Tyler James Burch

Applying research methods, statistics, machine learning, and Bayesian modeling to solve problems in professional baseball analytics.

Open Source Contributions in 2026

☕ Buy me a coffee

Below are changes merged in 2026 and active pull requests I’ve submitted to open-source projects this year. Dates show when merged changes landed or when active contributions were submitted.

5 contributions Last updated
  1. Merged

    tune — Clearer printing for control objects

    PR #1108

    A small quality-of-life addition to help out a library I've used a lot. Added dedicated print methods that use cli to show settings such as verbosity, parallel processing, prediction storage, and event levels as a readable list. This makes control objects (and their defaults) faster to inspect.

  2. Merged

    finetune — Preserve resample weights in racing methods

    PR #135

    Last year I contributed fold weights to tune to help enable variable-sized folds. While testing them with racing methods, I noticed that randomize_resamples() did not carry them correctly. This PR fixed that.

  3. Open

    arviz-stats — Add bivariate histogram and hexbin statistics

    PR #438

    Added public histogram2d() and hexbin() functions with support for NumPy, xarray, batched computation, weights, density normalization, explicit extents, and nonfinite samples. This gives every ArviZ plotting backend the same binning logic and the edges or centers needed to draw matching binned 2D distributions.

  4. Open

    arviz-stats — Support short chains in nested R-hat

    PR #442

    Made the minimum draw requirement for rhat_nested() depend on the estimator, allowing one draw per chain for the identity method and two for split-based methods (while handling several short-chain edge cases). This supports GPU-friendly sampling workflows that run many short chains in parallel. The previous shared validation did not support that use case, returning nan for valid inputs with fewer than four draws.

  5. Open

    bambi — Add nonlinear formulas

    PR #1006

    Added nonlinear expressions for likelihood parent parameters while retaining ordinary Bambi formulas for each named parameter, including common, group-specific, and offset terms. This brings nonlinear models into Bambi's existing formula, prior, prediction, and link machinery, enabling a new class of models. One example is the classic Gelman golf model, which I reproduced end to end.

Updated:

Cite This Post

@misc{burch2026open-source-contributions,
  author       = {Tyler James Burch},
  title        = {Open Source Contributions in 2026},
  year         = {2026},
  month        = {September},
  howpublished = {\url{https://tylerjamesburch.com/blog/open-source/open-source-contributions}},
}

You May Also Enjoy