docs(notebooks): add google colab badge and install cell to model construction notebook

[Arun-Niranjan]

It would be good to be able to run our notebooks in google colab (makes talks and live demos much easier).

This change modifies the model_construction_tutorial notebook so that this works.

I'll do the others in a follow-up PR, just wanted this one to be small to make the changes obvious

[Arun-Niranjan]

logic:

• we only run this install if we are in a colab space
• we can install the vanilla dependencies, as colab ships with JAX and CUDA support already there. (in fact if you try %pip install "inferactively-pymdp[gpu]" -q you get an error)

[Arun-Niranjan]

@conorheins I can't remember what we decided on notebook outputs? it does make the diffs much more painful to evaluate, but I think for tutorial notebooks especially they're worth keeping so that people can see what the expected results are without necessarily having to run it.

[conorheins]

This is a nice idea and agree will make quick experimentation for users and demo-ing during presentations much more frictionless.

@conorheins I can't remember what we decided on notebook outputs? it does make the diffs much more painful to evaluate, but I think for tutorial notebooks especially they're worth keeping so that people can see what the expected results are without necessarily having to run it.

good question -- yeah the policy was to keep saved outputs. the relevant prior work are PRs #367 and #375. Short version of the policy:

• every non-legacy notebook under examples/ lives in either test_notebooks/ci_notebooks.txt(PR-CI tier) or tests/notebooks/nightly_notebooks.txt (heavier nightly tier).
• both tiers keep saved outputs -- exactly for the reason you said
• to keep diffs tractable, a local precommit hook (scripts/notebook_precommit.py) does two things on commit: strips top-level kernelspec / language_info metadata, and standardizes execution counts (output-bearing cells get 1, 2, ..., N and empty-output cells get null). That way local kernel history doesn't get committed

So keeping outputs in the model_construction_tutorial as you did here is the right call. The one thing missing is running the hooks -- I pulled the branch down and the current state still has kernelspec / language_info populated and execution counts in the 33–48 range from your local kernel. If you run:

uv sync --group dev
uv run --group dev pre-commit install
uv run --group dev pre-commit run --all-files

it'll fix the metadata + renumber the execution counts, and the diff will get a lot easier to review. Worth installing before you push #407 too since it touches all 20 notebooks. Docs for all of this are in test/notebooks/README.md and the "Recommended local hooks" section of CONTRIBUTING.md.

FYI the hooks aren't wired into a CI job ( only the ruff linting runs there) so this won't block the merge, but it's the documented convention.

[Arun-Niranjan]

ok that's done for this one, once this is merged I'll update the PR stacked on top.

[conorheins]

LGTM!

One heads-up for #407: most of the other 19 notebooks haven't been touched since the notebook standardization rule landed in #375, so when you run the hooks on that stacked PR you'll see each notebook diff come out bigger than just "add 2 cells"; there'll be incidental kernelspec/language_info stripping and exec-count renumbering baked in alongside the badge + install cell. That's expected and fine, just don't be surprised by the diff sizes