generate_data.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import os\n",
    "import sys\n",
    "import numpy as np\n",
    "\n",
    "os.environ['KERAS_BACKEND'] = 'tensorflow'\n",
    "os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"7\"\n",
    "\n",
    "sys.path.append('/path/to/pommerman/playground/')\n",
    "sys.path.append('/path/to/evaluation_utils.py/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from copy import deepcopy\n",
    "from pommerman.agents import SimpleAgent\n",
    "from pommerman.configs import ffa_competition_env\n",
    "from pommerman.constants import BOARD_SIZE\n",
    "from pommerman.envs.v0 import Pomme\n",
    "from tqdm import tqdm_notebook as tqdm\n",
    "from matplotlib import pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 79,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "initial_rollouts   = 600\n",
    "min_episode_length = 35 \n",
    "\n",
    "train_data_path    = './dataset/'\n",
    "train_data_obs     = os.path.join(train_data_path, 'obs.npy')\n",
    "train_data_labels  = os.path.join(train_data_path, 'labels.npy')\n",
    "train_data_reward  = os.path.join(train_data_path, 'reward.npy')\n",
    "train_data_obs_map = os.path.join(train_data_path, 'obs_map.npy')\n",
    "train_data_obs_map_history = os.path.join(train_data_path, 'obs_map_history.npy')\n",
    "\n",
    "if not os.path.isdir(train_data_path):\n",
    "    os.makedirs(train_data_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class Episode:   \n",
    "    def __init__(self, agent_id, episode_id):\n",
    "        self.agent_id = agent_id\n",
    "        self.episode_id = episode_id\n",
    "        self.observations = []\n",
    "        self.actions = []\n",
    "        self.reward = []\n",
    "        self.done = False        \n",
    "        \n",
    "    def record(self, obs, action, reward):\n",
    "        self.observations.append(deepcopy(obs))\n",
    "        self.actions.append(deepcopy(action))\n",
    "        # Agent not win or not die\n",
    "        if reward != 0:\n",
    "            self.done = True\n",
    "            self.reward = [reward] * self.get_num_steps()\n",
    "        \n",
    "    def get_num_steps(self):\n",
    "        return len(self.observations)        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Environment wrapper\n",
    "class Stimulator:\n",
    "    def __init__(self, env):\n",
    "        self.env = env\n",
    "        self.episodes = []\n",
    "        \n",
    "    def record(self, agents, obs, actions, rewards):        \n",
    "        for agent_id in range(len(agents)): \n",
    "            agent = agents[agent_id]\n",
    "            # If an agent is dead (or win) we should not record a history\n",
    "            if not agent.done:    \n",
    "                agent.record(obs[agent_id], actions[agent_id], rewards[agent_id])\n",
    "\n",
    "    def stimulate(self, num_rollouts):\n",
    "        for cur_episode in tqdm(range(num_rollouts)):\n",
    "            # Create a history for each agent\n",
    "            agents = []                \n",
    "            for agent_id in range(4):\n",
    "                agents.append(Episode(agent_id, cur_episode))            \n",
    "            \n",
    "            done = False     \n",
    "            # Obtain initial observations\n",
    "            obs = self.env.reset()\n",
    "            try:\n",
    "                while not done:\n",
    "                    # FUCK self.env.act change \"POSITION\"!!!!\n",
    "                    obs_to_save = deepcopy(obs)                    \n",
    "                    # Produce actions\n",
    "                    actions = self.env.act(obs)\n",
    "                    # Make an episode step. Save an observations as new_obs, because we want to record previous one\n",
    "                    obs, rewards, done, _ = self.env.step(actions)\n",
    "                    # Record observations and actions\n",
    "                    self.record(agents, obs_to_save, actions, rewards)                    \n",
    "            except:\n",
    "                print(\"Error occurs\")\n",
    "                continue            \n",
    "            self.episodes.extend(agents)\n",
    "    \n",
    "    def get_episodes(self):\n",
