kooplearn Introduces Scikit‑Learn Compatible Tools for Evolution Operator Learning

Global: kooplearn Introduces Scikit‑Learn Compatible Tools for Evolution Operator Learning

A new open‑source library named kooplearn, developed by Giacomo Turri, Grégoire Pacreau, Giacomo Meanti, Timothée Devergne, Daniel Ordonez, Erfan Mirzaei, Bruno Belucci, Karim Lounici, Vladimir Kostic, Massimiliano Pontil, and Pietro Novelli, was submitted to arXiv on 24 December 2025. The library implements linear, kernel, and deep‑learning estimators for both discrete‑time Koopman/Transfer operators and continuous‑time infinitesimal generators, enabling spectral analysis and data‑driven modeling of dynamical systems.

Scikit‑Learn API Integration

kooplearn follows the scikit‑learn estimator interface, allowing users to employ familiar methods such as fit, transform, and predict. This design choice facilitates seamless incorporation into existing Python‑based machine‑learning pipelines and promotes reproducibility across research projects.

Supported Evolution Operators

The library provides algorithms for learning discrete‑time evolution operators (often referred to as Koopman or Transfer operators) as well as continuous‑time infinitesimal generators. By estimating these operators, practitioners can obtain spectral decompositions that reveal dominant modes, decay rates, and frequencies of the underlying dynamical system.

Benchmark Datasets and Reproducibility

To encourage transparent evaluation, kooplearn ships with a curated collection of benchmark datasets covering synthetic chaotic systems, fluid‑dynamics simulations, and real‑world time‑series. The datasets are versioned and documented, supporting fair comparison of algorithmic performance.

Use‑Case Scenarios

Potential applications highlighted by the authors include reduced‑order modeling of high‑dimensional physical processes, forecasting of observable quantities, and extraction of interpretable dynamical features for scientific discovery.

Availability and Documentation

The source code is hosted publicly and can be accessed via the URL provided in the arXiv abstract. Comprehensive documentation, example notebooks, and API references are included to assist both novice and experienced users.

Future Development

The authors indicate plans to extend kooplearn with additional deep‑learning architectures, automated hyper‑parameter tuning, and integration with GPU‑accelerated backends, aiming to broaden its applicability to large‑scale datasets.

This report is based on information from arXiv, licensed under Academic Preprint / Open Access. Based on the abstract of the research paper. Full text available via ArXiv.