2025 Hackathon Issue Assignments

[gmoss13]

Hi everyone, we’re excited for the start of the hackathon on Monday! Over the past few weeks, we have been curating issues for you to work on during the hackathon. We are now asking you all to have a look through the available issues, and think about which issues you would like to be working on during the hackathon. To make it easier to coordinate, we have divided the issues into high-level projects, and we provide an overview below. Each of the projects roughly corresponds to the GitHub issue label to make relevant issues easy to find. Note that most issues will be team efforts - you won’t need to tackle them alone!
Projects:

• Improving the test suite: The toolbox has a rather slow test suite with potential redundancies. We would like to streamline the test suite (Label: tests, coordinator: @janfb).
• New features: These are bigger projects to add a completely new feature to the toolbox, including methods for detecting model misspecification, or a bridge to probabilistic programming with pyro (Label: feature, coordinator: @janfb).
• Improving score-matching methods: Currently, our score-matching and flow-matching underperform. If you are familiar with these methods, there are various enhancements that can be made to our implementation to improve performance (Label: score-matching performance, coordinators: @gmoss13, @manuelgloeckler).
• More embedding nets: We would like to support more embedding nets, as well as embedding nets which can take in flexible input shapes (Label: embedding-nets, coordinator: @manuelgloeckler).
• Bug fixing: Miscellaneous bugs, which make good intermediate (potentially individual) projects (Label: bugs).
• Improving the general package architecture: Our code can be refactored to be more flexible and more readable in various ways. People tackling this should ideally have a solid software engineering background (Label: architecture, coordinator: @janfb).
• GPU-handling: Moving sbi objects to and from GPUs is clunky and can be annoying for investigating trained estimators which were trained on GPU on CPU (Label: gpu-handling, coordinator: @michaeldeistler ).
• Custom training loop: We should make it easier for users to write custom training loops by removing some code duplication and writing an explicit tutorial for how to write a custom training loop (Label: training-loop, coordinator: @michaeldeistler).
• Improving the documentation: We want to make a major refactoring of our documentation to make it more concise, consistent and structured. @michaeldeistler will do most of the work here, but he might need help from one person experienced with ReadTheDocs or design in general. (Label: docs with orthogonal smaller documentation issues, coordinator @michaeldeistler ).

In addition to the GitHub issues, we made a channel in the Discord server “Hackathon 2025” category for each of these categories. There you can already start discussing the issues and self-organize into groups working on different issues. This will help us hit the ground running at the start of the hackathon. Note that you can of course come up with your own issues to work on during the hackathon as well. Don’t hesitate to get in touch with any questions!