def setUp(self):
self.policy_params = {
'sess': tf.compat.v1.Session(),
'ac_space': Box(low=-1, high=1, shape=(1, ), dtype=np.float32),
'ob_space': Box(low=-2, high=2, shape=(2, ), dtype=np.float32),
'co_space': Box(low=-3, high=3, shape=(2, ), dtype=np.float32),
'layers': None,
'verbose': 2,
}
self.policy_params.update(SAC_PARAMS.copy())
self.policy_params.update(GOAL_CONDITIONED_PARAMS.copy())
def test_setup_model_goal_conditioned(self):
# Create the algorithm object.
policy_params = self.init_parameters.copy()
policy_params['policy'] = GoalConditionedPolicy
policy_params['_init_setup_model'] = True
alg = OffPolicyRLAlgorithm(**policy_params)
# check the policy_kwargs term
policy_kwargs = GOAL_CONDITIONED_PARAMS.copy()
policy_kwargs.update(TD3_PARAMS)
policy_kwargs['verbose'] = self.init_parameters['verbose']
policy_kwargs['env_name'] = self.init_parameters['env']
policy_kwargs['num_envs'] = self.init_parameters['num_envs']
self.assertDictEqual(alg.policy_kwargs, policy_kwargs)
with alg.graph.as_default():
expected_vars = sorted([var.name for var in get_trainable_vars()])
# Check that all trainable variables have been created in the
# TensorFlow graph.
self.assertListEqual(
expected_vars,
['level_0/model/pi/fc0/bias:0',
'level_0/model/pi/fc0/kernel:0',
'level_0/model/pi/fc1/bias:0',
'level_0/model/pi/fc1/kernel:0',
'level_0/model/pi/output/bias:0',
'level_0/model/pi/output/kernel:0',
'level_0/model/qf_0/fc0/bias:0',
'level_0/model/qf_0/fc0/kernel:0',
'level_0/model/qf_0/fc1/bias:0',
'level_0/model/qf_0/fc1/kernel:0',
'level_0/model/qf_0/qf_output/bias:0',
'level_0/model/qf_0/qf_output/kernel:0',
'level_0/model/qf_1/fc0/bias:0',
'level_0/model/qf_1/fc0/kernel:0',
'level_0/model/qf_1/fc1/bias:0',
'level_0/model/qf_1/fc1/kernel:0',
'level_0/model/qf_1/qf_output/bias:0',
'level_0/model/qf_1/qf_output/kernel:0',
'level_0/target/pi/fc0/bias:0',
'level_0/target/pi/fc0/kernel:0',
'level_0/target/pi/fc1/bias:0',
'level_0/target/pi/fc1/kernel:0',
'level_0/target/pi/output/bias:0',
'level_0/target/pi/output/kernel:0',
'level_0/target/qf_0/fc0/bias:0',
'level_0/target/qf_0/fc0/kernel:0',
'level_0/target/qf_0/fc1/bias:0',
'level_0/target/qf_0/fc1/kernel:0',
'level_0/target/qf_0/qf_output/bias:0',
'level_0/target/qf_0/qf_output/kernel:0',
'level_0/target/qf_1/fc0/bias:0',
'level_0/target/qf_1/fc0/kernel:0',
'level_0/target/qf_1/fc1/bias:0',
'level_0/target/qf_1/fc1/kernel:0',
'level_0/target/qf_1/qf_output/bias:0',
'level_0/target/qf_1/qf_output/kernel:0',
'level_1/model/pi/fc0/bias:0',
'level_1/model/pi/fc0/kernel:0',
'level_1/model/pi/fc1/bias:0',
'level_1/model/pi/fc1/kernel:0',
'level_1/model/pi/output/bias:0',
'level_1/model/pi/output/kernel:0',
'level_1/model/qf_0/fc0/bias:0',
'level_1/model/qf_0/fc0/kernel:0',
'level_1/model/qf_0/fc1/bias:0',
'level_1/model/qf_0/fc1/kernel:0',
'level_1/model/qf_0/qf_output/bias:0',
'level_1/model/qf_0/qf_output/kernel:0',
'level_1/model/qf_1/fc0/bias:0',
'level_1/model/qf_1/fc0/kernel:0',
'level_1/model/qf_1/fc1/bias:0',
'level_1/model/qf_1/fc1/kernel:0',
'level_1/model/qf_1/qf_output/bias:0',
'level_1/model/qf_1/qf_output/kernel:0',
'level_1/target/pi/fc0/bias:0',
'level_1/target/pi/fc0/kernel:0',
'level_1/target/pi/fc1/bias:0',
'level_1/target/pi/fc1/kernel:0',
'level_1/target/pi/output/bias:0',
'level_1/target/pi/output/kernel:0',
'level_1/target/qf_0/fc0/bias:0',
'level_1/target/qf_0/fc0/kernel:0',
'level_1/target/qf_0/fc1/bias:0',
'level_1/target/qf_0/fc1/kernel:0',
'level_1/target/qf_0/qf_output/bias:0',
'level_1/target/qf_0/qf_output/kernel:0',
'level_1/target/qf_1/fc0/bias:0',
'level_1/target/qf_1/fc0/kernel:0',
'level_1/target/qf_1/fc1/bias:0',
'level_1/target/qf_1/fc1/kernel:0',
'level_1/target/qf_1/qf_output/bias:0',
'level_1/target/qf_1/qf_output/kernel:0']
)