Florian Pommerening

Looks like this is not an option. I ran the experiment twice and got different outcomes (both used a very low but slightly different number of trials). One thing I...

I debugged this a bit further. As far as I understand, the intensifier fails to find new trials because the `fails` variable is increased beyond the allowed `retries`. This happens...

Sounds like a great addition. I think a separate command is a good compromise between waiting for custom features to finally make it out of preview mode and hacking something...

Originally reported to Chiliproject ([Link to issue](https://www.chiliproject.org/issues/934))

Hi Ambros, thanks for looking into this. I only changed two unrelated parameters: ```C++ soplex.setIntParam(SoPlex::VERBOSITY, SoPlex::VERBOSITY_ERROR); soplex.setIntParam(SoPlex::SIMPLIFIER, SoPlex::SIMPLIFIER_OFF); ``` The path to `_rationalLUSolver.Clear()` is * `void SoPlexBase::changeLhsReal(int i, const R&...

Maybe the part of the profile below the `clear` method also helps:  The time seems to be spend mostly with memory management (free, alloc) but I'm not 100% certain...

I'm also on holidays right now but will be happy to look into this further next month.

Hi Ambros, I'm back now and started looking into this again. I couldn't use #20 directly because our surrounding code also needs 522b8a18 which seems to be merged in `master`...

The old code talked to SoPlex via OSI, where the new code directly uses the C++ interface. The use through OSI is a bit of a blackbox to me, so...

I enabled the log output with both versions and uploaded the results here. In this setting, there is an LP with a bunch of "permanent" constraints that is solved over...