I highly recommend clicking through to the second post in the series. It discusses optimizing the models -- an area where I've had some success and a lot of fa

I highly recommend clicking through to the second post in the series. It discusses optimizing the models -- an area where I've had some success and a lot of failure. There are some really complicated tradeoffs to be managed, and it can be incredibly hard to measure whether you've made a genuine optimization or whether you've accidentally tipped the solver into a happy path for one particular instance of the problem you're trying to solve. Some solvers use PRNGs, and you have to be careful to try enough seeds to get a clear picture of the solver's performance. This can be challenging when the time to solution is exponentially distributed with a mean of several hours. The most important thing OP does, in my opinion, is to solve a smaller (but not otherwise simpler) version of the problem. This is crucial if you're going to find a tractable representation -- or rather, if you're going to eliminate completely bogus representations of the problem.