I am trying to use the scipy.optimize package to optimize a discrete optimization problem (global optimization). Acc to the doc, simulated annealing implemented in scipy.optimize.anneal should be a good choice for the same. But I am not sure how to force the optimizer to search only integer values of the search space. Can someone help?
An illustrative example:
f(x1,x2) = (1-0.4*x1)^2 + 100*(0.6*x2 -0.4*x1^2)^2
where, $x1, x2 \in I$
I've checked scipy.optimize.anneal, and I can't see a way to use discrete values. The way to implement it yourself, is to create a custom "move" function, but the way you have to specify the schedule (by a string) prevents you from doing so.
I think it is a big mistake, if you could just pass a custom schedule class as the parameter, you could customize it for using discrete variables and many more things.
The solution I found is to use this other implementation instead: https://github.com/perrygeo/python-simulated-annealing
Because you have to provide the function which modifies the state, you have control on what values it can have, or if they are discrete or continuous.
Hope it helps.
来源：https://stackoverflow.com/questions/16844423/discrete-optimzation-in-python