Scikit-Learn to PMML Converter

Warnings

• breaking The `sklearn2pmml.make_pmml_pipeline()` utility function has been removed. Additionally, the `escape_func` parameter was moved from `make_pmml_pipeline()` to the `sklearn2pmml()` function.
  Fix: Directly use `sklearn2pmml.pipeline.PMMLPipeline` for pipeline creation and pass `escape_func` (if needed) to `sklearn2pmml()`.
• gotcha sklearn2pmml is a Python wrapper for a Java library and requires a Java Runtime Environment (JRE) version 11 or newer to be installed and accessible via the system's PATH environment variable. Without Java, conversion attempts will fail with a `RuntimeError`.
  Fix: Ensure Java 11+ is installed and configured correctly on your system path.
• gotcha The `sklearn2pmml` library is designed for *exporting* Scikit-learn models to PMML. It does not provide functionality to *import* PMML files back into Scikit-learn objects or Python for native scoring. For Python-based PMML evaluation, consider using the `jpmml-evaluator-python` library.
  Fix: Use `jpmml-evaluator-python` or another PMML evaluation library if you need to score PMML files within Python.
• gotcha When training models within `PMMLPipeline`, using `pandas.DataFrame` or `pandas.Series` for `X` and `y` is recommended. This allows `sklearn2pmml` to correctly capture and embed meaningful feature and target names in the PMML file. If NumPy arrays are used, feature names will default to generic 'x1', 'x2', etc., and the target name to 'y'.
  Fix: Pass `X` and `y` as `pandas.DataFrame` and `pandas.Series` objects, respectively, to the `fit()` method of `PMMLPipeline`.
• gotcha Direct conversion of highly custom Python classes, especially for complex data preprocessing (e.g., advanced text feature extraction using third-party libraries), is generally not supported. PMML has a limited set of expressible transformations, and arbitrary Python code cannot be translated.
  Fix: Refactor complex custom preprocessing steps into simpler, PMML-compatible transformers where possible, or perform them external to the PMML pipeline.
• gotcha Conversion of large or complex models can lead to out-of-memory errors in the underlying Java process. This often manifests as a `RuntimeError` with Java-related stack traces.
  Fix: Increase the Java Virtual Machine (JVM) memory allocation by passing `java_opts` to the `sklearn2pmml()` function, e.g., `sklearn2pmml(pipeline, 'model.pmml', java_opts=['-Xms4096m', '-Xmx4096m'])` to allocate 4GB of heap space.

Install

• pip install sklearn2pmml Install from PyPI

Imports

• PMMLPipeline

  from sklearn2pmml.pipeline import PMMLPipeline

• sklearn2pmml

  from sklearn2pmml import sklearn2pmml

  The `make_pmml_pipeline` utility function has been removed or deprecated in favor of directly using `PMMLPipeline` and `sklearn2pmml()`.

Quickstart

This quickstart demonstrates how to create a Scikit-Learn pipeline, wrap it with `PMMLPipeline`, fit the model, and then export it to a PMML file using the `sklearn2pmml()` function. The `PMMLPipeline` class enhances the standard Scikit-Learn pipeline with PMML-specific functionalities like capturing feature names from Pandas DataFrames.

import pandas as pd
from sklearn.datasets import load_iris
from sklearn.tree import DecisionTreeClassifier
from sklearn2pmml.pipeline import PMMLPipeline
from sklearn2pmml import sklearn2pmml

# Load a sample dataset
iris = load_iris(as_frame=True)
X, y = iris.data, iris.target

# Create a PMMLPipeline (extends sklearn.pipeline.Pipeline)
pmml_pipeline = PMMLPipeline([
    ('classifier', DecisionTreeClassifier())
])

# Fit the pipeline
pmml_pipeline.fit(X, y)

# Convert the fitted pipeline to PMML
pmml_filepath = 'DecisionTreeIris.pmml'
sklearn2pmml(pmml_pipeline, pmml_filepath)

print(f"PMML model successfully exported to {pmml_filepath}")