    "        return deepcopy(self.episodes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[33mWARN: gym.spaces.Box autodetected dtype as <class 'numpy.float32'>. Please provide explicit dtype.\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "# Instantiate the environment\n",
    "config = ffa_competition_env()\n",
    "env = Pomme(**config[\"env_kwargs\"])\n",
    "\n",
    "# Random seed\n",
    "env.seed(0)\n",
    "\n",
    "# Add agents\n",
    "agents = [SimpleAgent(config[\"agent\"](agent_id, config[\"game_type\"])) for agent_id in range(4)]\n",
    "env.set_agents(agents)\n",
    "env.set_init_game_state(None)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "e8433f1fd7584f658b5ab4ef96702ddf",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=600), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "# Generate training data\n",
    "stimulator = Stimulator(env)\n",
    "stimulator.stimulate(initial_rollouts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "debbf9199d6d454cbecb529b4f01d422",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=2400), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "for episode in tqdm(stimulator.episodes):\n",
    "    for observation in episode.observations:\n",
    "        board = observation['board']\n",
    "        pos   = observation['position']    \n",
    "        if board[pos[0], pos[1]] != (episode.agent_id + 10):\n",
    "            print(\"ERROR\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Win/Lose in corners"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "wins = []\n",
    "loses = []\n",
    "\n",
    "for episode in stimulator.episodes:\n",
    "    if episode.reward[-1] == -1:\n",
    "        loses.append(episode.agent_id)\n",
    "    elif episode.reward[-1] == 1:\n",
    "        wins.append(episode.agent_id)\n",
    "    else:\n",
    "        print(\"FUCK\")\n",
    "\n",
    "plt.hist(loses, label='loses')\n",
    "plt.hist(wins, label='wins')\n",
    "plt.xticks(np.arange(4))\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([452, 496, 476, 519])"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.bincount(loses)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Episode length"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1224x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "episode_length = [episode.get_num_steps() for episode in stimulator.episodes]\n",
    "    \n",
    "plt.figure(figsize=(17,4))\n",
    "plt.bar(range(0, len(episode_length)), episode_length)\n",
    "plt.plot([0, len(episode_length)], [np.mean(episode_length), np.mean(episode_length)], 'r')\n",
    "plt.legend(['mean', 'episode length'])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10, 801, 151)"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.min(episode_length), np.max(episode_length), np.sum(np.array(episode_length) < min_episode_length)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Rewards"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1224x288 with 3 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "nr_wins = [0] * 4\n",
    "nr_loose = [0] * 4\n",
    "nr_tie = [0] * 4\n",
    "average_score = [[], [], [], []]\n",
    "\n",
    "for episode in stimulator.episodes:\n",
    "    if episode.reward[-1] == 1:\n",
    "        nr_wins[episode.agent_id] += 1\n",
    "    else:\n",
    "        nr_loose[episode.agent_id] += 1\n",
    "    average_score[episode.agent_id].append(episode.reward[-1])\n",
    "average_score[0] = np.mean(average_score[0])\n",
    "average_score[1] = np.mean(average_score[1])\n",
    "average_score[2] = np.mean(average_score[2])\n",
    "average_score[3] = np.mean(average_score[3])\n",
    "\n",
    "plt.figure(figsize=(17,4))\n",
    "plt.subplot(141)\n",
    "plt.bar(range(0,4), nr_wins)\n",
    "plt.title('nr_wins')\n",
    "plt.subplot(142)\n",
    "plt.bar(range(0,4), nr_loose)\n",
    "plt.title('nr_loose')\n",
    "plt.subplot(143)\n",
    "plt.bar(range(0,4), average_score)\n",
    "plt.title('average_score')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Consequtive actions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "actions_length = []\n",
    "for episode in stimulator.episodes:\n",
    "    current_action = episode.actions[0]\n",
    "    action_length = 0\n",
    "    for action in episode.actions:\n",
    "        if current_action != action:\n",
    "            actions_length.append(action_length)            \n",
