def policy_learn(self):
"""A learning iteration for the policy"""
all_discounted_returns = self.calculate_all_discounted_returns()
if self.hyperparameters["normalise_rewards"]:
all_discounted_returns = normalise_rewards(all_discounted_returns)
for _ in range(self.hyperparameters["learning_iterations_per_round"]):
all_ratio_of_policy_probabilities = self.calculate_all_ratio_of_policy_probabilities()
loss = self.calculate_loss([all_ratio_of_policy_probabilities], all_discounted_returns)
self.take_policy_new_optimisation_step(loss)
def policy_learn(self):
all_discounted_returns = self.calculate_all_discounted_returns()
if self.config.hyperparameters["normalise_rewards"]:
all_discounted_returns = normalise_rewards(all_discounted_returns)
# number of epochs
for _ in range(
self.config.hyperparameters["learning_iterations_per_round"]):
all_ratio_of_policy_probabilities = self.calculate_all_ratio_of_policy_probabilities(
)
loss = self.calculate_loss([all_ratio_of_policy_probabilities],
all_discounted_returns)
self.take_policy_new_optimisation_step(loss)
if self.config.save_model:
torch.save(self.policy_new.state_dict(), self.config.model_path)
self.init_batch_lists()