9caa24c58689c1d6f3d982f623ceab8f78e7362d,softlearning/algorithms/diayn.py,DIAYN,_init_critic_update,#DIAYN#,164

Before Change

        Q-function update rule.
        

        self._qf_t = self._qf.get_output_for(
            self._obs_pl, self._action_pl, reuse=True)  // N

        (obs, z_one_hot) = self._split_obs()
        if self._include_actions:

After Change

        if self._include_actions:
            logits = self._discriminator([obs, self._action_pl])
        else:
            logits = self._discriminator([obs])
        reward_pl = -1 * tf.nn.softmax_cross_entropy_with_logits(labels=z_one_hot,
                                                                 logits=logits)
        reward_pl = tf.check_numerics(reward_pl, "Check numerics (1): reward_pl")
        p_z = tf.reduce_sum(self._p_z_pl * z_one_hot, axis=1)

Instances