def obtain_samples(self, itr):
cur_params = self.algo.policy.get_param_values()
paths = parallel_sampler.sample_paths(
policy_params=cur_params,
max_samples=self.algo.batch_size,
max_path_length=self.algo.max_path_length,
scope=self.algo.scope,
)
if self.algo.whole_paths:
return paths
else:
paths_truncated = parallel_sampler.truncate_paths(
paths, self.algo.batch_size)
return paths_truncated
def obtain_is_samples(self, itr):
paths = []
for hist_policy_distribution, hist_paths in self.get_history_list(
self.n_backtrack):
h_paths = self.sample_isweighted_paths(
policy=self.algo.policy,
hist_policy_distribution=hist_policy_distribution,
max_samples=self.algo.batch_size,
max_path_length=self.algo.max_path_length,
paths=hist_paths,
hist_variance_penalty=self.hist_variance_penalty,
max_is_ratio=self.max_is_ratio,
ess_threshold=self.ess_threshold,
)
paths.extend(h_paths)
if len(paths) > self.algo.batch_size:
paths = random.sample(paths, self.algo.batch_size)
if self.algo.whole_paths:
return paths
else:
paths_truncated = parallel_sampler.truncate_paths(
paths, self.algo.batch_size)
return paths_truncated