dragen1860
Incorrect losses_q
losses_q is supposed to be all losses on query set across all tasks. But it is defined as below. That's all the losses on query set for all the update_step for task i, not all the task. losses_q = [0 for _ in range(self.update_step + 1)] Is this an error?
losses_q is supposed to be all losses on query set across all tasks. But it is defined as below. That's all the losses on query set for all the update_step for task i, not all the task. losses_q = [0 for _ in range(self.update_step + 1)] Is this an error?
so, you are suggesting losses_q = [0 for _ in range(self.update_step*task_num + 1)] to accumulate all query loss?
No it is not an error. They create a list the length of the number of inner update step then accumulate the loss on each task at that particular step i.e losses_q[3] will be the sum of loss of each task for the update step 4. He then proceed to take only the final value of the list and to average on that is done with loss_q = losses_q[-1] / task_num on line 134 .
