99153ad034108b7d0e759fc246cb3d05593ee897,texar/agents/pg_agent.py,PGAgent,_train_policy,#PGAgent#Any#,136

Before Change

        Args:
            TODO
        
        discount_factor = self._hparams.discount_factor
        qvalues = list(self._rewards)
        max_seq_length = len(qvalues)
        if max_seq_length >= 2:
            for i in range(max_seq_length - 2, -1, -1):
                qvalues[i] += discount_factor * qvalues[i + 1]

        q_mean = np.mean(qvalues)
        q_std = np.std(qvalues)
        qvalues = [(q - q_mean) / q_std for q in qvalues]

After Change

        qvalues = discount_reward(
            [self._rewards], discount=self._hparams.discount_factor,
            normalize=self._hparams.normalize_reward)
        qvalues = qvalues[0, :]

        fetches = dict(loss=self._train_op)
        feed_dict_ = {
            self._observ_inputs: self._observs,

Instances