musicmood
musicmood copied to clipboard
overfitting
Sebastian, isn't your model overfitting as the recall falls sharply when you going from train to test data.
Yup, I agree! I got more stable results with regularized logistic regression but couldn't find a way to get a lower train/test discrepancy with any of the naive Bayes models. I started to collect more data but never got around re-training the models, something that's still on my to-do list for some point in future!