sample.py

import gym
import universe
import random


#reinforcement learning step
def determine_turn(turn, observation_n, j, total_sum, prev_total_sum, reward_n):
	#for every 15 iterations, sum the total observations, and take the average
	#if lower than 0, change the direction
	#if we go 15+ iterations and get a reward each step, we're doing something right
	#thats when we turn
	if(j >= 15):
		if(total_sum/ j ) < 0:
			turn = True
		else:
			turn = False

		#reset vars
		j = 0
		prev_total_sum = total_sum
		total_sum = 0

	else:
		turn = False
	if(observation_n != None):
		#increment counter and reward sum
		j+=1
		total_sum += reward_n
	return(turn, j, total_sum, prev_total_sum)




def main():

	#init environment
	env = gym.make('CartPole-v0')
	#env.configure(remotes="vnc://localhost:5900+15901")
	observation_n = env.reset()

	#init variables
	# num of game iterations
	n = 0
	j = 0
	#sum of observations
	total_sum = 0
	prev_total_sum = 0
	turn = False

	#define our turns or keyboard actions
	left = [('KeyEvent', 'ArrowUp', True) ,('KeyEvent', 'ArrowLeft', True), ('KeyEvent', 'ArrowRight', False)]
	right = [('KeyEvent', 'ArrowUp', True) ,('KeyEvent', 'ArrowLeft', False), ('KeyEvent', 'ArrowRight', True)]
	Forward = [('KeyEvent', 'ArrowUp', True) ,('KeyEvent', 'ArrowLeft', False), ('KeyEvent', 'ArrowRight', False)]


	#main logic
	while True:
		#increment a counter for number of iterations
		n+=1

		#if at least one iteration is made, check if turn is needed
		if(n > 1):
			#if at least one iteration, check if a turn
			if(observation_n[0] != None):
				#store the reward in the previous score
				prev_score = reward_n[0]

				#should we turn?
				if(turn):
					#pick a random event
					#where to turn? 
					event = random.choice([left,right])
					#perform an action
					action_n = [event for ob in observation_n]
					#set turn to false
					turn = False

		elif(~turn):
			#if no turn is needed, go straight
			action_n = [Forward for ob in observation_n]


		#if there is an obseravtion, game has started, check if turn needed
		if(observation_n[0] != None):
			turn, j, total_sum, prev_total_sum = determine_turn(turn, observation_n[0], j, total_sum, prev_total_sum, reward_n[0])

		#save new variables for each iteration
		observation_n, reward_n, done_n, info = env.step(action_n)

		env.render()

if __name__ == '__main__':
	main()