Closed AlekseiPol closed 4 years ago
It's probably a Mac OS-specific issue - I can recall that Java's floating-point computations yield slightly different results on Mac than on Linux or Windows.
Probably won't fix.
You have two workarounds. First, there is no need to build JPMML-LightGBM locally, because there's a library JAR file available in the Maven Central repository, and an executable uber-JAR file in the "releases" page: https://github.com/jpmml/jpmml-lightgbm/releases
Second, you can always disable integration tests during the build:
$ mvn -Dmaven.test.skip=true clean install
Thank you!
Hi!
I've got the next error during installation. Fixed by editing ClassificationAuditInvalid for those 3 predictions. I guess difference in 16 decimal is not critical, but the program does not build without this test.
MAC OS - 10.15.2 Python 3.7.3 llightgbm 2.3.1 maven 3.6.3 java version "13.0.1" 2019-10-15 Java(TM) SE Runtime Environment (build 13.0.1+9) Java HotSpot(TM) 64-Bit Server VM (build 13.0.1+9, mixed mode, sharing)
One more small thing: lgbm.fit(boston.data, boston.target, feature_name = boston.feature_names) The type of boston.feature_names is 'numpy.ndarray' and the one should convert it to list before passing to fit function. (at least for my python and modules versions)