search

kimhc6028 / pathnet-pytorch

PyTorch implementation of PathNet: Evolution Channels Gradient Descent in Super Neural Networks
BSD 3-Clause "New" or "Revised" License
80 stars 18 forks source link

[question] Training/validation/test splits #3

Closed oarriaga closed 7 years ago

oarriaga commented 7 years ago

Hello again! I was looking at your mnist_dataset.py file and I was wondering if you create any validation/test splits for this task. I am asking since I am trying to figure out if the fitness/loss value is calculated using the training data itself or if you calculate it using a validation/test split. And in case you create this splits, are you adding the salt and pepper noise also to them? Thank you!

Maybe @jaesik817 can comment on this issue as well.

oarriaga commented 7 years ago

I just found the answer in the paper: The fitness of that pathway is the proportion of correct examples classified on the training set during this training period. Thank you