def rollout(self):
obs, actions, rewards, action_dists_mu, action_dists_logstd = [], [], [], [], []
self.old_observation = None
plain_obs = self.env.reset(difficulty = 2, seed = int(time.time()))
processed_observation, self.old_observation = go(plain_obs, self.old_observation, step=1)
ob = filter(processed_observation)
for i in xrange(self.args.max_pathlength - 1):
obs.append(ob)
action, action_dist_mu, action_dist_logstd = self.act(ob)
action = np.clip(action,a_max=1.0,a_min=0.0)
actions.append(action)
action_dists_mu.append(action_dist_mu)
action_dists_logstd.append(action_dist_logstd)
res = self.env.step(action)
processed_observation, self.old_observation = go(res[0], self.old_observation, step=1)
ob = filter(processed_observation)
rewards.append((res[1]))
if res[2] or i == self.args.max_pathlength - 2:
path = {"obs": np.concatenate(np.expand_dims(obs, 0)),
"action_dists_mu": np.concatenate(action_dists_mu),
"action_dists_logstd": np.concatenate(action_dists_logstd),
"rewards": np.array(rewards),
"actions": np.array(actions)}
return path
break
def obg(self, plain_obs):
# observation generator
# derivatives of observations extracted here.
processed_observation, self.old_observation = go(plain_obs,
self.old_observation,
step=self.stepcount)
return np.array(processed_observation)
def obg(plain_obs):
nonlocal old_observation, stepno, ob_log
# log csv observation into string
ob_log += ','.join([str(i) for i in plain_obs]) + '\n'
processed_observation, old_observation = go(plain_obs,
old_observation,
step=stepno)
return np.array(processed_observation)
def obg(plain_obs):
nonlocal old_observation, stepno
processed_observation, old_observation = go(plain_obs,
old_observation,
step=stepno)
return np.array(processed_observation)