c702968d83449f3a6aad3a5cec6cc0b2aafd78c5,mushroom/algorithms/dqn.py,RDQN,fit,#RDQN#Any#Any#,228

Before Change

            q_tilde_next = self._target_approximator.predict_all(
                next_state) * (1. - next_absorbing)
            r_tilde_next = np.zeros(q_tilde_next.shape)
            for i in xrange(r_tilde_next.shape[0]):
                r_tilde_next[i, next_action] = next_reward[i]
            q_next = r_tilde_next + self._gamma * q_tilde_next
            max_q_next = np.max(q_next, axis=1)

            self.approximator.train_on_batch(

After Change

            q_tilde_next = self._target_approximator.predict_all(
                next_state) * (1. - next_absorbing.reshape(-1, 1))
            r_tilde_next = np.zeros(q_tilde_next.shape)
            r_tilde_next[:, next_action.ravel().astype(np.int)] = next_reward
            q_next = r_tilde_next + self._gamma * q_tilde_next
            max_q_next = np.max(q_next, axis=1)

Instances