 8150\n",
      "\n",
      "Win probability: CT: 1.4e-05 T: 1.0\n",
      "\n",
      "round_status Normal\n",
      "map de_dust2\n",
      "round_status_time_left 54.890625\n",
      "alive_players_t 3\n",
      "alive_players_ct 3\n",
      "health_t 277\n",
      "health_ct 300\n",
      "money_ct 2400\n",
      "money_t 8150\n",
      "\n",
      "Win probability: CT: 8.7e-06 T: 1.0\n",
      "\n"
     ]
    }
   ],
   "source": [
    "tf.get_logger().setLevel('WARN')\n",
    "\n",
    "def predict_idx(data_df, idx):\n",
    "    def input_function():\n",
    "        ds = tf.data.Dataset.from_tensor_slices(dict(data_df.filter(items=feature_names)))\n",
    "        ds = ds.skip(idx).take(1).batch(1)\n",
    "        return ds\n",
    "    return input_function\n",
    "\n",
    "def predict_index(ds, idx):\n",
    "    for feature in feature_names:\n",
    "        print(feature, ds[feature][idx])\n",
    "        \n",
    "    print(\"\")\n",
    "    \n",
    "    for pred in linear_est.predict(predict_idx(ds, idx)):\n",
    "        probs = pred['probabilities']\n",
    "        print(\"Win probability: CT: {:.2} T: {:.2}\".format(probs[0], probs[1]))\n",
    "\n",
    "    print(\"\")\n",
    "\n",
    "wt_names = linear_est.get_variable_names()\n",
    "for name in wt_names:\n",
    "    if name.endswith(\"weights\"):\n",
    "        val = linear_est.get_variable_value(name)\n",
    "        print(name, val)\n",
    "\n",
    "print(feature_columns)\n",
    "\n",
    "for i in range(100,120):\n",
    "    predict_index(dfeval, i)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}