Cannot reproduce the results of the paper

[zaccharieramzi]

Hi @aravindr93 ,

Thank you so much for providing this repo and the code. I was able to run it in relatively little amount of time and get some results.

However, even when raising the number of tasks to 200,000 and using regularization of 2.0 in the evaluation, I get the following accuracy for validation: 98.204%, which is far from the reported 99.5% on the 5 way 1 shot case of Omniglot.

Do you know if there is any explanation to this? Do you have the weights of models trained to get 99.5% accuracy? I would be very interested in those.