{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# Replay Memory"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "\n",
    "\n",
    "Now we build the experience replay buffer which is used for storing all the agent's\n",
    "experience. We sample minibatch of experience from the replay buffer for training the\n",
    "network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class ReplayMemoryFast:\n",
    "\n",
    "\n",
    "    # first we define init method and initialize buffer size\n",
    "    def __init__(self, memory_size, minibatch_size):\n",
    "\n",
    "        # max number of samples to store\n",
    "        self.memory_size = memory_size\n",
    "\n",
    "        # mini batch size\n",
    "        self.minibatch_size = minibatch_size\n",
    "\n",
    "        self.experience = [None]*self.memory_size  \n",
    "        self.current_index = 0\n",
    "        self.size = 0\n",
    "\n",
    "\n",
    "    # next we define the function called store for storing the experiences\n",
    "    def store(self, observation, action, reward, newobservation, is_terminal):\n",
    "\n",
    "        # store the experience as a tuple (current state, action, reward, next state, is it a terminal state)\n",
    "        self.experience[self.current_index] = (observation, action, reward, newobservation, is_terminal)\n",
    "        self.current_index += 1\n",
    "\n",
    "        self.size = min(self.size+1, self.memory_size)\n",
    "               \n",
    "        # if the index is greater than  memory then we flush the index by subtrating it with memory size\n",
    "\n",
    "        if self.current_index >= self.memory_size:\n",
    "            self.current_index -= self.memory_size\n",
    "\n",
    "\n",
    "\n",
    "    # we define a function called sample for sampling the minibatch of experience\n",
    "\n",
    "    def sample(self):\n",
    "        if self.size <  self.minibatch_size:\n",
    "            return []\n",
    "\n",
    "        # first we randomly sample some indices\n",
    "        samples_index  = np.floor(np.random.random((self.minibatch_size,))*self.size)\n",
    "\n",
    "        # select the experience from the sampled index\n",
    "        samples = [self.experience[int(i)] for i in samples_index]\n",
    "\n",
    "        return samples"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:anaconda]",
   "language": "python",
   "name": "conda-env-anaconda-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}