Surprisingly slow on fairly small problem about ojalgo HOT 4 CLOSED

There is no trick. What you need is for the linear solver to make use of sparse data structures. Currently it does not (the convex solver does to some extent). Implementing this shouldn't be hard.

The ojAlgo solvers are not comparable to native (commercial) solvers in terms of performance and resource consumption. They're pure Java alternatives for smaller/simpler cases.

If you want solve the linear relaxation you have several options:

1. Create the model that way

2. Call model.relax(boolean);

3. Supply your own Optimisation.Integration using ExpressionsBasedModel.addIntegration(...) If you write your code in the org.ojalgo.optimisation package you can instantiate and delegate to any of the existing implementation.

I guess writing a GLPK integration shouldn't be too hard either, provided they have a Java interface.

With your little trick to solve the relaxed problem execution time goes down from 8s to 6s. This makes me think you never actually make use of the MIP solver.

from ojalgo.

Aha! Sparse vs dense certainly seems like it could account for the difference.

If I relax the model using methods 1 or 2, it actually does not speed up at all. Not sure why...

As for integrations, I in fact started with ojAlgo in an effort to eliminate unpure code (I was using GLPK before via https://www.xypron.de/projects/linopt/), but perhaps it's unavoidable...

Unless you want to keep this ticket open to track sparse linear solving, I'm fine with it being closed. Thank you for your quick response!

from ojalgo.

from ojalgo.

from ojalgo.