Inspired by the original function most_confused() in ClassificationInterpretation class, I added a new feature to see if there is any class confused strongly with the other one mutually.
For example, this is one of the results returned by most_confused_mutual() as shown below:
(‘Birman’, ‘Ragdoll’, 3, 7, 10)
which means that the model incorrectly predict ‘Birman’ as ‘Ragdoll’ for 3 times, and 7 times for the opposite case, 10 times in total.
This interpretation can show us that some classes in the dataset may share some common characteristics even can’t be easily differentiated by human.