def stats(policy, batch_tensors):
values_batched = _make_time_major(policy,
policy.value_function,
drop_last=policy.config["vtrace"])
stats_dict = {
"cur_lr":
tf.cast(policy.cur_lr, tf.float64),
"policy_loss":
policy.loss.pi_loss,
"entropy":
policy.loss.entropy,
"var_gnorm":
tf.global_norm(policy.var_list),
"vf_loss":
policy.loss.vf_loss,
"vf_explained_var":
explained_variance(tf.reshape(policy.loss.value_targets, [-1]),
tf.reshape(values_batched, [-1])),
}
if policy.config["vtrace"]:
is_stat_mean, is_stat_var = tf.nn.moments(policy.loss.is_ratio, [0, 1])
stats_dict.update({"mean_IS": is_stat_mean})
stats_dict.update({"var_IS": is_stat_var})
if policy.config["use_kl_loss"]:
stats_dict.update({"kl": policy.loss.mean_kl})
stats_dict.update({"KL_Coeff": policy.kl_coeff})
return stats_dict
def make_time_major(*args, **kw):
return _make_time_major(policy, *args, **kw)
def make_time_major(*args, **kw):
return _make_time_major(policy, train_batch.get("seq_lens"), *args,
**kw)
