def test_baseline(self):
"""Test the baseline initialization."""
box_env = GarageEnv(DummyBoxEnv())
deterministic_mlp_baseline = ContinuousMLPBaseline(env_spec=box_env)
gaussian_mlp_baseline = GaussianMLPBaseline(env_spec=box_env)
self.sess.run(tf.compat.v1.global_variables_initializer())
deterministic_mlp_baseline.get_param_values()
gaussian_mlp_baseline.get_param_values()
box_env.close()
def test_param_values(self):
box_env_spec = GarageEnv(DummyBoxEnv(obs_dim=(2, ))).spec
cmb = ContinuousMLPBaseline(env_spec=box_env_spec)
new_cmb = ContinuousMLPBaseline(env_spec=box_env_spec,
name='ContinuousMLPBaseline2')
# Manual change the parameter of ContinuousMLPBaseline
with tf.compat.v1.variable_scope('ContinuousMLPBaseline', reuse=True):
bias = tf.compat.v1.get_variable('mlp/hidden_0/bias')
bias.load(tf.ones_like(bias).eval())
old_param_values = cmb.get_param_values()
new_param_values = new_cmb.get_param_values()
assert not np.array_equal(old_param_values, new_param_values)
new_cmb.set_param_values(old_param_values)
new_param_values = new_cmb.get_param_values()
assert np.array_equal(old_param_values, new_param_values)
def test_param_values(self, obs_dim):
box_env = GarageEnv(DummyBoxEnv(obs_dim=obs_dim))
with mock.patch(('garage.tf.baselines.'
'continuous_mlp_baseline.'
'ContinuousMLPRegressor'),
new=SimpleMLPRegressor):
cmb = ContinuousMLPBaseline(env_spec=box_env.spec)
new_cmb = ContinuousMLPBaseline(env_spec=box_env.spec,
name='ContinuousMLPBaseline2')
# Manual change the parameter of ContinuousMLPBaseline
with tf.compat.v1.variable_scope('ContinuousMLPBaseline2', reuse=True):
return_var = tf.compat.v1.get_variable('SimpleMLPModel/return_var')
return_var.load(1.0)
old_param_values = cmb.get_param_values()
new_param_values = new_cmb.get_param_values()
assert not np.array_equal(old_param_values, new_param_values)
new_cmb.set_param_values(old_param_values)
new_param_values = new_cmb.get_param_values()
assert np.array_equal(old_param_values, new_param_values)
def test_baseline(self):
"""Test the baseline initialization."""
box_env = TfEnv(DummyBoxEnv())
deterministic_mlp_baseline = ContinuousMLPBaseline(env_spec=box_env)
gaussian_mlp_baseline = GaussianMLPBaseline(env_spec=box_env)
discrete_env = TfEnv(Resize(DummyDiscrete2DEnv(), width=64, height=64))
gaussian_conv_baseline = GaussianConvBaseline(
env_spec=discrete_env,
regressor_args=dict(
conv_filters=[32, 32],
conv_filter_sizes=[1, 1],
conv_strides=[1, 1],
conv_pads=['VALID', 'VALID'],
hidden_sizes=(32, 32)))
self.sess.run(tf.global_variables_initializer())
deterministic_mlp_baseline.get_param_values(trainable=True)
gaussian_mlp_baseline.get_param_values(trainable=True)
gaussian_conv_baseline.get_param_values(trainable=True)
box_env.close()