Is your feature request related to a problem? Please describe.
I would like to plot top losses for predictions made by a trained model. This would be really useful to help me figure out where my trained model is making mistakes. With show_results I can view a model's predictions however loss would also be useful as loss encompasses how confident the model is. With this additional information I could better clean my dataset and work towards improving accuracy.
Describe the solution you'd like
I would like to write a function that accepts samples and predictions as input parameters and outputs an ordered dictionary pairing samples with loss values.
Describe alternatives you've considered
Additional context
It doesn't seem possible to calculate loss when model is in evaluation mode.
Hey @lgvaz, can you please assign me to this one?
Feature implemented here.
Can we close?
Awesome work!! Look forward to working with this feature. Thanks so much :D