Support for PyCaret transformers

[szymoonl]

According to this comment, I tried to convert Pycaret model as follow:

prep_pipe = get_config('prep_pipe')
dt = create_model('dt')
final_dt = finalize_model(dt)

from sklearn2pmml.pipeline import PMMLPipeline

pmml_pipeline = PMMLPipeline([
	("prep_pipe", prep_pipe),
	("final_model", final_dt )
])

import pickle

with open("test_pmml.pkl", "wb") as pf:
	pickle.dump(pmml_pipeline, pf)

sklearn: 0.23.2
sklearn pandas: 2.2.0
sklearn2pmml: 0.84.1
pycaret: 2.3.6
openjdk version "11.0.15" 2022-04-19

The following exception occurred during conversion using a .jar:

java -jar /jpmml-sklearn/pmml-sklearn-example/target/pmml-sklearn-example-executable-1.7-SNAPSHOT.jar --pkl-input /test_pmml.pkl --pmml-output model_pmml.pmml
Jun 21, 2022 8:06:26 AM org.jpmml.sklearn.example.Main run
INFO: Parsing PKL..
Jun 21, 2022 8:06:26 AM org.jpmml.sklearn.example.Main run
INFO: Parsed PKL in 57 ms.
Jun 21, 2022 8:06:26 AM org.jpmml.sklearn.example.Main run
INFO: Converting PKL to PMML..
Jun 21, 2022 8:06:26 AM sklearn2pmml.pipeline.PMMLPipeline initTargetFields
WARNING: Attribute 'sklearn2pmml.pipeline.PMMLPipeline.target_fields' is not set. Assuming y as the name of the target field
Jun 21, 2022 8:06:26 AM org.jpmml.sklearn.example.Main run
SEVERE: Failed to convert PKL to PMML
java.lang.IllegalArgumentException: The transformer object (Python class pycaret.internal.preprocess.DataTypes_Auto_infer) is not a supported Transformer
at org.jpmml.python.CastFunction.apply(CastFunction.java:47)
at sklearn.pipeline.Pipeline$1.apply(Pipeline.java:108)
at sklearn.pipeline.Pipeline$1.apply(Pipeline.java:95)
at com.google.common.collect.Lists$TransformingRandomAccessList.get(Lists.java:638)
at sklearn2pmml.pipeline.PMMLPipeline.getHead(PMMLPipeline.java:629)
at sklearn2pmml.pipeline.PMMLPipeline.getHead(PMMLPipeline.java:642)
at sklearn2pmml.pipeline.PMMLPipeline.encodePMML(PMMLPipeline.java:198)
at org.jpmml.sklearn.example.Main.run(Main.java:226)
at org.jpmml.sklearn.example.Main.main(Main.java:151)
Caused by: java.lang.ClassCastException: Cannot cast net.razorvine.pickle.objects.ClassDict to sklearn.Transformer
at java.base/java.lang.Class.cast(Class.java:3605)
at org.jpmml.python.CastFunction.apply(CastFunction.java:45)
... 8 more

Exception in thread "main" java.lang.IllegalArgumentException: The transformer object (Python class pycaret.internal.preprocess.DataTypes_Auto_infer) is not a supported Transformer
at org.jpmml.python.CastFunction.apply(CastFunction.java:47)
at sklearn.pipeline.Pipeline$1.apply(Pipeline.java:108)
at sklearn.pipeline.Pipeline$1.apply(Pipeline.java:95)
at com.google.common.collect.Lists$TransformingRandomAccessList.get(Lists.java:638)
at sklearn2pmml.pipeline.PMMLPipeline.getHead(PMMLPipeline.java:629)
at sklearn2pmml.pipeline.PMMLPipeline.getHead(PMMLPipeline.java:642)
at sklearn2pmml.pipeline.PMMLPipeline.encodePMML(PMMLPipeline.java:198)
at org.jpmml.sklearn.example.Main.run(Main.java:226)
at org.jpmml.sklearn.example.Main.main(Main.java:151)
Caused by: java.lang.ClassCastException: Cannot cast net.razorvine.pickle.objects.ClassDict to sklearn.Transformer
at java.base/java.lang.Class.cast(Class.java:3605)
at org.jpmml.python.CastFunction.apply(CastFunction.java:45)
... 8 more

How to solve this? 🤔
Thank you in advance!

[vruusmann]

According to this comment, I tried to convert Pycaret model as follow:

Great to see that this old workaround is still valid!

However, I wonder if PyCaret has "systematized" their workflows, so that they could be programmatically converted to standard Scikit-Learn pipeline objects.

Exception in thread "main" java.lang.IllegalArgumentException: The transformer object (Python class pycaret.internal.preprocess.DataTypes_Auto_infer) is not a supported Transformer

Just as the exception message points out - there is a custom PyCaret transformer class pycaret.internal.preprocess.DataTypes_Auto_infer in your pipeline.

Potential solutions:

1. Write a SkLearn2PMML/JPMML-SkLearn handler for this custom transformer class.
2. Remove this step from the pipeline. Maybe if you specify column names/types explicitly, then PyCaret will have all the necessary information available, and won't perform any inference work by itself.

