This is a Go package for calculating feature contributions for XGBoost models. The code is ported from the XGBoost C++ code. This package exists to make it possible to calculate contributions with pure Go code rather than having to use cgo or another language.
While the code is ported from the XGBoost C++ code, not all code was ported. If you find results differ between the two implementations, it is possible that code relevant to your model was not ported.
In particular, we have not yet ported:
- Code involving condition and condition_features
- Code involving tree weights
- Some of the code involving the bias
- Code involving categories
It is also possible that XGBoost's code has changed since this code was written. We will be attempting to keep this implementation up to date.
modelFile := "/path/to/model.json"
predictor, err := xgbshap.NewPredictor(modelFile)
if err != nil {
return err
}
features := []*float32{...your features here...}
contributions, err := predictor.PredictContributions(features)
if err != nil {
return err
}Please report bugs by filing an issue with our GitHub issue tracker at https://github.com/maxmind/xgbshap/issues.
This software is Copyright (c) 2023 - 2024 by MaxMind, Inc.
Much of the code is ported from the XGBoost project, so it is also Copyright (c) 2017-2023 by XGBoost Contributors.
This is free software, licensed under the Apache License, Version 2.0.