Closed PavelVesely closed 7 months ago
Good idea. Depending on how large the ILP is we could also provide some standard approx. alg. for Set Cover like greedy/primal-dual alg. But if the ILP is small, this is probably not relevant.
If we find, we want the approx ilp solver, we can reopen this issue.
I think it's not worth trying to implement a good approximation algorithm for set cover, at least not now. In the end, it may require more resources than a good ILP solver on the data that we have now. (Actually, we tried just glpk and not something better, right?)
For very large datasets (e.g., the human genome), solving the ILP to minize the number of runs of 1s may be too time or memory consuming. I suggest to provide a heuristic version that only runs
HeuristicPreSolve
and then sets1
s in all other intervals, on which we now apply the ILP. We can call itrunsapprox
or something like that