def test_setup_model_feedforward(self):
# Create the algorithm object.
policy_params = self.init_parameters.copy()
policy_params['policy'] = FeedForwardPolicy
policy_params['_init_setup_model'] = True
alg = RLAlgorithm(**policy_params)
# check the policy_kwargs term
policy_kwargs = FEEDFORWARD_PARAMS.copy()
policy_kwargs.update(TD3_PARAMS)
policy_kwargs['verbose'] = self.init_parameters['verbose']
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,
['model/pi/fc0/bias:0',
'model/pi/fc0/kernel:0',
'model/pi/fc1/bias:0',
'model/pi/fc1/kernel:0',
'model/pi/output/bias:0',
'model/pi/output/kernel:0',
'model/qf_0/fc0/bias:0',
'model/qf_0/fc0/kernel:0',
'model/qf_0/fc1/bias:0',
'model/qf_0/fc1/kernel:0',
'model/qf_0/qf_output/bias:0',
'model/qf_0/qf_output/kernel:0',
'model/qf_1/fc0/bias:0',
'model/qf_1/fc0/kernel:0',
'model/qf_1/fc1/bias:0',
'model/qf_1/fc1/kernel:0',
'model/qf_1/qf_output/bias:0',
'model/qf_1/qf_output/kernel:0',
'target/pi/fc0/bias:0',
'target/pi/fc0/kernel:0',
'target/pi/fc1/bias:0',
'target/pi/fc1/kernel:0',
'target/pi/output/bias:0',
'target/pi/output/kernel:0',
'target/qf_0/fc0/bias:0',
'target/qf_0/fc0/kernel:0',
'target/qf_0/fc1/bias:0',
'target/qf_0/fc1/kernel:0',
'target/qf_0/qf_output/bias:0',
'target/qf_0/qf_output/kernel:0',
'target/qf_1/fc0/bias:0',
'target/qf_1/fc0/kernel:0',
'target/qf_1/fc1/bias:0',
'target/qf_1/fc1/kernel:0',
'target/qf_1/qf_output/bias:0',
'target/qf_1/qf_output/kernel:0']
)
def setUp(self):
sess = tf.compat.v1.Session()
self.policy_params = {
'sess': sess,
'ac_space': Box(low=-1, high=1, shape=(1, )),
'ob_space': Box(low=-2, high=2, shape=(2, )),
'co_space': Box(low=-3, high=3, shape=(3, )),
'verbose': 0,
}
self.policy_params.update(FEEDFORWARD_PARAMS.copy())
def setUp(self):
self.policy_params = {
'ac_space': Box(low=-1, high=1, shape=(1,)),
'ob_space': Box(low=-2, high=2, shape=(2,)),
'co_space': Box(low=-3, high=3, shape=(3,)),
'scope': None,
'verbose': 0,
}
self.policy_params.update(TRPO_PARAMS.copy())
self.policy_params.update(FEEDFORWARD_PARAMS.copy())
self.policy_params["model_params"]["model_type"] = "fcnet"
def setUp(self):
sess = tf.compat.v1.Session()
self.policy_params = {
'sess': sess,
'ac_space': Box(low=-1, high=1, shape=(1, )),
'ob_space': Box(low=-2, high=2, shape=(2, )),
'co_space': Box(low=-3, high=3, shape=(3, )),
'verbose': 0,
'buffer_size': 200000,
'batch_size': 128,
'actor_lr': 3e-4,
'critic_lr': 3e-4,
'tau': 0.005,
'gamma': 0.99,
'use_huber': False,
}
self.policy_params.update(FEEDFORWARD_PARAMS.copy())