Hi. Thanks for this paper. It's quite inspiring. I am trying to test different models on the DeepSea data set you used in this paper. However, due to the dominant 0 in the y labels, my model does not seem to able to predict any 1. Through out the training, the model accuracy stays around 0.98. At the end, when I did prediction on the test dataset with the trained model, I would get pretty much all 0 predictions.
Have you ever encountered such issue when training your model. If so, how did you resolve it?