jpmml/sklearn2pmml

Python library for converting Scikit-Learn pipelines to PMML

What it solves

SkLearn2PMML addresses the lack of portability in native Scikit-Learn estimators. Because Scikit-Learn models often lack adequate metadata (such as feature names), they are difficult to move between different environments. This tool allows users to convert Scikit-Learn pipelines into the Predictive Model Markup Language (PMML) format, enabling them to be deployed in diverse environments regardless of the original training platform.

How it works

The package acts as a Python wrapper for the JPMML-SkLearn library. It can convert fitted estimator objects or pickle files directly into PMML files using a utility function or a command-line interface.

To enhance portability and functionality, it provides PMMLPipeline, a meta-estimator that extends the standard Scikit-Learn pipeline. This extension allows for:

  • Metadata Collection: Automatically capturing feature and label names during the fit process.
  • Post-processing: Applying transformations to predictions via specific transformer attributes.
  • ** lemmas ** Model Verification: Embedding sample data within the model for self-checking during deployment.
  • Customization: Modifying the PMML XML fragments directly.

Additionally, the library provides specialized PMML-oriented transformers and predictors (e.g., ExpressionTransformer, CategoricalDomain, and GBDTLRClassifier) to handle complex data types and ensemble methods.

Who it’s for

Data scientists and ML engineers who use Scikit-Learn to build models but need to deploy them in environments that require PMML for interoperability and portability.

Highlights

  • Broad Compatibility: Works with Scikit-Learn versions 0.17 and newer.
  • Safe Unpickling: Uses a custom Java component to read pickle files, making it safer to use with unvetted files.
  • Enhanced Metadata: Captures feature and target names to ensure models are self-describing.
  • Flexible Deployment: Supports both a Python library API and a standalone command-line tool.

Related

  • Project
    jpmml/jpmml-sklearn

    A Java library and command-line tool that converts Scikit-Learn machine learning pipelines into the PMML standard for portable model deployment.

  • Project
    feature-engine/feature_engine

    A Python library providing Scikit-learn compatible transformers for comprehensive feature engineering and selection to improve machine learning model performance.

  • Project
    koaning/scikit-lego

    A collection of custom scikit-learn compatible transformers, metrics, and models to extend the functionality of machine learning pipelines.

  • Project
    csinva/imodels

    Python library offering a large suite of scikit‑learn‑compatible, interpretable models (rule sets, rule lists, trees, sparse linear models, etc.) with utilities for post‑hoc distillation and AutoML.

  • Project
    fnnx-ai/scikit-llm

    A library that integrates Large Language Models like ChatGPT into scikit-learn, enabling enhanced text analysis tasks using a familiar ML pipeline API.