def _init_model(self):
"""init model from parameters"""
self.env, env_continuous, num_states, num_actions = get_env_info(
self.env_id)
tf.keras.backend.set_floatx('float64')
# seeding
np.random.seed(self.seed)
tf.random.set_seed(self.seed)
self.env.seed(self.seed)
if env_continuous:
self.policy_net = Policy(num_states, num_actions) # current policy
else:
self.policy_net = DiscretePolicy(num_states, num_actions)
self.running_state = ZFilter((num_states, ), clip=5)
if self.model_path:
print("Loading Saved Model {}_reinforce_tf2.p".format(self.env_id))
self.running_state = pickle.load(
open(
'{}/{}_reinforce_tf2.p'.format(self.model_path,
self.env_id), "rb"))
self.policy_net.load_weights("{}/{}_reinforce_tf2".format(
self.model_path, self.env_id))
self.collector = MemoryCollector(self.env,
self.policy_net,
render=self.render,
running_state=self.running_state,
num_process=self.num_process)
self.optimizer_p = optim.Adam(lr=self.lr_p, clipnorm=20)
def _init_model(self):
"""init model from parameters"""
self.env, env_continuous, num_states, num_actions = get_env_info(
self.env_id
)
tf.keras.backend.set_floatx("float64")
# seeding
np.random.seed(self.seed)
tf.random.set_seed(self.seed)
self.env.seed(self.seed)
if env_continuous:
self.policy_net = Policy(num_states, num_actions)
else:
self.policy_net = DiscretePolicy(num_states, num_actions)
self.value_net = Value(num_states, l2_reg=1e-3)
self.running_state = ZFilter((num_states,), clip=5)
if self.model_path:
print("Loading Saved Model {}_trpo_tf2.p".format(self.env_id))
self.running_state = pickle.load(
open(
"{}/{}_trpo_tf2.p".format(self.model_path, self.env_id),
"rb",
)
)
self.policy_net.load_weights(
"{}/{}_trpo_tf2_p".format(self.model_path, self.env_id)
)
self.value_net.load_weights(
"{}/{}_trpo_tf2_v".format(self.model_path, self.env_id)
)
self.collector = MemoryCollector(
self.env,
self.policy_net,
render=self.render,
running_state=self.running_state,
num_process=self.num_process,
)
self.optimizer_v = optim.Adam(lr=self.lr_v)