This is exciting. It works: https://output.circle-artifacts.com/output/job/24f2114f-3815-45f3-8e2e-aec20d01b25a/artifacts/0/rtd_html/auto_examples/plot_0_sin.html#sphx-glr-auto-examples-plot-0-sin-py .

If at some point you can test a documentation with more examples such as matplotlib or scikit-learn's, it would be very useful to get a feeling of how well it works.

If at some point you can test a documentation with more examples such as matplotlib or scikit-learn's, it would be very useful to get a feeling of how well it works.

I'll take the bait :) Tested for MNE-Python (192 examples) in mne-tools/mne-python#11619. Spot checking I see that:

• A "background on temporal filtering" example points to a bunch of stuff about preprocessing
• An example about how to plot a cortical parcellation points to other examples about visualizing brain stuff.

So it seems to work well!

FYI I do get some warnings like:

auto_examples/time_frequency/time_frequency_simulated.recommendations:3: CRITICAL: Title level inconsistent:

More related examples
^^^^^^^^^^^^^^^^^^^^^

Thanks a lot, @larsoner ! I don't know MNE-Python that well. At a quick glance it seems that the recommendations are imperfect but definitely useful!

@ArturoAmorQ a couple of quick comments:

• I would use as a section title "Related examples", instead of "More related examples".

• You might consider using the rubric directive (https://docutils.sourceforge.io/docs/ref/rst/directives.html#rubric ) instead of a title. The reason is related to the warning reported by Eric above: it's impossible to get the title level consistent as the meaning of the type of underline depends on the order in which the underlines appear.

• I think that controlling the number of related examples display might be interesting

• Also, I was wondering: I notice that as you use scipy.sparse matrices, this feature will require the have scipy. That's a bit of a heavy dependency. Do you think that using plain numpy arrays is feasible? It will increase the memory cost, of course, but I wonder if this will lead to actually large memory costs

• If the recommendation code comes with additional dependencies, we need to make sure that if the feature is not enabled in the configuration file, the corresponding code is not imported (to avoid triggering those additional dependencies)

@ArturoAmorQ let me know when this is ready to go in!

@ArturoAmorQ let me know when this is ready to go in!

@lucyleeow On my side it's good to go :D

I hope that all the comments have been addressed. Please let me know otherwise.

Note CI failure due to: scipy/docs.scipy.org#80 (we use 'http://docs.scipy.org/doc/scipy/wrong_url' in our tinybuild)

Will merge once green, just in case.

@ArturoAmorQ I've made the last small changes.

I've made the last small changes.

Thanks @lucyleeow :)

Would be good to get this out there -- @lucyleeow do you have time to cut a release next week? If not, I can do it

Tentative yes. I'll ping you if not.

Sorry to post about it here, but I don't have folks emails. For GSOC this year, one of the proposed projects is an image based search. I think it could work as an extension of this feature/built on top if it, so I was wondering if any of the folks here had time/interest in being subject area mentors? Thanks!

Hmm... I'm not totally convinced it would be useful to other projects that use SG like scikit-learn, MNE, etc. Would it?

If it's mostly (or only) useful for matplotlib, it might be better to figure out how to interface with SG properly to insert some image-based search stuff. I could help with that at least. (Sphinx might already have the appropriate hooks to modify the reST files generated by SG for example, in which case nothing would even be needed at the SG end!)

it might be better to figure out how to interface with SG properly to insert some image-based search

The reason I was thinking custom solution is primarily b/c my end goal is a detexify type interface & also possibly a component based search. Granted like a detexify interface could hook into an image-based search - and also yeah that might be generally useful b/c I don't think my motivation of "folks may not know what a chart is called" is necessarily specific to matplotlib.

@lucyleeow would you be interested in mentoring this potentially? You might be a better fit

I was thinking the scikit-learn folks would be more knowledgable about machine learning algorithms but @story645 email me, we could be able to work something out.

Actually do you have my email? I'll make it visible on my github account.