Interest in GSoC 2026: Priorsense improvements

Google Summer of Code
1

Hi everyone

My name is Emanuele, I am an applied math and data science master’s student and I am interested in the GSoC 2026 project Priorsense improvements.

I submitted a small Pull Request. It resolves an explicit TODO in R/plots.R by optimizing the expansion of indexed variables.

Additionally, I have been thinking about the scalability of the package for high-dimensional models . I have put together a short Notebook as proof of concept outlining a potential mathematical optimization as well as a more classic software engineering one. For the former, my core idea is to introduce a screening phase to filter out and avoid performing the Importance Sampling on every single parameter. This is done by using the variance of the log-likelihood thanks to which we could analytically approximate the KL divergence for small perturbations.

Since this is my first time participating in Google Summer of Code, I have questions on how to draft the proposal and process for you. Specifically:

  1. Is there a specific proposal template preferred by Stan that I should follow?

  2. How precise and granular should the weekly timeline and technical implementation details be at this stage?

  3. Would you be open to taking a quick look at my Notebook or draft of the proposal to let me know if this direction aligns with your vision for the project?

I look forward to your feedback

Emanuele

1 Like
2

Thanks for your interest! Tagging @n-kall, @avehtari, @Florence_Bockting, and @Osvaldo_Martin who know more about the priorsense project.

3

Hi, thanks for reaching out. We would be happy to review your notebook/proposal. The weekly timeline does not need to have a lot of detail. You can share a draft, and we can iterate over it.

1 Like
4

thank you so much! here you can find both the notebook ipynb and and the proposal in pdf : GitHub - Emanuele-Elias/priorsense_notebook · GitHub . Looking forward your feedback!!

2 Likes
5

@n-kall, @avehtari, @Florence_Bockting, @Osvaldo_Martin

6

Thank you @Emanuele-Elias for your interest in the project, and sorry for the delayed response from my side, I was travelling. I’m going through the proposal and will provide feedback shortly

7

Thanks again for sending the proposal and PoC @Emanuele-Elias . It looks great; well thought out and elaborated. I think this would be a great addition to the package (and method in general).

For the proposal, I only have very minor comments:

  • In the section on the broader ecosystem, you mention power-scaling sensitivity checks are available in ArviZ. As we (mainly @Osvaldo_Martin) are trying to keep ArviZ up to parity with priorsense, you could highlight here that improvements you will make to priorsense will benefit ArviZ too, as they can be ported over.
  • As your optimizations would make it feasible to perform sensitivity checks on thousands of variables, perhaps something can be mentioned about modifying the output of the user-facing functions to present results of thousands of checks more sensibly
1 Like
8

thank you for the feedback. I submitted my proposal. fingers crossed!

2 Likes