Skip to content

ENH, DOC: Add support for interactive examples for NumPy with jupyterlite-sphinx #26745

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 61 commits into from
Jan 28, 2025

Conversation

agriyakhetarpal
Copy link
Contributor

@agriyakhetarpal agriyakhetarpal commented Jun 18, 2024

Description

This PR adds support for WASM-powered interactive examples with JupyterLite and jupyterlite-sphinx in the NumPy documentation.

A similar enhancement was made for the documentation for SciPy (scipy/scipy#20019) and that for PyWavelets (PyWavelets/pywt#728).

How to test this

Tip

I have a GitHub Pages website deployed at https://agriyakhetarpal.github.io/numpy/ through which these changes can be experimented with. The CircleCI deployment does not render the examples properly, however, a fix is available as reported in #26745 (comment) and will be fixed upstream in jupyterlite/jupyterlite-sphinx#182.

Update, 24/07/2024: this fix is now available in jupyterlite-sphinx version 0.16.2, and the CircleCI deployment should render the examples properly as well.

Key changes (so far)

  1. Configuration for jupyterlite-sphinx in doc/source/conf.py and doc/Makefile, coupled with changes to styling in doc/source/_static/numpy.css
  2. Updates to the API documentation under files in numpy/ under the "Examples" section in the docstrings for various modules

Note

I have currently left out the following: _datasource.py, _utils_impl.py, _version.py, _array_api_info.py, ctypeslib.py, files under numpy/_typing/, and numpy/_core/numerictypes.py.

Tip

The "NumPy user guide" section has not been modified at this time due to the size of the PR (in terms of the number of files modified). However, it can be modified or adapted to a Jupytext notebook similar to scipy/scipy#20303 to make it work well in an interactive set-up.

Important

The NumPy documentation will have to be rebuilt and reuploaded in https://github.com/numpy/doc/ for these changes to show up in the official documentation.

What issue does this PR address?

Closes #25969

Additional context

@charris
Copy link
Member

charris commented Jun 18, 2024

This could use a release note.

Comment on lines 470 to 485
.. try_examples::

>>> import numpy as np

>>> a = np.memmap('newfile.dat', dtype=float, mode='w+', shape=1000)
>>> a[10] = 10.0
>>> a[30] = 30.0
>>> del a

>>> b = np.fromfile('newfile.dat', dtype=float)
>>> print(b[10], b[30])
10.0 30.0

>>> a = np.memmap('newfile.dat', dtype=float)
>>> print(a[10], a[30])
10.0 30.0
Copy link
Contributor Author

@agriyakhetarpal agriyakhetarpal Jun 18, 2024

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I haven't tested whether every example works because there are way too many, but I did test this and this is one of the examples that doesn't, and fails with the error

ValueError: mmap length is greater than file size

and reducing the shape from 1000 to 100 doesn't help: OSError: [Errno 43] No such device (edit: this error occurs on subsequent runs on the same cell because newfile.dat cannot be loaded)

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I'm not sure if there is a way to enable the Pyodide file system or interact with Pyodide directly through a Pythonic interface like this.

[skip cirrus] [skip azp]
@agriyakhetarpal
Copy link
Contributor Author

This could use a release note.

@charris, I addressed this in 484fe56. Please let me know if a "Highlight" note would be okay or if I should switch this to new_feature.

[skip cirrus] [skip azp]
@agriyakhetarpal
Copy link
Contributor Author

I forgot to skip Azure Pipelines and Cirrus in my previous commit – please don't mind the force-push, my apologies

@agriyakhetarpal agriyakhetarpal marked this pull request as ready for review June 18, 2024 17:48
@agriyakhetarpal
Copy link
Contributor Author

agriyakhetarpal commented Jun 18, 2024

I notice that for some reason the CircleCI build isn't able to load the examples, most likely because the files in source/contents/ weren't copied or that jupyterlite-pyodide-kernel was missing. I shall investigate.

Edit: I think I figured it out – CircleCI is storing the notebooks in a different path, but GitHub Pages does not do so.

Please use my GitHub Pages site to test things out: https://agriyakhetarpal.github.io/numpy/reference/maskedarray.generic.html, I have another branch at agriyakhetarpal#2 and I shall keep both of them in sync based on further commits that I push here.

Copy link
Member

@melissawm melissawm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I also don't see the try examples button locally - should I do anything differently other than spin docs?

@agriyakhetarpal
Copy link
Contributor Author

I also don't see the try examples button locally - should I do anything differently other than spin docs?

@melissawm, I'm not sure – I think you'll need to refresh your conda/mamba environment, to load the new dependencies or install requirements/doc_requirements.txt, perhaps? I am able to see the buttons locally.

@melissawm
Copy link
Member

Nevermind - I do see it now!

