search

dwavesystems / penaltymodel

Utilities and interfaces for using penalty models.
https://docs.ocean.dwavesys.com/projects/penaltymodel/en/latest
Apache License 2.0
18 stars 21 forks source link

Is penaltymodel-lp 0.1.5 compatible with numpy 1.19.1 and scipy 1.5.2? #134

Closed ccalaza closed 2 years ago

ccalaza commented 2 years ago

Current Problem We are currently trying to install dwave-ocean-sdk 4.2.0 in our HPC software stack at Forschungszentrum Juelich. Our environment has Python 3.8.5, numpy 1.19.1 and scipy 1.5.2, but unfortunately the installation fails because penaltymodel-lp 0.1.5 requires numpy >=1.19.4 and scipy >= 1.5.4.

Proposed Solution Would it be possible for you to test and confirm if penaltymodel-lp 0.1.5 works correctly with numpy 1.19.1 and scipy 1.5.2?

Many thanks in advance!

arcondello commented 2 years ago

Hi @ccalaza, looking into it now, will get back to you shortly.

arcondello commented 2 years ago

It appears that those versions do work. I have made a PR to slightly lower the minimum bound, see https://github.com/dwavesystems/penaltymodel/pull/135.

Do you need a release of penaltymodel-lp and/or the dwave-ocean-sdk? Or does knowing that it will work suffice?

ccalaza commented 2 years ago

I just needed confirmation, I'll patch the installation myself. Once again thanks a lot for the super-quick reaction!

arcondello commented 2 years ago

Hi @ccalaza , following up on this. We'd like to raise our minimum supported SciPy version to 1.6.0 to take advantage of the 'highs' linear programming method, see link.

I am curious what the impact on y'all would be? How much of a blocker would this be to using Ocean?

arcondello commented 2 years ago

I was able to find a workaround (see #138 ), but still interested to know for future reference.

ccalaza commented 2 years ago

Hi Alex,

Happy new year and thanks for checking with us, that's very much appreciated!

Unfortunately, as of now we are stuck with Scipy 1.5.2. But the plan is to upgrade to a new environment with Scipy 1. 7.1 in the coming months, hopefully by the beginning of Q2.

X Carlos

arcondello commented 2 years ago

Hi @ccalaza , checking in again! Any chance it's been updated?

ccalaza commented 2 years ago

Hi Alex!

We are currently almost done switching to our software stage "2022" with Python 3.9.6, NumPy 1.21.3 and SciPy 1.7.1. If everything goes according to plan we will stick to these versions at least until around this time next year, when we switch to stage "2023".

Unless of course there's some bug in these packages or their dependencies which forces us to upgrade before then, or for some unlikely reason we need to deploy a "2022b" stage in the 2nd half of 2022. Even then we may not need to upgrade NumPy and Scipy, though.

TLDR, we are extremely likely to stick to NumPy 1.21.3 and SciPy 1.7.1 until at least Q2 2023.

X Carlos