You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
A new primal heuristic was recently proposed in The granularity concept in mixed-integer optimization. The idea is to define a subset of the feasible set so it is are guaranteed that at-least one rounding exists that is in the feasible set, optimize this relaxation and then round once to get to a feasible solution faster. This is trivial for Linear constraints but doing it for non-linear constraints might be harder or require an iterative approach, (i haven't done the math). I will also suggest this for Alpine.
The text was updated successfully, but these errors were encountered:
A new primal heuristic was recently proposed in The granularity concept in mixed-integer optimization. The idea is to define a subset of the feasible set so it is are guaranteed that at-least one rounding exists that is in the feasible set, optimize this relaxation and then round once to get to a feasible solution faster. This is trivial for Linear constraints but doing it for non-linear constraints might be harder or require an iterative approach, (i haven't done the math). I will also suggest this for Alpine.
The text was updated successfully, but these errors were encountered: