def _build_critic(self):
self.q_func = create_continuous_q_function(
self.observation_shape,
self.action_size,
self.critic_encoder_factory,
n_ensembles=self.n_critics,
q_func_type=self.q_func_type,
bootstrap=self.bootstrap,
share_encoder=self.share_encoder)
def _build_critic(self):
self.q_func = create_continuous_q_function(
self.observation_shape,
self.action_size,
n_ensembles=self.n_critics,
use_batch_norm=self.use_batch_norm,
q_func_type=self.q_func_type,
bootstrap=self.bootstrap,
share_encoder=self.share_encoder,
encoder_params=self.encoder_params)
def test_compute_max_with_n_actions(observation_shape, action_size,
n_ensembles, batch_size, n_quantiles,
n_actions, lam, q_func_type):
q_func = create_continuous_q_function(observation_shape,
action_size,
n_ensembles,
n_quantiles,
q_func_type=q_func_type)
x = torch.rand(batch_size, *observation_shape)
actions = torch.rand(batch_size, n_actions, action_size)
y = compute_max_with_n_actions(x, actions, q_func, lam)
if q_func_type == 'mean':
assert y.shape == (batch_size, 1)
else:
assert y.shape == (batch_size, n_quantiles)
def test_create_continuous_q_function(observation_shape, action_size,
batch_size, n_ensembles, n_quantiles,
embed_size, use_batch_norm,
share_encoder, q_func_type):
q_func = create_continuous_q_function(observation_shape,
action_size,
n_ensembles,
n_quantiles,
embed_size,
use_batch_norm,
q_func_type,
share_encoder=share_encoder)
assert isinstance(q_func, EnsembleContinuousQFunction)
for f in q_func.q_funcs:
if q_func_type == 'mean':
assert isinstance(f, ContinuousQFunction)
elif q_func_type == 'qr':
assert isinstance(f, ContinuousQRQFunction)
elif q_func_type == 'iqn':
assert isinstance(f, ContinuousIQNQFunction)
elif q_func_type == 'fqf':
assert isinstance(f, ContinuousFQFQFunction)
assert f.encoder.use_batch_norm == use_batch_norm
# check share_encoder
encoder = q_func.q_funcs[0].encoder
for q_func in q_func.q_funcs[1:]:
if share_encoder:
assert encoder is q_func.encoder
else:
assert encoder is not q_func.encoder
x = torch.rand((batch_size, ) + observation_shape)
action = torch.rand(batch_size, action_size)
y = q_func(x, action)
assert y.shape == (batch_size, 1)