only reproduce 42.94 in miniimagenet about maml-pytorch HOT 3 CLOSED

I guess probably not.
the stop_gradient is for using 1-order derivative approximation. I did not includes it since I only implement the 2nd order version, which should performance better.

from maml-pytorch.

Thanks for your reply!
I achieved 0.4644 with more epochs.

A few more question, I noticed grad = torch.autograd.grad(loss, fast_weights) can be used to calculate the 1nd order gradient and then use fast_weights = list(map(lambda p: p[1] - self.update_lr * p[0], zip(grad, fast_weights))) to calculate the 2nd gradient.

if we only use grad = torch.autograd.grad(loss, fast_weights, create_graph=True), is it can be used to calculate 2nd gradient?

from maml-pytorch.

maybe you can add me wechat: dragen1860 . I dnt get your point clearly.

from maml-pytorch.