fastai Interpretation

Classification Interpretation

interp = ClassificationInterpretation.from_learner(learn)
interp.plot_confusion_matrix()
interp.plot_top_losses(5, nrows=1)

• Ordered by loss
• If predicted correctly but still shown, then low confidence

Cleaner

cleaner = ImageClassifierCleaner(learn)
cleaner
 
for idx in cleaner.delete(): cleaner.fns[idx].unlink()
for idx,cat in cleaner.change(): shutil.move(str(cleaner.fns[idx]), path/cat)

Get All Classes and Their Probabilities

def classify_image(img):
	pred,idx,probs = learn.predict(img)
 
return dict(zip(categories, map(float,probs)))
 
classify_image(im)