Query on number of inner loop iterations in MAML

[klnavaneet]

Hi, thanks for the great repo!

In MAML 2nd order training, what values do you use for the number of inner loop iterations for mini-Imagenet dataset? The paper uses 1 and 5 iterations for Omniglot and mini-Imagenet respectively in the meta-training stage. Here, in the default arguments (https://github.com/oscarknagg/few-shot/blob/master/experiments/maml.py), inner-train-steps is set to 1. I was curious to know if 1 iteration during training would still result in good performance for mini-Imagenet.