Beams2D not detecting all constraint violations

[arthurdrake1]

At the moment, the constraint check for beams2d appears to only detect violations for rmin. Here's the code to reproduce this issue in v0.py:

if __name__ == "__main__":

    from engibench.problems.beams2d.v0 import Beams2D

    problem = Beams2D()

    dataset = problem.dataset

    params_train = dataset["train"].select_columns(problem.conditions_keys)

    design, idx = problem.random_design()

    config = params_train[idx]

    ############################################################
    # Example of three violating configs
    # Expected behavior: 3 violated constraints
    # Actual behavior: only rmin is violated
    print("Original configs:", config)
    new_cfg = {"volfrac": -1.0, "rmin": -1.0, "forcedist": -1.0}
    config.update(new_cfg)
    print("New configs:", config)
    ############################################################

    violated_constraints = problem.check_constraints(design, config)

    print("Violated constraints:", violated_constraints)

Current output on my system:

Original configs: {'volfrac': 0.225, 'rmin': 1.5, 'forcedist': 0.45, 'overhang_constraint': 0}
New configs: {'volfrac': -1.0, 'rmin': -1.0, 'forcedist': -1.0, 'overhang_constraint': 0}
Violated constraints: rmin_bound: Params.rmin: -1.0 ∉ (0, max(nelx, nely)]

Let me know if you are able to reproduce this. Thanks!

[g-braeunlich]

I am able to reproduce if I change

-   print("Violated constraints:", violated_constraints)
+   print("Violated constraints:", violated_constraints.violations)

[ffelten]

@g-braeunlich for what it's worth, I added an str method to violation: https://github.com/IDEALLab/EngiBench/blob/main/engibench/constraint.py#L187

[ffelten]

Just checking the status on this as we might want to launch our constraint checking experiments this week-end.

[g-braeunlich]

This is again something very technical.
It will affect all problems which have from __future__ import annotation. This converts all typehints to strwhich prevents runtime introspection (without parsing strings). Would it be an option to require>= python 3.10? In this case we no longer would need from future import annotation.
Python 3.9 will have EOL in October.

[ffelten]

It is acceptable in my opinion, yes.

[ffelten]

#121

[arthurdrake1]

Thanks. The new output for the test is:

Original configs: {'volfrac': 0.2875, 'rmin': 1.75, 'forcedist': 0.75, 'overhang_constraint': 0}
New configs: {'volfrac': -1.0, 'rmin': -1.0, 'forcedist': -1.0, 'overhang_constraint': 0}
Violated constraints: check: Config.volfrac: -1.0 ∉ [0.0, 1.0]
check: Config.volfrac: -1.0 ∉ [0.1, 0.9]
check: Config.rmin: -1.0 ∉ (0.0, ∞)
check: Config.rmin: -1.0 ∉ [1.0, 10.0]
check: Config.forcedist: -1.0 ∉ [0.0, 1.0]
rmin_bound: Params.rmin: -1.0 ∉ (0, max(nelx, nely)]

This looks a lot better but let me know if anything is still off here. Otherwise we can go ahead and close the issue.