search

ds4dm / ecole

Extensible Combinatorial Optimization Learning Environments
https://www.ecole.ai
BSD 3-Clause "New" or "Revised" License
321 stars 68 forks source link

Strong branching score of -1e+20 for candidate variables #348

Open SelinBayramoglu opened 2 years ago

SelinBayramoglu commented 2 years ago

Describe the bug

Hi,

I use the the observation function ecole.observation.StrongBranchingScores to extract strong branching scores of candidate variables in an application. I collect scores for only variables with a fractional solution, i.e. no pseudo candidates. I see that some candidates get a score of -1e+20. According to the SCIP source code branch_vanillafullstrong.c, all candidate scores are initialized at this value and later obtain a valid score after strong branching. By definition, a branching score should be nonnegative, so I am confused as to why this happens. If this is the result of an LP error during strong branching computation, could you suggest a method to debug this?

Thank you.

Setting

To Reproduce

Expected behavior

I would expect all strong branching scores to be >= 0 and I would expect a message if no scores could be extracted for a variable due to some error.

Additional context

Yige-Lei commented 4 months ago

Have you solved this problem? I've encountered similar problem. Some of my strong branching scores of candidate variables are "nan". I'm confused too. TT