forgetting metric error ?

[rolsheng]

your code is :

if task_id > 0:
  forgetting = np.mean((np.max(acc_matrix, axis=1) -
                        acc_matrix[:, task_id])[:task_id])
 backward = np.mean((acc_matrix[:, task_id] - diagonal)[:task_id])

Maybe, I think the correct code for forgetting metric, which is lower the better.
forgetting = np.mean((diagonal - acc_matrix[:, task_id])[:task_id]))

Looking forward for your reply.