    "            current_action = action\n",
    "            action_length = 0\n",
    "        action_length += 1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1.6890224514546033, 1.0, 515)"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(actions_length), np.median(actions_length), np.max(actions_length)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1224x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(17,10))\n",
    "plt.plot(actions_length)\n",
    "plt.plot([np.mean(actions_length)] * len(actions_length))\n",
    "plt.title('action length')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Actions frequency"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1224x720 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "actions_count = np.zeros(6)\n",
    "for episode in stimulator.episodes:    \n",
    "    for action in episode.actions:\n",
    "        actions_count[action] += 1\n",
    "        \n",
    "plt.figure(figsize=(17,10))\n",
    "plt.bar(range(0,6), actions_count)\n",
    "plt.title('actions count')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Some data preprocessing"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Remove episodes with too small length. Maybe it is suicide."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "2251"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "episodes = [episode for episode in stimulator.episodes if episode.get_num_steps() > min_episode_length or episode.reward[-1] == 1]\n",
    "len(episodes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's remove consequtive actions. I will use threshold 12."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "05d8b36b07b747e293db0625f8e906b6",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=2202), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n"
     ]
    }
   ],
   "source": [
    "threshold = 12\n",
    "\n",
    "for episode in tqdm(episodes):\n",
    "    # New observations\n",
    "    obs = []\n",
    "    act = []\n",
    "    rew = []\n",
    "    \n",
    "    # Initial values\n",
    "    cur_action = episode.actions[0]\n",
    "    seq_length = 0\n",
    "    for ind in range(len(episode.actions)):\n",
    "        # Current action\n",
    "        action = episode.actions[ind]\n",
    "        \n",
    "        # If an action the same\n",
    "        if action == cur_action:\n",
    "            seq_length += 1\n",
    "        else:\n",
    "            seq_length = 0\n",
    "            cur_action = action\n",
    "        # If a sequence is less than threashold\n",
    "        if seq_length <= threshold:\n",
    "            obs.append(episode.observations[ind])\n",
    "            act.append(episode.actions[ind])\n",
    "            rew.append(episode.reward[ind])\n",
    "    \n",
    "    # Save preprocessed observations\n",
    "    episode.actions      = act\n",
    "    episode.observations = obs\n",
    "    episode.reward       = rew"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's rerun plots."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Save stuff"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "observations = []\n",
    "actions = []\n",
    "rewards = []\n",
    "for episode in stimulator.episodes:\n",
    "    observations.append(episode.observations)\n",
    "    actions.append(episode.actions)\n",
    "    rewards.append(episode.reward)\n",
    "\n",
    "observations_merged = np.concatenate(observations)\n",
    "actions_merged = np.concatenate(actions)\n",
    "rewards_merged = np.concatenate(rewards)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "np.save(train_data_obs, observations_merged)\n",
    "np.save(train_data_labels, actions_merged)\n",
    "np.save(train_data_reward, rewards_merged)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "observations_merged = np.load(train_data_obs)\n",
    "actions_merged      = np.load(train_data_labels)\n",
    "rewards_merged      = np.load(train_data_reward)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((670725,), (670725,), (670725,))"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "observations_merged.shape, actions_merged.shape, rewards_merged.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'alive': [10, 11, 12, 13],\n",
       " 'ammo': 1,\n",
       " 'blast_strength': 2,\n",
       " 'board': array([[ 0,  0,  0,  2,  1,  1,  1,  0,  0,  1,  2],\n",
