CI for pypi release#6
Conversation
|
I also recommend you set up branch protections. Don't allow pushing directly to main, require reviews before merge, and require linear git history. Disable the "merge pull request" option. |
|
I also recommend you update you python pins (>=3.12, <3.15). 3.11 and below are not supported by the scientific stack per NEP 29 |
jessegrabowski
force-pushed
the
release-ci
branch
from
a57496a to
df0bf85
Compare
|
Tests (both rust and python) were not passing. I cleaned them up a bit but I'm not sure what the intent of them is/was. |
|
Thanks @jessegrabowski. This was a project started by @GStechschulte and then continued by @OttoVintola. So some mismatches have not been resolved yet, like the tests. I have added branch protection. We discussed with Otto about following https://scientific-python.org/specs/spec-0000/ |
| BATCH_SIZE = (0.1, 0.1) | ||
| NUM_TREES = 50 | ||
| NUM_PARTICLES = 10 | ||
| NUM_TREES = 10 |
There was a problem hiding this comment.
We could lower this more, even 2 should work
| with pm.Model() as model: | ||
| alpha = pm.Exponential("alpha", 1.0) | ||
| mu = pmb.BART("mu", X, np.log(Y), m=NUM_TREES) | ||
| mu = pmb.BART("mu", X, np.log(Y), m=NUM_TREES, response="gaussian") |
There was a problem hiding this comment.
Unrelated to this PR, but mentioning it here before I forget: @OttoVintola, we should use the value "constant" instead of "gaussian", both for backward compatibility and because I find "Gaussian" potentially confusing. During fitting/training, we use a Gaussian, but after that, the response is a constant value.
There was a problem hiding this comment.
The default value errors fyi
3 participants
Since we're on pymc-devs, trusted publisher needs to be configured. We can ask @maresb nicely to help.
When you do a release, you will have to remember to bump the version in cargo.toml. Not perfect. The alternative is to do a sentinel version (0.0.0) and automatically write from a github tag on release, but I felt like that was a bit too much for this PR.