Turicreate Text Classifier predicts lower accuracy than CreateML Text Classifier

[LeeCenY]

The same training data, turicreate 80% prediction accuracy is lower than createml 95%, but the creation of objects occupy memory, turicreate is lower than createml, memory very good, the Turicreate .mlmodel file 1.5MB created object memory occupancy exceeds 6MB, the Createml .mlmodel file 100KB created object memory occupancy exceeds 6MB. Turicreate How should I improve the prediction accuracy?

https://apple.github.io/turicreate/docs/userguide/text_classifier/ https://developer.apple.com/documentation/createml/creating_a_text_classifier_model

[TobyRoseman]

To improve the performance of Turi Create's text classifier, there are few things you could try:

1 - Try increasing the word_count_threshold parameter of text_classifier.create. This will also decrease the size of the .mlmodel file.

2 - Trying increasing he max_iterations iteration parameter.

3 - The text classifier wraps logistic regression. You could preform the preprocessing yourself then try using different parameters when creating the logistic regression model.