I’m posting here since as I submitted a PR I’ve been required to talk about it here on
fastai dev forum section.
I submitted a pull request with my version of
plot_top_losses() which works upon multilabeled datasets. I named it
It basically behaves like
plot_top_losses(), but since multiple class labels tended to overlap in a grid plot, I preferred plotting a single column with big pics (by default).
Regardless, it plots the first N misclassified samples (sorted by loss magnitude), with predicted class, actual class, loss, and probability.
I hope you’ll accept it.
In time, I’ll try and code the corresponding versions of
most_confused() and the remaining stuff.