Closed rezabayani closed 1 year ago
The mps files to reproduce the results are attached here: mps_format.zip
Update: I think I have found the cause of this behavior. It appears that SCIP gets rid of some variables in the presolve stage (although I tried disabling presolve actions through setting maxrestarts, maxrounds, abortfac, and restartminred to 0). Therefore, I do not belive this is an issue with Ecole and rather how SCIP's root node is formed.
Hi @rezabayani could you please elaborate how you disable the presolve stage eventually? Did you use specific scip_parameters in the env.reset() function? Thank you!
[Reposting this from discussions as I believe it is a bug rather than a feature] Hi, My Ecole (v8) is having trouble reading observations from the optimization model stored as mps file. Basically, it returns None observation and True done flag when I reset the environment. I did the same approach with Ecole v7 and it was okay back then. Interestingly, when I tried feeding other mps files found online, it is okay with some of them and not with others. I have attached 2 images. Number 1 is when Ecole is acting as intended, using an mps model found online. Number 2 is when Ecole is acting funny, with an mps file I generated myself and was okay with Ecole v7.
Originally posted by @rezabayani in https://github.com/ds4dm/ecole/discussions/353