       "        [ 0, 10,  0,  0,  2,  2,  2,  0,  0, 13,  2],\n",
       "        [ 0,  0,  0,  2,  1,  1,  2,  2,  1,  0,  0],\n",
       "        [ 2,  0,  2,  0,  0,  2,  1,  2,  1,  0,  2],\n",
       "        [ 1,  2,  1,  0,  0,  0,  1,  2,  1,  2,  2],\n",
       "        [ 1,  2,  1,  2,  0,  0,  1,  1,  1,  2,  1],\n",
       "        [ 1,  2,  2,  1,  1,  1,  0,  1,  1,  2,  2],\n",
       "        [ 0,  0,  2,  2,  2,  1,  1,  0,  0,  0,  1],\n",
       "        [ 0,  0,  1,  1,  1,  1,  1,  0,  0,  0,  0],\n",
       "        [ 1, 11,  0,  0,  2,  2,  2,  0,  0, 12,  0],\n",
       "        [ 2,  2,  0,  2,  2,  1,  2,  1,  0,  0,  0]], dtype=uint8),\n",
       " 'bomb_blast_strength': array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]]),\n",
       " 'bomb_life': array([[0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n",
       "        [0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0.]]),\n",
       " 'can_kick': False,\n",
       " 'enemies': [<Item.Agent1: 11>, <Item.Agent2: 12>, <Item.Agent3: 13>],\n",
       " 'position': (1, 1),\n",
       " 'teammate': <Item.AgentDummy: 9>}"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "observations_merged[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Featurize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def make_centered(board, position, view_size=BOARD_SIZE * 2 - 1, crop=False):\n",
    "    # If it is a perk channel, just return the resized full array\n",
    "    if np.all(board == 1):\n",
    "        return np.full((view_size, view_size, 1), 1)\n",
    "    \n",
    "    # make sure of odd view_size\n",
    "    view_size = view_size + 1 if view_size % 2 == 0 else view_size \n",
    "    \n",
    "    # TODO: what should be the value of the off world? maybe defining an edge channel?\n",
    "    agent_view = np.zeros((view_size, view_size, 1)) # agent centric full-world coverage\n",
    "    center = (view_size // 2 + 1, view_size // 2 + 1)\n",
    "    \n",
    "    # copy board to the new view\n",
    "    offset_y = center[0] - position[0] - 1\n",
    "    offset_x = center[1] - position[1] - 1\n",
    "    agent_view[offset_y:offset_y+BOARD_SIZE, offset_x:offset_x+BOARD_SIZE, :] = board\n",
    "    \n",
    "    # finalize view size\n",
    "    r = BOARD_SIZE // 2\n",
    "    start, end = center[0]-r-1, center[0]+r\n",
    "    if crop:\n",
    "        agent_view = agent_view[start:end, start:end, :]\n",
    "    \n",
    "    return np.array(agent_view, dtype=np.float32)\n",
    "\n",
    "def featurize(obs, center=True, crop=False):\n",
    "    shape = (BOARD_SIZE, BOARD_SIZE, 1)\n",
    "\n",
    "    def get_matrix(board, key):\n",
    "        res = board[key]\n",
    "        return res.reshape(shape).astype(np.float32)\n",
    "\n",
    "    def get_map(board, item):\n",
    "        map = np.zeros(shape)\n",
    "        map[board == item] = 1\n",
    "        return map\n",
    "\n",
    "    board = get_matrix(obs, 'board')\n",
    "\n",
    "    path_map       = get_map(board, 0)          # Empty space\n",
    "    rigid_map      = get_map(board, 1)          # Rigid = 1\n",
    "    wood_map       = get_map(board, 2)          # Wood = 2\n",
    "    bomb_map       = get_map(board, 3)          # Bomb = 3\n",
    "    flames_map     = get_map(board, 4)          # Flames = 4\n",
    "    fog_map        = get_map(board, 5)          # TODO: not used for first two stages Fog = 5\n",
    "    extra_bomb_map = get_map(board, 6)          # ExtraBomb = 6\n",
    "    incr_range_map = get_map(board, 7)          # IncrRange = 7\n",
    "    kick_map       = get_map(board, 8)          # Kick = 8\n",
    "    skull_map      = get_map(board, 9)          # Skull = 9\n",
    "\n",
    "    position = obs[\"position\"]\n",
    "    my_position = np.zeros(shape)\n",
    "    my_position[position[0], position[1], 0] = 1\n",
    "\n",
    "    team_mates = get_map(board, obs[\"teammate\"].value) # TODO during documentation it should be an array\n",
    "\n",
    "    enemies = np.zeros(shape)\n",
    "    for enemy in obs[\"enemies\"]:\n",
    "        enemies[board == enemy.value] = 1\n",
    "\n",
    "    bomb_blast_strength = get_matrix(obs, 'bomb_blast_strength')\n",
    "    bomb_life           = get_matrix(obs, 'bomb_life')\n",
    "\n",
