Analyze model prediction errors

This recipe shows how to analyze model prediction errors in the FiftyOne App.

1. Make predictions

Through mistakenness: 1. Compute mistakenness 2. Filter in the app by mistakeness and select most frequent errors 3. Open another app instance and review training data 4. Tag labels as fix, del, etc.

Through classification report: 1. Compute classification report 2. Review the report and select most frequent errors 3. Open another app instance and review test errors 4. Open another app instance and review training data 5. Tag labels as fix, del, etc.