Failing to prune XGBoost tree models about jpmml-converter HOT 3 OPEN

Looks like you've trained an XGBoost model, which contains no-op nodes.

The JPMML-XGBoost automatically tries to eliminate those nodes (because they are provably unreachable under any and all scenarios) by applying a special tree model pruning algorithm implementes as org.jpmml.converter.visitors.TreeModelPruner.

I was sure that the tree pruning code will always succeed. However, you've managed to train an XGBoost model that contains such an unusual "internal structure" that the tree pruning code still fails.

Can you share your model file so that I could take a look at this unusual "internal structure" myself? Or if it's trained on proprietary data, can you reproduce the pruning error using some publicly available toy dataset?

from jpmml-converter.

As a workaround, you should dsable tree pruning by specifying the prune = False conversion option:

pipeline = PMMLPipeline([
  ("xgb", XGBClassifier())
])
pipeline.fit(X, y)
# THIS - specify conversion options right after fitting the pipeline
pipeline.configure(prune = False)

sklearn2pmml(pipeline, "XGBoost.pmml")

from jpmml-converter.

Thanks a lot! It works now

from jpmml-converter.