After updating version , I used the same code and data set, but the AUC changed a lot

[NEXUSkeyis]

Today, I updated MMLSpark from V0.18 to V1.0.0-rc3, and then I tried to run the same code. But the results were quite different. Here are the parameters and results.

lgbm = LightGBMClassifier(numIterations=50, objective='binary', learningRate=0.1,lambdaL1=0.6, lambdaL2=0.01, baggingFraction=0.1, maxDepth=6, featureFraction= 0.6, numLeaves=31, labelCol=cur_label, verbosity=3,isProvideTrainingMetric=True )

AUC on the new version group AUC 0 train 0.723570 1 test 0.594123

AUC on the old version group AUC 0 train 0.845109 1 test 0.811383

I ran them separately more than once, but the gap was still large, and it didn't seem to be an accident. I wonder what might have caused it.

AB#1984518

[Ereebay]

same with me, upgrade from 0.18.1 to 1.0.0 rc3

[candalfigomoro]

See https://github.com/microsoft/SynapseML/issues/919