Cannot find the saved native string model though SUCCSESS files are created

Describe the bug I trained a mmlspark.lightgbm regressor which works greats and generates outputs on the test set. When saving it in a local directory on an azure/k8s pod using the following command:

reg_fitted.saveNativeModel(‘./savemodel/model02/’,overwrite=True)

I do not get the txt file containing the model parameters. I can see the content of that directory which has the _SUCCESS file, but cannot find the txt file that contains the model itself:

ls savemodel/model02/ -la

total 12 drwxr-xr-x 2 root root 4096 Jan 6 22:37 . drwxr-xr-x 4 root root 4096 Jan 6 22:37 … -rw-r–r-- 1 root root 8 Jan 6 22:37 ._SUCCESS.crc -rw-r–r-- 1 root root 0 Jan 6 22:37 _SUCCESS

To Reproduce Not sure how you can exactly reproduce this issue since no error is generated but nothing is saved either.

Expected behavior I expected to see a txt file in the same directory that describes the model. I have done it before with success on azure/databricks cluster, but it seems weird on an azure/k8s deployment.

Info (please complete the following information):

• MMLSpark Version: [e.g. v0.18.1]
• Spark Version [e.g. 2.4.4]
• Spark Platform [e.g. Azure/k8s]

** Stacktrace**

Please post the stacktrace here if applicable

If the bug pertains to a specific feature please tag the appropriate CODEOWNER for better visibility

Additional context Add any other context about the problem here.

AB#1984591

[OscarDPan]

FYI this is what I ended up doing:

        def call_java(model, name, *args):
            from pyspark import SparkContext
            from pyspark.ml.common import _java2py, _py2java
            m = getattr(model._java_obj, name)
            sc = SparkContext._active_spark_context
            java_args = [_py2java(sc, arg) for arg in args]
            return _java2py(sc, m(*java_args))

        def get_booster(model):
            return call_java(model, "getModel")

        def get_dump(lightgbm_model: LightGBMClassificationModel):
            jxgb = JavaWrapper(get_booster(lightgbm_model))
            return jxgb._call_java("model")

I think the output of get_dump() in the drive is exactly what I need - only not sure if there’s a better way

[OscarDPan]

Maybe let me put it this way. I want to train a lightGBM with mmlspark/pyspark on a Spark cluster but save/extract the native model and save the binary (lgb.Booster.save_model) locally for future usage.

How should I approach this?