neural_net.py

import tensorflow as tf
import numpy as np
import gym

num_inputs = 4
num_hidden = 4
# Prob to go left   1-left = right
num_outputs = 1 

initializer = tf.contrib.layers.variance_scaling_initializer()


x = tf.placeholder(tf.float32, shape=[None, num_inputs])
hidden_layer_one = tf.layers.dense(x, num_hidden, activation=tf.nn.relu, kernel_initializer=initializer)
hidden_layer_two = tf.layers.dense(hidden_layer_one, num_hidden, activation=tf.nn.relu, kernel_initializer=initializer)
output_layer = tf.layers.dense(hidden_layer_two, num_outputs, activation=tf.nn.sigmoid, kernel_initializer=initializer)


probabilities = tf.concat(axis=1, values=[output_layer, 1-output_layer])
action = tf.multinomial(probabilities, num_samples=1)

init = tf.global_variables_initializer()

epi = 50
step_limit = 500

env = env = gym.make('CartPole-v0')
avg_steps = []


with tf.Session() as sess:
    init.run()

    for i_episode in range(epi):
        obs = env.reset()

        for step in range(step_limit):
            env.render()
            action_val = action.eval(feed_dict={x: obs.reshape(1, num_inputs)})
            obs, reward, done, info = env.step(action_val[0][0])
            # print(obs, reward, done, info)

            if done:
                avg_steps.append(step)
                # print("Done after {} steps".format(step))
                break

print("After {} episodes, average steps per game was {}".format(epi, np.mean(avg_steps)))
env.close()