Meet our 2023 PyMC Interns#

Daniel Saunders#

Project Name

Expand support for spatial models in PyMC

Project Description

This project will improve PyMCs support for modeling spatial processes. There are many possible algorithms one may choose to work on, such as Gaussian process based methods for point processes like Nearest Neighbor GPs or the Vecchia approximation, and models that are types of Gaussian Markov Random Fields, like CAR, ICAR and BYM models. Implementions of these can be found in the R package CARBayes and INLA. Past work in PyMC has shown promising results and this project would build on that.

Info

  • Expected outcome: An implementation of one or more of the methods listed above, along with one or more notebook examples that can be added to the PyMC docs demonstrating these techniques.

  • Internship tag: GSoC

Mentors

photo of Daniel Saunders

Bio

I’m a philosophy PhD student at the University of British Columbia. My academic work looks at the foundations of behavioral science - what frameworks are best for understanding and modeling human behavior? What’s the right way to evaluate abstract theoretical models against data? I’m interested in Bayesian statistics because it presents some novel ways of thinking about those questions.

Connecting

Find out more at Daniel’s GSoC blog.

  1. What motivated you to apply for the intership with PyMC?

    A few years ago, I started reading Richard McElreath’s marvelous book Statistical Rethinking and fell in love with probabilistic programming. My prior coding experience was in Python so PyMC was a natural choice of probabilistic programming language to pick up. Since then, I’ve really grown to love the package and wanted to learn how to contribute to it in a serious way.

  2. Why did you choose your specific project topic?

    I knew I would be better suited to work on the modeling side than the backend or visualization side. So that left only a couple of suggested projects from the list put out by the PyMC team. Spatial statistics involves working with large covariance matrices which I think are really neat. So it was the logical choice.

  3. How did you get involved in open source software?

    I participated in a PyMC sprint in July 2022, organized with Data Umbrella. I remember Reshama Shaikh, Ravin Kumar, Rowan Schaefer, and Oriol Abril Pla being really nice and super helpful. They taught me how git works and how to tidy up doc strings.

    The world of open source was definitely overwhelming at first but I enjoyed being in a completely foreign environment. I just lurked on GitHub for the next few months before applying to GSoC.

  4. What are you expecting or hoping to get out of your internship experience?

    I want to grow my skills and my community. Working on large, collaborative software projects is a completely different challenge so I would like to get a feel for how they work. Similarly, my project is really going to push my knowledge of probabilistic programming forward. Getting to know who works in this space and what projects excite them is the other thing I’m really looking forward to.

  5. What are your career goals? How do you see the internship program moving you towards them?

    I’d like to work in probabilistic programming, either in academia or the private sector. My prior experience has been heavily academic so I’m eager to step just a bit outside that world and get connected with people who work on the private sector side. GSoC will diversify my experience in a way I hope opens new doors.

Shreyas Singh#

Project Name

Support Automatic Derivation of Arbitrary Censoring logp

Project Description

PyMC can automatically derive the logp of certain censoring processes such as left and/or right censoring, and rounding. This project would extend the ability to arbitrary forms of censoring of which left/right and rounding are just special cases. This would include interval censoring and binning.

This project will require learning how to parse PyTensor graphs, which is the backend used by PyMC. See https://www.pymc.io/projects/docs/en/v5.0.2/learn/core_notebooks/pymc_pytensor.html for more details. An understanding of probability theory is helpful but not a requirement (you can learn as you go)

Info

  • Expected outcome: PyMC logprob submodule will understand arbitrary censoring encoding models.

  • Internship tag: PyMC research

Mentors

photo of Shreyas Singh

Bio

I am an incoming Master’s student in Scientific Computing at the University of Pennsylvania and my areas of interest include symbolic computation and probabilistic programming. I have worked on using statistical analysis tools in computational physics followed by software development at Accenture Japan. I am fascinated by the computational backend structures used in data science and to that end, I have been enjoying contributing to PyTensor and PyMC in my relatively new open-source journey.

Connecting

Find out more at Shreyas’s internship blog.

  1. What motivated you to apply for the intership with PyMC?

    My undergraduate major was Engineering Physics, but it was while doing a minor in Mathematics that I realized my affinity for statistics and computational math. Although I had prior development experience in Java at my job, I wanted to pursue probabilistic programming with more rigor.

    While going through the organizations that participated in GSoC, I found out about PyMC. The extensive PyMC examples, PyMCon Web Series, and an active community both on GitHub and Discourse were what appealed to me the most.

  2. Why did you choose your specific project topic?

    I was quite fascinated by PyTensor, the computational backend of PyMC, and how it accounts for various kinds of operations and optimizations under the hood. The concept of log-probabilities, one of the central blocks in PyMC, and graph computation were all very intriguing. Additionally, the usage of arbitrary censoring in survival analysis, especially in infectious disease research was a driving factor too.

  3. How did you get involved in open source software?

    Despite having prior experience in software development and data analytics, it was only recently that I got introduced to the world of open source. A few friends suggested contributing to open-source software as the learning curve is sharp but rewarding. I have been amazed by the diversity of contributions and the willingness of developers from all backgrounds to work together towards a common goal and share their knowledge with those who wish to learn.

  4. What are you expecting or hoping to get out of your internship experience?

    I hope to add the log-probability inference for as many cases of arbitrary censoring as possible, taking in some special edge cases into consideration as well. The end goal would also be to port these likelihoods into Bambi and proper documentation for the entire logprob submodule. Overall, I expect to learn a lot from my mentors and other members of the organization, and become a regular contributor to PyMC along with learning the best practices in open source.

  5. What are your career goals? How do you see the internship program moving you towards them?

    I aspire to become a proficient Data Scientist in the sectors pertaining to sciences, such as meteorology, astronomy or healthcare. The internship program with PyMC and especially a project as engaging would provide me with a strong foundation for statistical modeling and machine learning. I had already got to learn a great deal during the application phase of GSoC and the variety of opportunities in open-source projects would keep my curiosity piqued.

Gabriel Stechschulte#

Project Name

Better tools to interpret complex Bambi regression models

Project Description

Bambi allows building Generalized Linear Models for Location, Scale, and Shape. The interpretation of parameter estimates can be challenging, especially when the model contains several predictors of different nature, possibly transformed, and model parameters are modified with link functions. To simplify the understanding, researchers often prefer to concentrate on simpler and easily interpretable quantities and visualizations. However, calculating these quantities and their standard errors is both time-consuming and non-trivial. Bambi currently has some visualization features currently has some visualization features to aid in comprehending model predictions, and this plan aims to enhance these features. A useful reference for our goals is the R library marginaleffects.

Info

  • Expected outcome: An implementation of one or more plotting functions to aid in the interpretation of Bambi’s models

  • Internship tag: GSoC

Mentors

photo of Gabriel Stechschulte

Bio

I have a MSc. in Data Science and work as a Systems Engineer in the supply chain department of an elevator manufacturing / production company. I develop hierarchical regression models to analyze costs and profitability of our different product lines and configurations as well as perform optimizations to reduce material waste and cost.

Connecting

Find out more at Gabriel’s GSoC blog.

  1. What motivated you to apply for the internship with PyMC?

    A personal objective of mine for 2023 was to begin contributing to open source probabilistic programming libraries to: (1) deepen my knowledge and skill sets within Bayesian statistics and software development, and (2) to “give back” to the PPL open source community after having used the software for the previous 1-2 years, and (3) to meet like minded people within the probabilistic programming field.

  2. Why did you choose your specific project topic?

    Although the model building portion of the Bayesian workflow has become easier, the interpretation of these models has not. Interpretation of generalized linear models are cumbersome even for the modeler; add on top the need of explainability to management and other non-technical stakeholders. Thus, I see my project topic as a way to automate certain aspects of model interpretability and as a way to present complex models to a non-technical audience more effectively.

  3. How did you get involved in open source software?

    I follow most of the PyMC, Bambi, Aesara, and Blackjax core devs on Twitter and GitHub. Seeing how supportive they (and the communities) are towards beginner developers wanting to contribute showed I should not be afraid. My first merged PR was documentation related in Blackjax haha.

  4. What are you expecting or hoping to get out of your internship experience?

    I am expecting to:

    • Improve software engineering knowledge and skill sets such as writing tests, robust code (error handling and shape handling), and object oriented programming.

    • Dive into the Bambi and xarray libraries in greater depth.

    • Meet, communicate, and learn from the other devs of the Bambi library.

    • To merge all three of my project deliverables on time.

  5. What are your career goals? How do you see the internship program moving you towards them?

    One of my career goals is to work at a company where we not only utilize open source probabilistic programming (and related) libraries, but are allocated a certain percentage of resources (time and money) to further develop and improve those libraries.

PyMC: Past, Present, and Future