environment: anaconda pinned at 2024.06, two years behind the rest of the series

[mmcky]

This repo pins anaconda=2024.06 in environment.yml, while the rest of the Python series has moved on: lecture-python-programming.myst, lecture-python.myst, lecture-python-advanced.myst, lecture-python-intro and lecture-jax all now pin anaconda=2026.07. That is a two-year gap, and it means this series builds against a materially different scientific-Python stack from every other series.

This surfaced while closing QuantEcon/meta#171, the 2025.06 rollout wave. That issue never listed this repo, so the drift has never actually been tracked anywhere.

Worth deciding rather than assuming. The pin may be deliberate — this series is used in taught courses where a stable environment across a semester matters more than currency. If so, the right outcome is to record that intent in the file rather than close the gap, so the next person auditing environments does not read it as neglect.

One thing to keep separate. The quantecon-book-theme==0.8.0 pin immediately below carries the comment "jupyterhub links need to be the older style for datascience lectures". That is an explicit, reasoned hold and should not be swept along with an Anaconda bump — the two pins are independent decisions.

Possibly related: #191 (error creating the conda environment for this repo).

Suggested next step: confirm whether the 2024.06 pin is intentional, then either bump toward 2026.07 with a fresh execution run, or annotate the pin with the reason it is held.

[mmcky]

Decision: the pin is not deliberate, but the upgrade is deferred. It was drift rather than a considered hold — but bringing this series up to anaconda=2026.07 needs real work, so the pin stays at 2024.06 for now rather than being bumped opportunistically.

That answers the question this issue asked. What follows is what the work would involve, recorded so the next attempt does not start from scratch.

Three dependencies that will not survive the bump as-is

Package	Latest release	Problem
descartes	1.1.0, January 2017	Nine years unmaintained. It exists in this file to provide PolygonPatch for geopandas plotting, and that broke with modern matplotlib. This is the hard blocker
quandl	3.7.0, November 2021	Deprecated in favour of Nasdaq Data Link — and nasdaq-data-link (1.0.4, August 2022) is pinned in the same file. Carrying both is redundant; the migration was never finished
arviz==0.13.0	current is 1.2.0, June 2026	Pinned three major versions back. 1.2.0 requires Python ≥3.12, which this repo already pins, so the constraint is our pin rather than the package

Less urgent but worth a look during the same pass: pyLDAvis (3.4.1, April 2023) and interpolation (2.2.7, July 2024) are both stale, though neither is abandoned. gensim is fine — actively maintained, 4.4.0 in October 2025.

One blocker that has just cleared

pandas-datareader is pinned here and imported by tools/classification.md and tools/visualization_rules.md. Until recently it would have been a hard stop: it was incompatible with pandas 3, which anaconda=2026.07 brings. That changed in June 2026 — 0.11.0 restored pandas 3 compatibility by vendoring the private pandas API it depended on. So this is no longer a reason to hold back. Related discussion in QuantEcon/data-lectures#26.

Sequencing: fix the CI gate first

This repo is one of six where a broken notebook currently passes CI green — its Build Download Notebooks (sphinx-tojupyter) step carries neither -n -W nor set -eo pipefail, so a CellExecutionError is a non-fatal warning and the step's exit code comes from the trailing rsync regardless. See QuantEcon/meta#340.

That matters here specifically: attempting an Anaconda bump against a build that cannot fail would let a broken upgrade land looking green. The CI fix should go in before the environment work starts, so the upgrade attempt gets honest feedback.

Unchanged

The quantecon-book-theme==0.8.0 pin below stays exactly as it is. Its comment — "jupyterhub links need to be the older style for datascience lectures" — is an explicit, reasoned hold and is independent of the Anaconda question.

Also still relevant: #191, on errors creating the conda environment for this repo.