plot_top_losses in vison.learn can show pictures for us to see if what the model predicted was reasonable or not.
Is there any method like
plot_top_losses for text classification can show how well the model have done with source sentences presented?
To do some error analysis after some text-based multi-label classification, I incorporated the true labels, predictions and metric in a single dataframe, and sorted in ascending order of metric (in my case f-score). This way I can analyse what’s going on, starting with the worst predictions. See the end of my kernel.
Thanks for sharing! Learned a lot.
But what I want is presenting the interpretation of validation dataset just like
plot_top_losses does. So I’ve wrote a similar method to do it called show_top_losses(). I’ve send my PR and waiting for reply.
Nice! I would definitely use it!