gym_test.py

import gym
import numpy as np
import time


def run_test_env(env_name, perform_action = True, exit_on_done = True):

    env = gym.make(env_name)


    for i_episode in range(20):
        observation = env.reset()

        for t in range(100):
            env.render()

            if not perform_action:
                continue

            action = env.action_space.sample()

            print(action)

            observation, reward, done, info = env.step(action)
            '''Returns:
                observation (object): agent's observation of the current environment
                reward (float) : amount of reward returned after previous action
                done (boolean): whether the episode has ended, in which case further step() calls will return undefined results
                info (dict): contains auxiliary diagnostic information (helpful for debugging, and sometimes learning)
            '''

            #print("Observation %s reward %s done %s info $s" % (1, reward, done, 1))
            print(observation, reward, done, info)

            if done:
                print('Episode {} done in {} steps'.format(i_episode, t))
                if exit_on_done:
                    break

# use hill climbing algorithm, initialize weights randomly
# use memory to save good weights
def run_episode(env, parameters):
    observation  = env.reset()
    total_reward = 0

    for _ in range(200):
        env.render()
        # move left (0) or right (1)
        # initialize random weights
        action = 0 if np.matmul(parameters, observation) < 0 else 1
        observation, reward, done, info = env.step(action)

		# wait a little to make it easier to view the action behavior
        #time.sleep(0.2)

        total_reward += reward

        if done:
            print('Done')
            break

    return total_reward


def random_parameters(param_count):
    return np.random.rand(param_count) * 2 -1

#hill climbing
def train(submit):
    env = gym.make('CartPole-v0')

    noise_scaling = 0.1
    parameters = random_parameters(4)

    new_parameters = None
    best_reward = 0

    for i in range(2000):
        new_parameters = parameters + random_parameters(4) * noise_scaling
        reward = run_episode(env, new_parameters)

        print("Reward %d best %d : %f %f %f %f" % (reward, best_reward, *parameters))

        if reward > best_reward:
            print("New reward %d" % reward)
            best_reward = reward
            parameters = new_parameters

            if reward == 200:
                print(i)
                break

    return parameters

#r = train(submit = False)

#view different start positions
#run_test_env('CartPole-v0', perform_action=False, exit_on_done=False)

#view behavior of cart with random actions, do not exit environment 
#even if pole is past the recovery point
#run_test_env('CartPole-v0', perform_action=True, exit_on_done=False)


#run_test_env('CartPole-v0')