split dependencies into inference (mandatory), training and linters

[aditya0by0]

• chebifier has many (too many) dependencies python-chebifier#7

[sfluegel05]

If I understand the extras-system correctly, given the dependencies

dependencies = ["A"]
[project.optional-dependencies]
opt1 = ["B"]
opt2 = ["C"]

you will get the following install options:

pip install my_package -> installs A
pip install my_package[opt1] -> installs A and B
pip install my_package[all] -> installs A, B and C

Given this structure, I would set the dependencies you put under inference as dependencies and make the rest optional (please correct me if I got this wrong)

[aditya0by0]

@sfluegel05, Thanks for pointing this out. I have made changes as per your suggestion. Please review.

[aditya0by0]

As this PR and the change (https://github.com/ChEB-AI/python-chebai/pull/105/files#diff-faff1af3d8ff408964a57b2e475f69a6b7c7b71c9978cccc8f471798caac2c88L27-R28) in another PR has conflicting change, it will be better to merge the another PR first and then merge the dev with this PR, to keep the corresponding change from this PR.

Final change to this dev should be the below
python -m pip install -e .[dev]

[sfluegel05]

can you split the dev dependencies into two parts? I see 2 distinct use cases: a) use chebai to generate datasets and train models, b) develop chebai and commit things to the repo (only the latter need things like pre-commit)

[sfluegel05]

this should be fine now. I removed some dependencies that appeared twice and pinned lightning to 2.5.1 (#124).

Also, I had some performance issues while installing chebai. It turns out that I had some large files in the repository (data, model checkpoints, ...) that were scanned during installation. I changed [tool.setuptools.packages.find] s.t. only chebai is scanned