MBM/Hull on InfiniteGDP disjunctions where constraints carry different infinite parameters (or none)

[dnguyen227]

While sampling M values for InfGDP problems, the possibility of finite disjunction constraints was not considered. Small fix applied here.

using DisjunctiveProgramming, InfiniteOpt, HiGHS
model = InfiniteGDPModel(HiGHS.Optimizer)
@infinite_parameter(model, t in [0, 1], num_supports = 11)
@variable(model, 0 <= x <= 10, Infinite(t))   # operating variable
@variable(model, 0 <= d <= 10)                # design variable
@variable(model, Y[1:2], Logical)

@constraint(model, x >= 1 - t)
@constraint(model, d >= 3, Disjunct(Y[1]))    # finite: no infinite variable
@constraint(model, d >= 5, Disjunct(Y[2]))
@disjunction(model, Y)
@objective(model, Min, d + ∫(x, t))

optimize!(model, gdp_method = MBM(HiGHS.Optimizer))

Upon running you'd get

Unrecognized sampler value ExhaustiveSampler() for MBM on an infinite model

ALSO ADDED

• Variable/constraint carries the indicator's parameters (x(t) or x(t, xi) under Y(t)): unchanged
• Finite indicator, infinite constraints (x(xi) <= xi under finite Y): unchanged other than fixing error.
• Variable has fewer parameters (finite d under Y(t), z(t) under Y(t, xi)): was wrong for MBM, MBM-GP and Hull; now the copy contains the indicator's parameters.
• Parameters partly or not at all shared (v(xi) under Y(t)): was wrong; now the copy contains the union.

The workflow is per variable, so d + x(t) <= 2 - 2t under Y(t) copies d and leaves x(t) alone.

using DisjunctiveProgramming, InfiniteOpt, HiGHS

model = InfiniteGDPModel(HiGHS.Optimizer)
@infinite_parameter(model, t ∈ [0, 1], supports = [0.0, 0.25, 0.5, 0.75, 1.0])
@variable(model, 0 <= d <= 3)                    # finite
@variable(model, Y[1:2], InfiniteLogical(t))     # indicator over t
@constraint(model, d <= 2 - 2t, Disjunct(Y[1]))
@constraint(model, d <= 1 + 2t, Disjunct(Y[2]))
@disjunction(model, Y)
@objective(model, Max, d)

optimize!(model, gdp_method = BigM(10.0)); objective_value(model)          # 1.5
optimize!(model, gdp_method = MBM(HiGHS.Optimizer)); objective_value(model)  # was 1.0
optimize!(model, gdp_method = Hull()); objective_value(model)              # was 1.0

At each t the better disjunct gives d <= max(2 - 2t, 1 + 2t), so the optimum is min_t max(...) = 1.5.

No work needed for MBM-GP because it samples based on sub problems made from the MBM code.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 99.69%. Comparing base (eca397e) to head (7ca2445).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #146   +/-   ##
=======================================
  Coverage   99.69%   99.69%           
=======================================
  Files          21       21           
  Lines        2265     2284   +19     
=======================================
+ Hits         2258     2277   +19     
  Misses          7        7           

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

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[dnguyen227]

@pulsipher ready for review

[pulsipher]

Please merge with main to get to resolve conflicts

[dnguyen227]

@pulsipher ready for another look.

Not sure why I can't request review on this PR specifically (the button just isn't there).

[pulsipher]

Looks reasonable, just want to make the tests a little more robust.

[pulsipher]

For all the tests, can we check more than the objective value to make sure that it is reformulating correctly, not that it just runs.