Multiclass classification with mixed effects

fabsig / GPBoost

Combining tree-boosting with Gaussian process and mixed effects models

Other

527 stars 42 forks source link

Multiclass classification with mixed effects #67

Open ML-Nielsen opened 2 years ago

ML-Nielsen commented 2 years ago

Hello,

Thanks for a great library!

When I try to train a GPModel with with mixed effects (repeated measures/panel data) and pass params = {'objective' : 'multiclass', 'num_class' : 3}, I get the following error message:

GPBoostError: The GPBoost algorithm can currently not be used for objective=multiclass. If this is desired, contact the developer or open a GitHub issue.

Is multiclass classification not yet implemented, or am I doing it wrong?

Thanks a lot!

fabsig commented 2 years ago

The GPBoost / LaGaBoost algorithm is currently not yet implemented for categorical data with more than two categories. You might consider a "one-against-all" approach where you create K - 1 binary variables being one if the label is a certain category and zero otherwise (K = number of categories) and you train K - 1 models separately.

I will keep this issue open and add an enhancement label. However, I will probably not have time to work on this in the near future. Contributions are welcome.