def act(self, action_idx):
"""
takes an action and executes it
returns (next_state_observation, reward, done, extra_info)
"""
# convert the action index into the actual action
action_idx = int(action_idx)
action = self.actions[action_idx]
self.game.act(action)
self.frame += 1
r = self.game.reward()
config = self.game.observe()
obs, info = config['observation']
featurizers.grid_one_hot(self.game, obs)
obs = np.array(obs)
done = self.game.is_over() or self.frame >= self.max_frame
total_reward = self.game.reward_so_far()
return (obs, r, done, total_reward)
def reset(self):
"""
reset the game and gives the first observation
"""
self.game.reset()
config = self.game.observe()
obs, info = config['observation']
featurizers.grid_one_hot(self.game, obs)
obs = np.array(obs)
return obs
def _observe(self):
# returns s, r, is_done, info
game_data = self.game.observe()
state, info = game_data['observation']
state = featurizers.grid_one_hot(self.game, state, np=np)
state = state.transpose((2,0,1)) #images go in the [C,H,W] shape for pytorch
return state, game_data['reward'], self.game.is_over(), info
def observe(self):
game_observation = self.game.observe()
# Logic borrowed from:
# https://github.com/facebook/MazeBase/blob/23454fe092ecf35a8aab4da4972f231c6458209b/py/example.py#L192
obs, info = game_observation[OBSERVATION]
featurizers.grid_one_hot(self.game, obs)
obs = np.array(obs)
featurizers.vocabify(self.game, info)
info = np.array(obs)
game_observation[OBSERVATION] = np.concatenate((obs, info),
2).flatten()
is_episode_over = self.game.is_over()
return Observation(id=game_observation[ID],
reward=game_observation[REWARD],
state=game_observation[OBSERVATION],
is_episode_over=is_episode_over)
return actions[ind]
frame = 0
game.display()
sleep(.1)
system('clear')
while True:
print("r: {}\ttr: {} \tguess: {}".format(game.reward(),
game.reward_so_far(),
game.approx_best_reward()))
config = game.observe()
pp.pprint(config['observation'][1])
# Uncomment this to featurize into one-hot vectors
obs, info = config['observation']
featurizers.grid_one_hot(game, obs)
obs = np.array(obs)
featurizers.vocabify(game, info)
info = np.array(obs)
config['observation'] = obs, info
game.display()
id = game.current_agent()
actions = game.all_possible_actions()
action = action_func(actions)
game.act(action)
sleep(.1)
system('clear')
print("\n")
frame += 1