However, if the constraint was continuous, it would solve Much faster, and return a relaxed solution. Since we are minimizing the sum of the $y$ variables, we will get a solution that may contain extra reactions, but not too many. This set of extra reactions can be used as the universal matrix, which will significantly reduce the runtime of the full gapfill.
Currently on line 52 of gapfill.py, we declare a binary constraint for $y$: https://github.com/cdanielmachado/carveme/blob/master/carveme/reconstruction/gapfilling.py#L52
However, if the constraint was continuous, it would solve Much faster, and return a relaxed solution. Since we are minimizing the sum of the $y$ variables, we will get a solution that may contain extra reactions, but not too many. This set of extra reactions can be used as the universal matrix, which will significantly reduce the runtime of the full gapfill.