Should validation set be used to select the best model?

joshuacwnewton / d-SNE-PyTorch

A PyTorch port of the existing MXNet implementation for the 2019 CVPR paper "d-SNE: Domain Adaptation Using Stochastic Neighborhood Embedding."

7 stars 3 forks source link

Should validation set be used to select the best model? #2

Open charliezon opened 4 years ago

charliezon commented 4 years ago

I noticed there is a difference between your implementation and the original MXNET implementation. A validation set is used in your implmentation, while the original implementation didn't use.

Since there are only a few labeled samples in target training set for few-shot SDA problem, it seems the validation set will be unavailable in real scenario. I think the original implementation is more reasonable.