This question arises often when first learning about Gurobi's Seed parameter and how it can affect the solution path. The answer is that there is no best choice, at least not one that can be known in advance. Wanting to know the best seed is like wanting to know the best lottery numbers. You will certainly know this after the lottery takes place, but you cannot know beforehand as both are related to random processes that cannot be predicted. Likewise, there may be one value of Seed that will work best for your model but it cannot be known without testing all possible values first (and there are two billion such values).
But let's assume that, through experimentation, the best value of Seed for your model had been discovered - the value which leads to the fastest solve time on your production machine, for example. The moment there is any change to the model (or the data, or the parameters, or the machine it is run on, or the Gurobi version) then we will no longer know anything about how the solver will perform with this value of Seed. The pursuit of the best value of Seed is not one that makes sense.
Comments
0 comments
Article is closed for comments.