For starters, try to convert the model without pre-processing. When you can get this part working, only then start adding complexity (such as pre-processing).

[szymoonl]

When trying to convert the model alone without preprocessing, the following error appears:

Aug 16, 2022 12:48:12 PM org.jpmml.sklearn.example.Main run
INFO: Parsing PKL..
Aug 16, 2022 12:48:12 PM org.jpmml.sklearn.example.Main run
SEVERE: Failed to parse PKL
net.razorvine.pickle.PickleException: failed to setstate()
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:395)
at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:220)
at org.jpmml.python.CustomUnpickler.dispatch(CustomUnpickler.java:31)
at org.jpmml.python.PickleUtil$1.dispatch(PickleUtil.java:64)
at net.razorvine.pickle.Unpickler.load(Unpickler.java:109)
at org.jpmml.python.PickleUtil.unpickle(PickleUtil.java:85)
at org.jpmml.sklearn.example.Main.run(Main.java:163)
at org.jpmml.sklearn.example.Main.main(Main.java:151)
Caused by: java.lang.NoSuchMethodException: net.razorvine.pickle.objects.ClassDict.setstate(java.lang.Integer)
at java.base/java.lang.Class.getMethod(Class.java:2108)
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:392)
... 7 more

Exception in thread "main" net.razorvine.pickle.PickleException: failed to setstate()
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:395)
at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:220)
at org.jpmml.python.CustomUnpickler.dispatch(CustomUnpickler.java:31)
at org.jpmml.python.PickleUtil$1.dispatch(PickleUtil.java:64)
at net.razorvine.pickle.Unpickler.load(Unpickler.java:109)
at org.jpmml.python.PickleUtil.unpickle(PickleUtil.java:85)
at org.jpmml.sklearn.example.Main.run(Main.java:163)
at org.jpmml.sklearn.example.Main.main(Main.java:151)
Caused by: java.lang.NoSuchMethodException: net.razorvine.pickle.objects.ClassDict.setstate(java.lang.Integer)
at java.base/java.lang.Class.getMethod(Class.java:2108)
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:392)
... 7 more

The model is a random forest but trained with GPU, so it is a cuml object:

RandomForestClassifier()
<class 'cuml.ensemble.randomforestclassifier.RandomForestClassifier'>

Converting the model alone without GPU as a sklearn object works without problem. 🤔



When converting model with preprocessing pipeline I got below exception:

Aug 16, 2022 12:48:32 PM org.jpmml.sklearn.example.Main run
INFO: Parsing PKL..
Aug 16, 2022 12:48:32 PM org.jpmml.sklearn.example.Main run
SEVERE: Failed to parse PKL
net.razorvine.pickle.PickleException: failed to setstate()
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:395)
at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:220)
at org.jpmml.python.CustomUnpickler.dispatch(CustomUnpickler.java:31)
at org.jpmml.python.PickleUtil$1.dispatch(PickleUtil.java:64)
at net.razorvine.pickle.Unpickler.load(Unpickler.java:109)
at org.jpmml.python.PickleUtil.unpickle(PickleUtil.java:85)
at org.jpmml.sklearn.example.Main.run(Main.java:163)
at org.jpmml.sklearn.example.Main.main(Main.java:151)
Caused by: java.lang.reflect.InvocationTargetException
at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.base/java.lang.reflect.Method.invoke(Method.java:566)
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:393)
... 7 more
Caused by: net.razorvine.pickle.PickleException: Expected 8 attribute(s), got 9 attribute(s)
at org.jpmml.python.CustomPythonObject.createAttributeMap(CustomPythonObject.java:81)
at numpy.DType.setstate(DType.java:50)
... 12 more

Exception in thread "main" net.razorvine.pickle.PickleException: failed to setstate()
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:395)
at net.razorvine.pickle.Unpickler.dispatch(Unpickler.java:220)
at org.jpmml.python.CustomUnpickler.dispatch(CustomUnpickler.java:31)
at org.jpmml.python.PickleUtil$1.dispatch(PickleUtil.java:64)
at net.razorvine.pickle.Unpickler.load(Unpickler.java:109)
at org.jpmml.python.PickleUtil.unpickle(PickleUtil.java:85)
at org.jpmml.sklearn.example.Main.run(Main.java:163)
at org.jpmml.sklearn.example.Main.main(Main.java:151)
Caused by: java.lang.reflect.InvocationTargetException
at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at java.base/jdk.internal.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:62)
at java.base/jdk.internal.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.base/java.lang.reflect.Method.invoke(Method.java:566)
at net.razorvine.pickle.Unpickler.load_build(Unpickler.java:393)
... 7 more
Caused by: net.razorvine.pickle.PickleException: Expected 8 attribute(s), got 9 attribute(s)
at org.jpmml.python.CustomPythonObject.createAttributeMap(CustomPythonObject.java:81)
at numpy.DType.setstate(DType.java:50)
... 12 more

[vruusmann]

@szymoonl Please open a new issue for each unsupported ML framework.

Otherwise I'll classify all your messages as "spam", and send to trash.