Hold-Out Validation Dataset & Test Dataset

[GloryyrolG]

Thank authors for all the development and release of this lib. Btw, may I know do you guys use hold-out validation dataset which should be distinguished from test dataset. For example, training and testing the model on CIFAR-10 dataset, for choice and decision of hype-params, the original training dataset is divided into trainset and valset usually with the ratio 9:1. But unfortunately, I find the repo may not implement this function, just using the valset identical to testset. Thanks.

[andrewilyas]

Thanks for the comment! We will likely eventually get to this but we are quite resource-constrained and it's not on our immediate roadmap. We would welcome a pull request, if you can find an elegant way to integrate it with the code!