TST: run generic tests

[mdhaber]

@JacobHass8 this was the beginning of a proof of concept of what a distribution library might look like.

Note how there is currently just one test of the PDF method. The thought is that the reference _pdf function is very easy to verify manually against a reference, and test_pdf confirms that the package's Normal.pdf matches the reference.

The other thing to do would be to run generic tests of all the methods against one another. We have a really thorough set of tests in https://github.com/scipy/scipy/blob/214919ec21fe7115fe8f1ce88b8e508cf034c8f7/scipy/stats/tests/test_continuous.py#L201. Would you be interested in helping me figure out how we could import that class and run this distribution through the suite?

One issue is the import at the top of https://github.com/scikit-stats/normal/blob/main/Normal/tests/test_distribution.py should be from Normal import Normal. But besides that, I'm having trouble with import scipy.stats.tests from the Pixi environment, even though scipy is installed and that same command works with other installations, say on Google Colab. Maybe SciPy needs to be installed with optional components?

I will create a separate library like this one for the stable distribution¹; I just thought we could pave the road a bit here in the meantime.

Footnotes

1. In the future, the idea is when contributor proposes adding a new distribution and the idea is accepted, a scikit-stats maintainer would create a new repo (like this normal one) from a template here in the scikit-stats organization. The contributor would fork the repo, adapt the template to implement the distribution, and open it as a PR against the original repo. When merged, the author of the PR would become a maintainer of the repo, and the distribution would be added to the scikit-stats meta-library. The intended advantages of this (admittedly somewhat complex) scheme are discussed in these slides. ↩

[JacobHass8]

Sorry for the stupid question, but I don't exactly know how to set up a Pixi environment. I'm running ubuntu so I added linux-64 to the pixi.toml file. However, running pixi shell, I get the error

Error:   × Failed to update PyPI packages for environment 'default'
  ├─▶ Failed to prepare distributions
  ├─▶ Failed to build `normal @ file:///home/jacob/Documents/normal`
  ├─▶ The build backend returned an error
  ╰─▶ Call to `flit_core.buildapi.build_editable` failed (exit status: 1)
      
      [stderr]
      Traceback (most recent call last):
        File "<string>", line 14, in <module>
          requires = get_requires_for_build({})
        File "/home/jacob/.cache/rattler/cache/uv-cache/builds-v0/.tmpGQY9bK/
      lib/python3.14/site-packages/flit_core/buildapi.py", line 31, in
      get_requires_for_build_wheel
          module = Module(info.module, Path.cwd())
        File "/home/jacob/.cache/rattler/cache/uv-cache/builds-v0/.tmpGQY9bK/lib/
      python3.14/site-packages/flit_core/common.py", line 59, in __init__
          raise ValueError("No file/folder found for module {}".format(name))
      ValueError: No file/folder found for module normal
      
      hint: This usually indicates a problem with the package or the build environment.

[mdhaber]

Huh, that's not supposed to happen. I'll test on another machine or two.

[mdhaber]

I don't know what's going on there. I am not experiencing the problem on my either my Windows or macOS machine, so I can't debug. I don't want to use AI for code generation, but for making recommendations about these tomls, I don't have a problem with it. Maybe ask an LLM?

[mdhaber]

You could try changing the build system, because there was really no reason to use flit.

[build-system]
requires = ["setuptools"]
build-backend = "setuptools.build_meta"

Another thing I see is that pixi.toml thinks the version is 0.0.0 whereas it's 1.0.0 in pyproject.toml. Maybe see if harmonizing those helps?

[JacobHass8]

I think it was completely user error. First off I installed flint, but I'm not sure if that was necessary because I got pixi working just fine. I just had to rename the folder containing the__init__.py file from Normal to normal.

[lucascolley]

I'm having trouble with import scipy.stats.tests from the Pixi environment, even though scipy is installed and that same command works with other installations, say on Google Colab. Maybe SciPy needs to be installed with optional components?

pixi add scipy-tests, see https://discuss.scientific-python.org/t/wheel-size-is-big-but-are-including-unit-tests/2239/2

[mdhaber]

Thanks!

[mdhaber]

Haha creepy but thanks : P (JK but like... how did you see this?!)
Would love to have you around, though! I am still terrible at this library management stuff.

[lucascolley]

You could try changing the build system

I've heard good things about https://docs.astral.sh/uv/concepts/build-backend/

[lucascolley]

Haha creepy but thanks : P (JK but like... how did you see this?!

less creepy than would be funny... you won the algorithm lottery

[mdhaber]

Weee I win. Again, you're welcome around here : ) Could definitely use your help. But I know you're busy.

[lucascolley]

Weee I win. Again, you're welcome around here : ) Could definitely use your help. But I know you're busy.

can you take a look at scipy/scipy#24669?

pixi.toml thinks the version is 0.0.0

I think you can just remove the workspace.version entry in pixi.toml.

[lucascolley]

see gh-2

[JacobHass8]

More user error here. The parameter default-environment isn't recognized on Pixi version 0.60. Updating to version 0.65.0 fixed this.