def train(self):
return Agent.train(self)
def train(self):
# since the algorithm works with experience replay buffer (non-episodic),
# we cant use the policy optimization train method. we need Agent.train
# note that since in Agent.train there is no apply_gradients, we need to do it in learn from batch
return Agent.train(self)
def train(self):
self.ap.algorithm.num_consecutive_training_steps = \
int(self.current_episode_steps_counter * self.ap.algorithm.td3_training_ratio)
return Agent.train(self)
def train(self):
self.ap.algorithm.num_consecutive_training_steps = self.current_episode_steps_counter
return Agent.train(self)