Pyproject too restrictive

[alex47ST3]

Motivation: describe the problem to be solved
I am having a hard time trying to integrate nanny-ml in my library, basically I have conflicts with many different libraries, like dotenv, numpy, pydantic, pandas, scipy etc. I get that for some not widely used library it might be tricky, but I think that for the basic ones latest versions should be available.

Describe the solution you'd like
Maybe open the versions in your pyproject.toml or update the code to make sure it's able to run with more recent versions of all these basic libraries

Describe alternatives you've considered
I tried downgrading some libraries, but to be fair using such old versions of these libraries is not something we are very happy with, as we aren't able to use the latest features plus it could be a security risk

[nnansters]

Hey @alex47ST3 ,

yes to everything you mentioned. We've scheduled time to work on unpinning and updating the dependencies next week.
If there are any specific (major) versions of dependencies you'd like to see, definitely let us know.

Thanks for taking the time to try us out and log an issue!

[alex47ST3]

Thank you for the quick response @nnansters! Nanny is an amazing tool and it's pity to not be able to use it because some dependency conflict 😕

Taking a look to the pyproject and without about absolutely idea of the how it works internally (so take it with caution), I would suggest allowing minor versions in the following :

• python-dateutil
• scikit-learn
• joblib
• matplotlib
• plotly
• scipy
• seaborn
• kaleido
• pyarrow
• gcsfs
• pydantic
• rich
• click
• PyYAML
• Jinja2
• sqlmodel
• psycopg2-binary
• python-dotenv
• types-pyyaml
• types-python-dateutil

And major for:

• numpy
• pandas
• lightgbm

Thanks a lot again!

[nnansters]

Cheers! We'll take that into account. Also we have way too many dependencies (but we're solving that as we speak)!

[canoadri]

Hi @nnansters, any update on this? This is becoming a blocking issue for us.

[nnansters]

Starting on it right now!

[nnansters]

Take a look at #433 !

[alex47ST3]

thank you, but I believe that now nannyml is even more restrictive as that pyproject in #433 needs even newer version of some fundamental libraries.

For example it will only allow pandas >=2.2, <3.0, and in many libraries pandas 1 is still the needed version, so imo this will make very complicated to adopt NannyML.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[inigohidalgo]

Bump, the new versions are nice but is quite restrictive by default.

Good article about the issues with Poetry's restrictive pinning methodology: https://iscinumpy.dev/post/bound-version-constraints/

I know from experience how difficult it is to ensure compatibility with latest libraries, and I don't have a blocking issue here, but I wanted to keep the issue open as it was marked stale.

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.