Turicreate: Display falsely classified images

[johnyquest7]

Is there a way to display the images that were wrongly classified and they were classified right?
If turicreate does not have a function for this, creating one will help us understand the data and the model better.

[TobyRoseman]

@johnyquest7 - Here is how I would do that:

test_data['predicted_label'] = model.predict(test_data)
wrong_predictions = test_data[test_data['label_column_name'] != test_data['predicted_label']]
wrong_predictions[['image_column_name', 'label_column_name', 'predicted_label']].explore()

If you have ideas about how to make this easier, please let us know.

[johnyquest7]

Thanks @TobyRoseman. This will help me.
Fastai uses a bunch of code to display "most correctly classified" , "Most incorrectly classified". There was also code for displaying the "most ambiguous results" in which the predictions were near 0.5. I think having an inbuilt function for this in the future versions will help data scientists understand their data and the model in a better way. @znation please consider this as a feature request for future versions.

[znation]

@nickjong Assigning to you as I think this is more of a toolkit API issue. We could include something like @TobyRoseman's snippet above as a method in the toolkit; the existing explore should suffice I think.