Right now when training the model we use the built-in classification_report and confusion_matrix scikit-learn to judge model accuracy. We should add a "top five" accuracy measure to this, as when predicting we return the top 5 predicted classes by default. Top Five accuracy would show the percentage of test predictions which have the correctly predicted label in the top five returned classes.
Right now when training the model we use the built-in
classification_reportandconfusion_matrixscikit-learn to judge model accuracy. We should add a "top five" accuracy measure to this, as when predicting we return the top 5 predicted classes by default. Top Five accuracy would show the percentage of test predictions which have the correctly predicted label in the top five returned classes.