77b1d527138ea2385284ed8eae8322c589065d3f,texar/agents/seq_pg_agent.py,SeqPGAgent,_train_policy,#SeqPGAgent#Any#,108

Before Change

        
        discount_factor = self._hparams.discount_factor
        qvalues = list(self._rewards)
        for i in range(len(qvalues) - 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 = np.expand_dims(qvalues, -1)
        max_seq_length = self._samples_py.shape[1]
        if max_seq_length > 1:
            prefix = np.zeros(
                [qvalues.shape[0], max_seq_length-1], dtype=qvalues.dtype)
            qvalues = np.concatenate([prefix, qvalues])
            for i in range(max_seq_length - 2, -1, -1):
                qvalues[:, i] += discount_factor * qvalues[:, i + 1]

        q_mean = np.mean(qvalues)

Instances