Model not training: Model predicting same thing every time

[J_Johnson]

This is a very common problem. I wrote something in the FAQ which addresses it with a code example.

How do I get the accuracy of an unbalanced dataset or segmentation task?

This post is to define a Class Weighted Accuracy function(WCA). As questions related to this get asked often, I thought it might help people to post a tool torchers can use and reference here. Why? Take, for example, a classification dataset of kittens and puppies with a ratio of 0.2:0.8 kittens to puppies. A model trained on this dataset might show an overall accuracy of 80% by just predicting “puppies” every time. A WCA function resolves this problem by checking the accuracy of each class s…

In a nut shell, when you train a supervised task on an unbalanced dataset, your model will optimize for decreasing loss, which can mean only answering the option that is the most common in the dataset.

To address this, we need to reweight those class indices and pass that set of weights into the loss function.

EDIT: I just reviewed the FAQ in the link and it seems that only addresses assessing an unbalanced dataset for model accuracy. I had later intended to include a function within it to calculate the rescaling weights, but it seems I never got around to that. I will update it in the next 24 hours to include that.