frozen_lake2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Define the Next State and Error Corretion Techniques\n",
    "We will use Gradient optimizer to reduce the loss with a learning rate of 0.1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "Q2 = tf.placeholder(shape=[1,4],dtype=tf.float32)\n",
    "loss = tf.reduce_sum(tf.square(Q2 - Q1))\n",
    "gdo = tf.train.GradientDescentOptimizer(learning_rate=0.1)\n",
    "updatedweights = gdo.minimize(loss)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Lets build our model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "gamma = 0.9\n",
    "epsilon = 0.1\n",
    "episodes = 2000\n",
    "\n",
    "totalReward = 0\n",
    "\n",
    "session = tf.Session()\n",
    "session.run(tf.initialize_all_variables())\n",
    "for i in range(episodes):\n",
    "    state_now = env.reset()\n",
    "    done = False\n",
    "    reward = 0\n",
    "    for j in range(100):\n",
    "        #Lets find the dot product of the inputs with the weights and apply argmax on it.\n",
    "        action , Y = session.run([output, Q1], feed_dict = {inputs : [np.eye(16)[state_now]]})\n",
    "        if epsilon > np.random.rand(1):\n",
    "            action[0] = env.action_space.sample()\n",
    "            epsilon -= 10**-3\n",
    "        #Lets iterate to the next state Note: This can be random.\n",
    "        state_next , reward, done, _ = env.step(action[0])\n",
    "        Y1 = session.run(Q1, feed_dict = {inputs : [np.eye(16)[state_next]]})\n",
    "        change_Y = Y\n",
    "        change_Y[0, action[0]] = reward + gamma*np.max(Y1)\n",
    "        #Updating the weights \n",
    "        _,new_weights = session.run([updatedweights,weights],feed_dict={inputs:[np.eye(16)[state_now]],Q2:change_Y})\n",
    "        #Lets append the total number of rewards\n",
    "        totalReward += reward\n",
    "        state_now = state_next\n",
    "        if reward == 1:\n",
    "            print ('Episode {} was successful, Agent reached the Goal'.format(i))\n",
    "            \n",
    "session.close()"
   ]
  }
 ],
 "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
}