Fix random seed for model repeatability?

[astronobri]

Hi, I’m wondering if there’s a way to set the random seed so that I can reproduce the same REopt API run? I see that the HiGHS optimizer has a random_seed option, but I’m not seeing a straightforward way to implement that from running REopt.

[kurt-verdant]

If 'random_seed' = 0 is the default every time, maybe some explanation other than random_seed variation is responsible for variation in REopt results. There are a couple HiGHS settings that might be set:

In the REopt Julia code the word seed doesn't appear, but other HiGHS settings appear. For example: I wonder if something else is responsible for the variation: I'm just wondering. I don't know the answer!

[adfarth]

You could feasibly get two different results within the optimality tolerance, so I might suggest starting with decreasing the optimality tolerance (mip_rel_gap) setting (as kurt-verdant has suggested). The difference could also potentially be from updates to the underlying REopt.jl package, which would have a corresponding REopt_API "Release". You can view the releases here: https://github.com/NatLabRockies/REopt_API/releases.