    "    ammo           = np.full((BOARD_SIZE, BOARD_SIZE, 1), obs[\"ammo\"])\n",
    "    blast_strength = np.full((BOARD_SIZE, BOARD_SIZE, 1), obs[\"blast_strength\"])\n",
    "    can_kick       = np.full((BOARD_SIZE, BOARD_SIZE, 1), int(obs[\"can_kick\"]))\n",
    "    \n",
    "    maps = [my_position, enemies, team_mates, path_map, rigid_map, \n",
    "                          wood_map, bomb_map, flames_map, fog_map, extra_bomb_map,\n",
    "                          incr_range_map, kick_map, skull_map, bomb_blast_strength,\n",
    "                          bomb_life, ammo, blast_strength, can_kick]\n",
    "    \n",
    "    if center:\n",
    "        maps = [make_centered(m, position, crop=crop) for m in maps]\n",
    "    \n",
    "    obs = np.concatenate(maps, axis=2)\n",
    "    return obs.astype(np.int16)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "4015a83dee4346f8a9b35c77ca9d969a",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=2400), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "2400"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "observations_map = [np.array([featurize(obs, center=True, crop=False) for obs in episode]) for episode in tqdm(observations)]\n",
    "len(observations_map)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Check that our agent is in the center"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],\n",
       "      dtype=int16)"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "observations_map[0][0][:,:,0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 67,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 1, 1, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 1, 1, 0, 0, 0, 0, 1, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 1, 1, 0, 1, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1, 0],\n",
       "       [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]],\n",
       "      dtype=int16)"
      ]
     },
     "execution_count": 67,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "observations_map[0][0][:,:,3]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Add previous frames as history"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def get_history(observations, i, axis=0):\n",
    "    if i < 4:\n",
    "        n_first = 4 - i\n",
    "        return np.concatenate([np.array([observations[0] for _ in range(n_first)]), np.array(observations[:i]).reshape(i, 21, 21, 18)], axis=0)\n",
    "    return np.stack(observations[i-4:i], axis)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2, 4, 21, 21, 18)"
      ]
     },
     "execution_count": 71,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.array([get_history(observations_map[0], 0), get_history(observations_map[0], 1)]).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "application/vnd.jupyter.widget-view+json": {
       "model_id": "64f27ae2e48348889837a571903b6227",
       "version_major": 2,
       "version_minor": 0
      },
      "text/plain": [
       "HBox(children=(IntProgress(value=0, max=2400), HTML(value='')))"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "observations_map_history = [np.array([get_history(episode, i) for i in range(len(episode))], dtype=np.uint8) for episode in tqdm(observations_map)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "observations_map_history = np.concatenate(observations_map_history).astype(np.uint8)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 80,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(670725, 4, 21, 21, 18)"
      ]
     },
     "execution_count": 80,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "observations_map_history.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 81,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "np.save(train_data_obs_map_history, observations_map_history)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 82,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2400, 670725)"
      ]
     },
     "execution_count": 82,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "episode_lengths = np.array([len(episode) for episode in observations])\n",
    "len(episode_lengths), np.sum(episode_lengths)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "np.save('./dataset/episode_lengths.npy', episode_lengths)"
   ]
  }
 ],
 "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": "3.5.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}