search

openml / automlbenchmark

OpenML AutoML Benchmarking Framework
https://openml.github.io/automlbenchmark
MIT License
391 stars 130 forks source link

Switch to using Python 3.8 instead of Python 3.7 #511

Closed Innixma closed 1 year ago

Innixma commented 1 year ago

Python 3.7 is no longer supported by Python, and some frameworks such as AutoGluon only support Python 3.8+ in their latest releases. Further, many packages such as scikit-learn, pandas, numpy, etc. no longer support Python 3.7.

I'd recommend AMLB use Python 3.8 or Python 3.9 as the new default. (I've tested setting Python 3.9, works out of the box with AutoGluon).

PGijsbers commented 1 year ago

I think 3.9 would be a better choice as the scientific python community is about to drop releases for 3.8 in a month as per NEP 29. A secondary concern would be to provide support for installations of older frameworks, while the Python versions can be independent in principle, they are not in practice. I think it's now always fixed to be versions.python and there isn't a good mechanism for pinning Python versions based on framework versions. Suggestions welcome.

Innixma commented 1 year ago

I think 3.9 would be a better choice as the scientific python community is about to drop releases for 3.8 in a month as per NEP 29. A secondary concern would be to provide support for installations of older frameworks, while the Python versions can be independent in principle, they are not in practice. I think it's now always fixed to be versions.python and there isn't a good mechanism for pinning Python versions based on framework versions. Suggestions welcome.

3.9 should be fine, I actually switched to using 3.9 and it appears to work for AutoGluon (AutoGluon supports 3.8-3.10 in latest v0.7 release). Let me know if you'd like me to edit the PR to be 3.9, or else I think you should be able to directly commit to this PR and update it on your end.

Innixma commented 1 year ago

@PGijsbers Please feel free to close this PR in favor of #520