Installation issues with Python 3.9

[aegerton]

There appears to be an issue installing on Python 3.9 due to numba installation issues as detailed in the following Issue;-

https://github.com/numba/numba/issues/6345#issue-718596580

Is there a workaround available, please?

[jbogaardt]

Hi @aegerton , Thankd for reporting as I wasn't aware of this. numba is a hard requirement by way of the sparse package. chainladder doesn't have too many dependencies, but the ones it has are major ones.

• pandas
• scikit-learn
• sparse
• dill
• xlcompose (deprecated)

Most of these packages have professional development teams supporting them, numba included. So if they don't have a work-around, I doubt I'd be skilled enough to find one - specifically a workaround for using chainladder on python>=3.9. conda environments help and if I create one with chainladder as the only requirement, it gives me python==3.8.5 if you're okay with a downgrade.

#121 is another recent chainladder issue due to upgrading to the latest in other dependencies. I think what I need to do with this issue is to create a "bleeding edge" CI/CD pipeline that tests against the latest releases of all dependencies to at least be aware of these conflicts when they arise.

[jbogaardt]

Oh, it's already in, which means it should be in the next release.

Originally posted by @hameerabbasi in pydata/sparse#426 (comment)

It looks like this should resolve once the sparse package does its next release. It looks like their workaround is:

Sparse now installs for python 3.9 if the exact following version is used as a dependency:
numba==0.53.0rc1.post1. But a regular pip install sparse does not work unless numba==0.53.0rc1.post1 is already installed.

Originally posted by @JackWetherell in pydata/sparse#426 (comment)

[aegerton]

Thanks @jbogaardt for the usual prompt and helpful response. It is greatly appreciated.

I'll keep an eye open for the sparse release and try again then. In the meantime, I have it working with Python 3.8.6 after downgrading.

Is it worth adding a note about dependency issues with 3.9 to the README, please?

[jbogaardt]

Is it worth adding a note about dependency issues with 3.9 to the README, please?

Yes, of course.

[hameerabbasi]

It looks like this should resolve once the sparse package does its next release. It looks like their workaround is:

Small correction, sparse doesn't need a new release, numba does. sparse should magically "just work".

[jbogaardt]

Thank you @hameerabbasi. You do fantastic work on sparse!

[JackWetherell]

I will leave a comment here once the final numba/llvmlite release is published, and the suggested workaround from my sparse issue is no longer needed.

[aegerton]

Is it worth adding a note about dependency issues with 3.9 to the README, please?

Yes, of course.

Thanks again @jbogaardt. Saw the change. Congratulations on the release of 0.8.0

[JackWetherell]

The numba/llvmlite release is published. The above workaround is no longer needed. sparse now installs with python 3.9.
The current versions of the relevant dependencies are:

sparse==0.11.2
numba==0.53.0
llvmlite==0.36.0

[jbogaardt]

Thank you @JackWetherell for letting us know. I'll give it a try.

[jbogaardt]

Resolved by #139