I’m getting 1.0 accuracy on MLP training with weights resulting in NaNs. Either I’m using the wrong input file or Kaggle dataset changed (again?). I have *.jpg files named dog..jpg and cat..jpg, but I don’t see any code that extracts and encodes dog/cat information for training, resulting (I think) in all labels being 0.
Edit: sorry, I was using absolute path and had dogs-vs-cats subdir on it, causing everything to be labeled as cats! :facepalm: