Writing array-library-agnostic code using the Python array API standard (with a hint of cosmology)

[Saransh-cpp]

Title

Writing array-library-agnostic code using the Python array API standard (with a hint of cosmology)

Describe your Talk

Over the years, the lack of an array data type in Python has resulted in the creation of numerous array libraries, each specializing in unique niches but still having some interoperability between each other. NumPy has become the de facto array library of Python, and the other array libraries try to keep their API close to that of NumPy. However, this often becomes infeasible, and the libraries deviate out of necessity. To make Python's array libraries shake hands with each other without any inconsistencies, the Consortium for Python Data API Standards has formalized an Array API standard for libraries offering array creation and manipulation operations.

A user usually finds it trivial to choose an array library (based on the suggestions on the internet forums), but things get out of hand once they start exploring use cases that they had not thought about initially. For instance, a user might attempt to make their NumPy code differentiable but hit a blocker when they realize their functions do not work with the arrays of an automatic-differentiable library such as JAX. Similarly, specific applications require CPU and GPU support; hence, the code written should be able to work with both CPU and GPU arrays. The Array API standard allows users to write and use the same code for arrays belonging to any of the standard-conforming libraries.

Through this talk, we will explore the need for such standardization and discuss its salient features in detail. We will further delve into several real-life examples of large and relatively small libraries that have migrated their codebases to adopt this standard. These examples will be sourced from my experience working with and on multiple Python array libraries for scientific computing (mostly cosmology and particle physics). Ultimately, the audience can expect to leave the room with the knowledge to write device-agnostic and array-library-agnostic code for their use cases.

Pre-requisites & reading material

• familiarity with NumPy
• a rough idea of Python's array ecosystem might be beneficial
• some scientific curiosity is appreciated

Resources

No response

Time required for the talk

30 minutes; I can add more content if people want a 45-60 minute talk; cannot be a lightning talk

Link to slides/demos

No response

About you

Yearly pilgrimage to India and Delhi means yearly pilgrimage to a PyDelhi event :)

---

Previous talks at PyDelhi:

• Python packaging: from stone age to the future #215
• The scientific side of Python and Open-Source: using Python for High Energy Physics at CERN  #275

---

I am a research software engineer at University College London's Centre for Advanced Research Computing, where I am partly funded by grants from the Edinburgh Parallel Computing Centre and the UK Space Agency to work on GLASS, a cosmology library for full-universe simulations.

Before UCL, I was based at CERN (European Organisation for Nuclear Research), where my work revolved around high-performance+scientific computing in Python and computational high-energy physics. Before CERN, I was a B.Tech (IT and Maths) student at the University of Delhi.

See my website for a more detailed profile.

Availability

August 2025

Any comments

Strictly not available before and after August :(

My availability will become more clear with time, but I will most probably make it to the August event. As always, opening this issue in advance to secure a spot.

[agriyakhetarpal]

Hi @Saransh-cpp, thanks for proposing a talk! I would be happy to accept this for the August edition.

Could you please add some slides to the issue description whenever it's possible for you? This is not a strict requirement, considering your previous experience speaking at PyDelhi and that we know each other in our personal and professional lives. Still, we've recently been asking everyone to do so and we're trying to formalise this better, solely to preserve speakers' presentations and as a means of archival via these issues on GitHub. You may do so at any time close to or after the meetup, or add some draft slides from your previous talks on the topic – that would be completely alright.

Also, ccing @Schefflera-Arboricola, given her experience working with the array API standard and in case she would have anything to add. :)

[Saransh-cpp]

Thanks for the quick review, @agriyakhetarpal! All sounds good! I should be able to add my slides 1-2 weeks before the event in August. Looking forward to the meet-up! :)

[github-actions]

Hi there! This proposal has been automatically marked as "inactive" due to it
not receiving any activity for the last sixty days.

To you as a contributor to the PyDelhi community, thank you for your proposal
and apologies if this issue fell through the cracks! This ping was created to
help bring this proposal a fresh pair of eyes. If you need any help, you can
always reach out here on this issue, or email us at
pydelhi.org@gmail.com.

If you think that this talk is no longer relevant, or that you are no longer
available to deliver this talk at one of our meetups in the coming time, you
may either close this issue, or the currently active volunteers shall do it
at some point in the future. If you are still interested in delivering this
talk, please comment on this issue to indicate your interest, and the
"inactive" label will be dropped automatically.

If you are a volunteer for the PyDelhi community, and if you no longer have
the bandwidth to dedicate to reviewing this proposal, please consider
reaching out to your fellow volunteers to help you with this proposal on the
appropriate communication channels.

If this issue is not a talk proposal, please ignore this message, remove the
"inactive" label, and either continue with your work here as planned or
close this issue if it is no longer relevant.