Training with a small fraction of the dataset?

I am building a NN for image enhancement currently and I have a matching pair dataset of around 1mil images. I want to start experimenting with learning rates and different feature losses on a smaller subset of images (maybe 50k or 5% of the whole dataset). Is there a way to specify this in the data loader? I thought I remember you could do this but I can’t find the info anywhere. Thanks!

I figured it out in case anyone wants to do the same thing.

It was already here I just didn’t know what to search for! 100 tabs and a half a day later and I found this.

2 Likes