Some concerns about training

[zzpustc]

Hi Seongmin, great work!

I have run your code, it performs well. But diving into the training details, I found that each episode you just sample a task while in other metric-based methods/codes (like prototypical network) they usually sampled a certain number of tasks in one epoch and repeat several epochs. I noticed that you set episode as 50000, is this setting equivalent to multi-epochs?

Hope to receive your reply.