def compatible_2():
print('+++++++++++++++++++++++++++++++++++++++++++++++++')
fruit_env = GymEnvironment(env_name='CartPole-v1')
state = fruit_env.get_state_space()
print(state.get_range())
print(tuple(state.get_shape()))
print(fruit_env.get_action_space().get_range())
print(fruit_env.reset())
print(fruit_env.get_state())
print('+++++++++++++++++++++++++++++++++++++++++++++++++')
print('+++++++++++++++++++++++++++++++++++++++++++++++++')
env = OpenAIGym(level='CartPole-v1')
state = env.states()
print(state)
print(env.actions())
print(env.reset())
print(env.execute(0))
print(env.max_episode_timesteps())
print('+++++++++++++++++++++++++++++++++++++++++++++++++')
print('+++++++++++++++++++++++++++++++++++++++++++++++++')
env = TFEnvironment(fruit_environment=fruit_env)
print(env.states())
print(env.actions())
print(env.getrobotics_states())
print(env.execute(0))
print(env.max_episode_timesteps())
print('+++++++++++++++++++++++++++++++++++++++++++++++++')
def main():
#Creates a log for MineRL
#logging.basicConfig(level=logging.DEBUG)
# Create the environment
ENV_NAME = "MineRLTreechop-v0"
# Pre-defined or custom environment
env = gym.make(ENV_NAME)
environment = OpenAIGym(env)
agent = Agent.create(agent='ac',
environment=environment,
max_episode_timesteps=8000,
exploration=.03,
critic_optimizer='evolutionary')
sum_rewards = 0.0
rewards_by_episode = []
for _ in range(200):
states = environment.reset()
terminal = False
print("Training episode " + str(_))
while not terminal:
actions = agent.act(states=states, evaluation=True)
states, terminal, reward = environment.execute(actions=actions)
sum_rewards += reward
#print(actions)
print("Sum reward so far: " + str(sum_rewards))
rewards_by_episode.append((_, sum_rewards))
print("Ending episode ", _)
print(rewards_by_episode)
print('Mean episode reward:', sum_rewards / 200)
agent.close()
environment.close()
dict(type='dense', size=64),
dict(type='dense', size=64)
],
learning_rate=1e-3,
name='agent_loader'
)
# import ipdb;ipdb.set_trace()
agent = agent.load()
running_score = 0.0
# Train for 300 episodes
for i_epoch in range(50000):
game_score = 0.0
# Initialize episode
states = environment.reset()
terminal = False
while not terminal:
# Episode timestep
actions = agent.act(states=states,evaluation=True)
states, terminal, reward = environment.execute(actions=actions)
game_score+=reward
# agent.observe(terminal=terminal, reward=reward)
running_score = 0.95*running_score + 0.05*game_score
if i_epoch%5==0:
print("Game ", i_epoch, " game score %.2f"%game_score," running score %.2f"%running_score)