Ok we will give it a try, but out of curiosity, whats the point on doing a hyperparam search on the value of the loss on the last epoch of the experiment vs using the value of the min_loss on the full experiment?