Thomas J. Fan

Open-source work

Fan joined the scikit-learn team as a maintainer more than six years before a May 2025 retrospective post, and the project's About page lists Fan among its active maintainers; scikit-learn is released under the BSD 3-Clause License. In that retrospective Fan highlighted contributions such as native categorical and missing-value support in tree models, cost complexity pruning, pandas and Polars DataFrame output through the set_output API, and feature name propagation with get_feature_names_out.⁴²⁵

Other features Fan worked on include the TargetEncoder for categorical data, grouping of infrequent categories in scikit-learn's encoders, KNN-based missing value imputation, an HTML representation for visualizing estimators in Jupyter notebooks, a plotting API for inspecting estimators, and experimental GPU support through the Array API standard so that scikit-learn can run on PyTorch or CuPy arrays.⁴

Teaching and writing

Fan speaks regularly at Python data science conferences. Talks listed on Fan's site include a deep dive into scikit-learn's histogram-based gradient boosting at PyData NYC 2019, talks on skorch, which combines scikit-learn and PyTorch, and on Poisson regression at SciPy, and talks on parallelism in numerical Python libraries, scikit-learn on GPUs with the Array API, and pushing Cython to its limits in scikit-learn.⁶

Fan's blog covers performance and tooling topics, including PyTorch graph capture and data-dependent control flow, portable torch.compile caches, serializing models with torch.export, NumPy ufuncs written with Cython 3.0 and Python extensions in Rust. A July 2026 post set out five principles for using large language models in software engineering. Fan is currently a Member of Technical Staff at Modal.³¹

Work and contributions

• scikit-learn Open-source code

  Listed in the active Maintainers Team on the scikit-learn About page; Fan's GitHub profile describes the role as scikit-learn maintainer.²¹⁵

Timeline

1. 2019Presented a deep dive into scikit-learn's HistGradientBoosting estimators at PyData NYC 2019.⁶
2. 2019Presented skorch, a union of scikit-learn and PyTorch, at SciPy 2019.⁶
3. 2022Spoke on parallelism in numerical Python libraries at PyData NYC 2022.⁶
4. 2023Presented scikit-learn on GPUs with the Array API at PyData NYC 2023.⁶
5. 2024Presented "Pushing Cython to its Limits in Scikit-learn" at PyData NYC 2024.⁶
6. May 15, 2025Published a retrospective on six years as a scikit-learn maintainer.⁴
7. Jul 3, 2026Published "Five Principles for Using LLMs for Software Engineering".³

Where they work

• Modal 3 people in the wiki

Related people

• Erik BernhardssonAlso at Modal
• Akshat BubnaAlso at Modal
• Nikita ShulgaML frameworks and libraries · Developer tools
• Matthew WatsonML frameworks and libraries · Developer tools
• Bratin SahaML frameworks and libraries · Developer tools
• Tim DavisML frameworks and libraries · Developer tools
• Jake VanderPlasML frameworks and libraries · Developer tools
• Chip HuyenDeveloper tools · ML frameworks and libraries
• William FalconML frameworks and libraries · Developer tools
• Piero MolinoML frameworks and libraries · Developer tools