About validation in pytorch about hashnet

[caozhangjie]

1.You can use the model at 6000-7000 iterations.

2.The parameters are tuned in our experiments with cross-validation and I only set them as the optimal parameters for reproducing.

[soon-will]

@caozhangjie
Thanks for your reply.

About the first question, if there is no val set, then how do you ensure the model is the best model? The parameters, like learning rate, can be tuned with cross-validation, should we use val set to choose the better model , like training a classification model?

[caozhangjie]

Yes, you can divide the whole training set into 10 pieces and use 10-fold cross-validation as in classification.

[soon-will]

Thank you for your reply.