cog-imperial / entmoot

Multiobjective black-box optimization using gradient-boosted trees
https://entmoot.readthedocs.io/
BSD 3-Clause "New" or "Revised" License
56 stars 12 forks source link

Constraint Class #24

Closed TobyBoyne closed 1 year ago

TobyBoyne commented 1 year ago

This PR implements a Constraint class that enables a more user friendly handling of constraints, as well as a delay between constraint definition and application to a model.

This is currently a draft pending a little more documentation and a few test cases. A jupyter notebook constraint_classes.ipynb is provided to demonstrate the functionality.

Let me know what you think! The aim is for useability, so let me know how it reads. My biggest issue at the moment is that creating the constraint requires the ConcreteModel (for access to the variables) and ProblemConfig.feat_list (for the names of the variables). This feels slightly clunky, but I think is necessary for how ENTMOOT is written.

Also this currently only supports Pyomo, not Gurobi. If this works well for Pyomo then I'm happy to have a look at transferring over!

Thanks, Toby