Closed xixi2715 closed 6 years ago
As you correctly point out, the JPMML-SkLearn only supports the conversion of Apache Spark ML models and pipelines.
There are no plans to start supporting Apache Spark MLlib. It is my understanding that the Apache Spark team has decided to gradually phase out MLlib functionality, so it would be pointless to spend any resources on it.
I would personally recommend you to re-train your GBDT model using Apache Spark ML. Alternatively, you might want to open discussion with Apache Spark team, and see if they are willing to apply the PMMLExportable
trait to the GBDT model type. This trait doesn't seem to cover decision tree model types at the moment, so prospect is not so good.
Jpmml-sparkml can only convert Spark ML pipelines to PMML. But I trained a spark rdd based gbdt mllib model, how can i convert the mllib model to pmml model.
PMML model export - RDD-based API show that only KMeansModel, LinearRegressionModel, RidgeRegressionModel, LassoModel, SVMModel, Binary LogisticRegressionModel can be converted to pmml model. What about gbdt model, Is there no method to convert it to pmml model?
Can anyone help me?