1. Till Stensitzki
  2. pymls
  3. Issues
Issue #1 resolved


created an issue


I'm using the pymls solver in OpenOpt. The solver has crashed as described in http://forum.openopt.org/viewtopic.php?id=559

The issue is that m_free is 0, yet vector i_free contains one True. Is that OK?

What could I do to help debug the problem? (Unfortunately, I do not have a short reproducer (yet).)

Thanks, Stepan Kasal

Comments (6)

  1. kasal reporter

    I decided to try the following change (near the end of the file):

    if i_free[i_alpha]: i_free[i_alpha]=False m_free-=1

    this helps to maintain the invariant m_free == np.sum(i_free)

    This seems to fix the issue in my case, but I'm just guessing, I don't know if this change is complete, or even if it is correct.

  2. Till Stensitzki repo owner

    Thanks for you feedback, the code is not really well tested, i ll have a look at it later this week. You could also try bvls, which is well tested and also connected to openopt.

  3. kasal reporter

    Thank you very much for your kind answer.

    Unfortunately, I cannot build bvls.f on my platform: Windows64, EPD (based on mingw64 + Intel MKL). f2py builds, but the solver than goes crazy.

    But pymls seems to work now, and I can use converter to nlp if I encounter more problems.

  4. Log in to comment