LP diving heuristic that chooses fixings w.r.t. the active constraints the variable appear in.
Diving heuristic: Iteratively fixes some fractional variable and resolves the LP-relaxation, thereby simulating a depth-first-search in the tree. Active Constraint Diving chooses a variable based on the active LP rows (equations are counted twice here). The variable score is calculated as a convex combination of the number of constraints the variable appears in, the sum of all positive coefficients and the absolute sum of all negative coefficients. The chosen variable is then rounded to the closest integer. One-level backtracking is applied: If the LP gets infeasible, the last fixings is undone, and the opposite fixing is tried. If this is infeasible, too, the procedure aborts.
Definition in file heur_actconsdiving.h.
Go to the source code of this file.
|SCIP_RETCODE||SCIPincludeHeurActconsdiving (SCIP *scip)|