7856d22b09561e33522bdc0bd00218ae75b84bd7,softlearning/algorithms/sac.py,SAC,_init_actor_update,#SAC#,282

Before Change

        elif self._action_prior == "uniform":
            policy_prior_log_probs = 0.0

        log_target1 = self._qf1.output_for(
            self._observations_ph, actions, reuse=True)  // N
        log_target2 = self._qf2.output_for(
            self._observations_ph, actions, reuse=True)  // N
        min_log_target = tf.minimum(log_target1, log_target2)

        if self._reparameterize:
            policy_kl_loss = tf.reduce_mean(
                alpha * log_pi - min_log_target - policy_prior_log_probs)

After Change

        q_log_targets = tuple(
            q_function.output_for(
                self._observations_ph, actions, reuse=True)  // N
            for q_function in self._q_functions)
        min_q_log_target = tf.reduce_min(q_log_targets, axis=0)

        if self._reparameterize:

Instances