contrib.piecewise: Adding nonlinear-to-piecewise-linear transformation

[emma58]

Fixes # .

Summary/Motivation:

This PR adds a transformation that takes a MINLP model and automatically generated a piecewise-linear approximation of the model that can then be transformed to MILP using the other transformations in contrib.piecewise.

Changes proposed in this PR:

• Adds contrib.piecewise.nonlinear_to_pwl transformation
• Adds a lot of tests
• Sneaks in a fix to correctly handle LogicalConstraints in the transformations from piecewise linear to GDP/MIP

Legal Acknowledgement

By contributing to this software project, I have read the contribution guide and agree to the following terms and conditions for my contribution:

1. I agree my contributions are submitted under the BSD license.
2. I represent I am authorized to make the contributions and grant the license. If my employer has rights to intellectual property that includes these contributions, I represent that I have received permission to make contributions and grant the required license on behalf of that employer.

[codecov]

Codecov Report

Attention: Patch coverage is 97.59760% with 8 lines in your changes missing coverage. Please review.

Project coverage is 88.60%. Comparing base (29f8ede) to head (b3429e1).
Report is 325 commits behind head on main.

Files	Patch %	Lines
...mo/contrib/piecewise/transform/nonlinear_to_pwl.py	97.83%	7 Missing ⚠️
...omo/contrib/piecewise/piecewise_linear_function.py	85.71%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #3333      +/-   ##
==========================================
+ Coverage   88.55%   88.60%   +0.04%     
==========================================
  Files         877      878       +1     
  Lines       99262    99591     +329     
==========================================
+ Hits        87906    88238     +332     
+ Misses      11356    11353       -3     

Flag	Coverage Δ
linux	86.13% <97.59%> (+0.03%)	⬆️
osx	75.83% <97.59%> (+0.07%)	⬆️
other	86.64% <97.59%> (+0.04%)	⬆️
win	83.98% <97.59%> (+0.05%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[emma58]

Oops, I think this is failing coverage because we don't have lineartree installed anywhere for testing...

[emma58]

If this passes tests, it is ready for review!

[emma58]

@mrmundt @jsiirola Are the pypy failures my fault? They seem consistent.

[jsiirola]

@mrmundt @jsiirola Are the pypy failures my fault? They seem consistent.

Yes - I think so. Because you added linear-tree, that triggered pip to install scipy, which is known to have issues under pypy (which is why we exclude it).

I think the resolution is to add linear-tree to PYPY_EXCLUDE: scipy numdifftools seaborn statsmodels in .github/workflows/test_pr_and_main.yml and .github/workflows/test_branches.yml. Note that you need to update .github/workflows/test_branches.yml to add linear-tree to PYPI_ONLY , too...

[emma58]

Oooof, now I broke everything?

[blnicho]

@emma58 GitHub was having issues yesterday so I just kicked off the GHA tests again to see if that fixes things.

[jsiirola]

Overall, this looks pretty good. One thing that should be changed, though (defer the construction of the visitor, or move it to an instance attribute)

[jsiirola]

This gets instantiated upon module import (which will happen as part of pyomo.environ. It would be better to put this in a getter:

def visitor():
    if visitor._instance is None:
        visitor._instance = QuadraticRepnVisitor(
            subexpression_cache={}, var_map={}, var_order={}, sorter=None
        )
    return vistor._instance

[emma58]

I put it on the transformation instance.