// If the solution space is very large, this method will not return.
// The most likely result is that it will exceed memory capacity
// and then fail. As a result, we shouldn't use this method in
// any production optimization.
public IEnumerable <ProductionTarget> GetFeasibleTargets(double claySupply, double glazeSupply)
{
List <ProductionTarget> targets = new List <ProductionTarget>();
// Setup the Gurobi environment & Model
GRBEnv env = new GRBEnv();
GRBModel model = new GRBModel(env);
// Setup the decision variables
GRBVar xS = CreateSmallVasesVariable(claySupply, model);
GRBVar xL = CreateLargeVasesVariable(glazeSupply, model);
model.Update();
// Create Constraints
CreateConstraints(model, xS, xL, claySupply, glazeSupply);
// Find the greatest number of small vases we can make
var maxSmall = System.Math.Min(claySupply, glazeSupply);
// Find the greatest number of large vases we can make
var maxLarge = System.Math.Min(claySupply / 4.0, glazeSupply / 2.0);
// Find all feasible combinations of small and large vases
// Note: There are probably several better ways of doing this
// that are more efficient and organic. For example, we could make
// a tree that represents all of the possible decisions and let the
// optimizer find the solutions from within that tree.
var results = new List <ProductionTarget>();
for (int nSmall = 0; nSmall <= maxSmall; nSmall++)
{
for (int nLarge = 0; nLarge <= maxLarge; nLarge++)
{
// Force the solution to the target set of values
var c1 = model.AddConstr(xS == nSmall, $"xS_Equals_{nSmall}");
var c2 = model.AddConstr(xL == nLarge, $"xL_Equals_{nLarge}");
model.Update();
// See if the solution is feasible with those values
model.Optimize();
if (model.IsFeasible())
{
results.Add(new ProductionTarget()
{
Small = nSmall, Large = nLarge
});
}
model.Remove(c1);
model.Remove(c2);
model.Update();
}
}
return(results);
}
