def test_continuous_mlp_policy(self):
continuous_mlp_policy = ContinuousMLPPolicy(env_spec=self.env,
hidden_sizes=(1, ))
self.sess.run(tf.compat.v1.global_variables_initializer())
obs = self.env.observation_space.high
assert continuous_mlp_policy.get_action(obs)
def test_get_action(self, obs_dim, action_dim):
"""Test get_action method"""
env = TfEnv(DummyBoxEnv(obs_dim=obs_dim, action_dim=action_dim))
with mock.patch(('metarl.tf.policies.'
'continuous_mlp_policy.MLPModel'),
new=SimpleMLPModel):
policy = ContinuousMLPPolicy(env_spec=env.spec)
env.reset()
obs, _, _, _ = env.step(1)
action, _ = policy.get_action(obs.flatten())
expected_action = np.full(action_dim, 0.5)
assert env.action_space.contains(action)
assert np.array_equal(action, expected_action)
actions, _ = policy.get_actions(
[obs.flatten(), obs.flatten(),
obs.flatten()])
for action in actions:
assert env.action_space.contains(action)
assert np.array_equal(action, expected_action)