[skip actions] [skip azp] [skip cirrus]
@agriyakhetarpal
Copy link
Contributor Author

d818f8c disables source maps for JupyterLite, based on scikit-learn/scikit-learn#26246. This brings the size of the JupyterLite build in the doc/build/lite/ folder from 63.3 MiB (out of the total docs size of 180.7 MiB) to 22.4 MiB (with a total docs size of 140.2 MiB), which is a 22.4% reduction in build size. While doing this has the disadvantage of making debugging slightly harder for end users, it is favourable for the aspect of saving space, especially for GitHub Pages and https://github.com/numpy/doc. I've employed the same configuration for sympy/sympy#27419 as well.

This reverts commit ef3f14b.

[skip actions] [skip azp] [skip cirrus]
[skip actions] [skip azp] [skip cirrus]
[skip actions] [skip azp] [skip cirrus]
@agriyakhetarpal
Copy link
Contributor Author

Hi @rgommers, we published https://github.com/jupyterlite/jupyterlite-sphinx/releases/tag/v0.18.0 some moments ago (thanks @steppi!), and I've updated the versions of jupyterlite-sphinx and jupyterlite-pyodide-kernel in 0713994 and 369e27d – we are good to proceed now, thanks!

[skip actions] [skip azp] [skip cirrus]
[skip actions] [skip azp] [skip cirrus]
@agriyakhetarpal
Copy link
Contributor Author

Commit e7972a5 bumps to https://github.com/jupyterlite/pyodide-kernel/releases/tag/v0.5.1 that was released today, which brings the recent https://github.com/pyodide/pyodide/releases/tag/0.27.1.

(Apologies for the force push!)

Copy link
Member

@melissawm melissawm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hi @agriyakhetarpal ! The changes look good to me, I've successfully built this PR locally and it seems to work (although the jupyterlite blocks don't work locally).

The CircleCI version of the "Try it" works fine, but when I click "Open in new tab" I see this:

Captura de imagem_20250121_141043

Is this expected? (Maybe it's because it's in CI)

@agriyakhetarpal
Copy link
Contributor Author

agriyakhetarpal commented Jan 21, 2025

Hi @melissawm, thanks for the review! I don't think I can reproduce it with CircleCI. Could you please retry with incognito mode, or with a different internet connection or mobile data if you can? This is what I get:

JupyterLite notebook interface that displays an example from the NumPy documentation, more specifically, np.ma.HardMask, running in a kernel provided by the Pyodide project

For reference, I tried this link: https://output.circle-artifacts.com/output/job/4b11fff4-e43b-4010-af4e-de2e84deed66/artifacts/0/doc/build/html/lite/notebooks/index.html?path=88efbc39_8e7b_4e21_86c0_519bb66f4f6a.ipynb

Based on your screenshot, I guess you're using Firefox. I tried with it as well, both with incognito mode and without, and the JupyterLite UI loads perfectly.

although the jupyterlite blocks don't work locally

They should work locally as well. Could you please share any errors or output that you might be seeing?

@melissawm
Copy link
Member

I think they had a partial outage eariler today so that may have been just bad luck. I'll wait for the check to finish here and test again.

Copy link
Member

@melissawm melissawm left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Now it works fine! Thanks @agriyakhetarpal

Copy link
Member

@rgommers rgommers left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This looks pretty clean, nice work @agriyakhetarpal! Time to give this a go. Thanks to @melissawm and everyone else who contributed as well.

@agriyakhetarpal could you update us on the next steps here? I think it includes upgrading to NumPy 2.2 in Pyodide and enabling use of nightlies, and probably also enabling functionality on the User Guide? Anything else?

@rgommers rgommers merged commit 62b43b9 into numpy:main Jan 28, 2025
4 checks passed
@rgommers rgommers added this to the 2.3.0 release milestone Jan 28, 2025
@agriyakhetarpal agriyakhetarpal deleted the interactive-docs branch January 28, 2025 21:45
@agriyakhetarpal
Copy link
Contributor Author

agriyakhetarpal commented Jan 28, 2025

@agriyakhetarpal could you update us on the next steps here? I think it includes upgrading to NumPy 2.2 in Pyodide and enabling use of nightlies, and probably also enabling functionality on the User Guide? Anything else?

Thanks, @rgommers! The following steps here are indeed those, and also, we need to add some docs to resolve gh-27310, similar to @melissawm's SciPy PR (scipy/scipy#22385, which I'll be reviewing and we can adopt similar messaging for NumPy when done).

@rgommers
Copy link
Member

Thanks Agriya. Docs would be a nice next step here indeed, leveraging the work Melissa did for SciPy (the setups and considerations are almost identical). It'd also be fine with me to have a shorter explanation and link to the SciPy docs on the topic for more details.

Upgrading to NumPy 2.2 in Pyodide was last blocked due to [...]

This would be good to prioritize. It seems like there was a way forward (or two possibly) there, by reverting making symbols private for example.

Re User Guide: agreed with waiting for better tracebacks and doing 1-2 pages first.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

DOC: Add interactive examples with jupyterlite-sphinx
8 participants