Xgboost 4j feature interaction

#1

Hi,

I have a trained XGBoost model trained in Spark using XGBoost 4j. Is there a way for me to get the feature interactions based on the result/ model I have? Any idea will be helpful! Thankss!

#2

Do you mean SHAP values for feature interactions? No, SHAP is not yet available in XGBoost4J.

#3

Yeah, I think that’s the one. Is any other relative api I can call within XGBoost4j? I am not very familiar with it. But the data size in general is huge, so it’s unreal to re-run it in some perspective. I might be only able to use the model saved.

#4

Hi,

I figured to export the xgboost 4j model to the python version. Using the method below.

val nativeModelPath = “/tmp/nativeModel”
xgbClassificationModel.nativeBooster.saveModel(nativeModelPath)

import xgboost as xgb
bst = xgb.Booster({‘nthread’: 4})
bst.load_model(nativeModelPath)

I met an error after I loaded the model and tried to save it using “nativeBooster”. The error says “nativeModelPath doesn’t exist”. Should this just be a new location+model name I define? I don’t know why this happens. Please kindly let me know if you have the answer.

Thanks a bunch in advanced!!!

Best,
Jo