def UnclippedObjectiveMean(dist_inputs, values,
returns, actions, old_log_probs):
"""Unclipped objective Mean from the PPO algorithm."""
advantages = returns - values
probs_ratio = rl_layers.ProbsRatio(
dist_inputs, actions, old_log_probs,
log_prob_fun=self._policy_dist.log_prob)
unclipped_objective = rl_layers.UnclippedObjective(
probs_ratio, advantages)
return jnp.mean(unclipped_objective)
def f(dist_inputs, values, returns, actions, old_log_probs):
"""Unclipped objective Mean from the PPO algorithm."""
advantages = returns - values
probs_ratio = rl_layers.ProbsRatio(
dist_inputs, actions, old_log_probs,
log_prob_fun=self._policy_dist.log_prob)
# advantages are of the shape [128,1,1]
# and probs_ratio are of the shape [128,1]
advantages = advantages.squeeze(axis=2)
unclipped_objective = rl_layers.UnclippedObjective(
probs_ratio, advantages)
return jnp.mean(unclipped_objective)