output_nonlinearity=None,
)
policy = GaussianConvFeaturePolicy(
"conv_feature_policy",
env_spec=env_spec,
feature_network=policy_cnn,
hidden_sizes=(128, 64),
clip_action=False,
)
baseline = NNBaseline(
env_spec=env_spec,
feature_network=baseline_cnn,
hidden_sizes=(128, 64),
hidden_nonlinearity=tf.nn.relu,
init_lr=1e-4,
n_itr=5,
train_feature_network=True,
)
batch_size = 9600
algo = TRPO(env=env,
policy=policy,
baseline=baseline,
batch_size=batch_size,
whole_paths=True,
max_path_length=1000,
n_itr=4000,
step_size=0.01,
subsample_factor=0.2,
hidden_nonlinearity=tf.nn.relu,
output_nonlinearity=None,
)
policy = GaussianConvFeaturePolicy(
"conv_feature_policy",
env_spec=env_spec,
feature_network=cnn,
hidden_sizes=(128,64),
output_nonlinearity=tf.nn.tanh,
)
baseline = NNBaseline(
env_spec=env_spec,
feature_network=cnn,
hidden_sizes=(128,64),
hidden_nonlinearity=tf.nn.relu,
init_lr=0.001,
n_itr=5,
)
batch_size = 2400
algo = TRPO(
env=env,
policy=policy,
baseline=baseline,
batch_size=batch_size,
whole_paths=True,
max_path_length=200,
n_itr=2000,
step_size=0.01,
subsample_factor=1.0,
