def cql_stats(policy: Policy,
train_batch: SampleBatch) -> Dict[str, TensorType]:
sac_dict = stats(policy, train_batch)
sac_dict["cql_loss"] = torch.mean(torch.stack(policy.cql_loss))
if policy.config["lagrangian"]:
sac_dict["log_alpha_prime_value"] = policy.log_alpha_prime_value
sac_dict["alpha_prime_value"] = policy.alpha_prime_value
sac_dict["alpha_prime_loss"] = policy.alpha_prime_loss
return sac_dict
def cql_stats(policy: Policy,
train_batch: SampleBatch) -> Dict[str, TensorType]:
# Get SAC loss stats.
stats_dict = stats(policy, train_batch)
# Add CQL loss stats to the dict.
stats_dict["cql_loss"] = torch.mean(
torch.stack(*policy.get_tower_stats("cql_loss")))
if policy.config["lagrangian"]:
stats_dict["log_alpha_prime_value"] = torch.mean(
torch.stack(policy.get_tower_stats("log_alpha_prime_value")))
stats_dict["alpha_prime_value"] = torch.mean(
torch.stack(policy.get_tower_stats("alpha_prime_value")))
stats_dict["alpha_prime_loss"] = torch.mean(
torch.stack(policy.get_tower_stats("alpha_prime_loss")))
return stats_dict