ENH: Return last population of scipy.optimize.differential_evolution

Is your feature request related to a problem? Please describe.

Differential evolution is population based algorithm. However, scipy.optimize.differential_evolution returns result as OptimizeResult which gives only the best elite of the population ignoring the rest of its generation.

How can someone preserve information about the last population?

Describe the solution you’d like.

result.x should be the last population in sorted way. Otherwise, OptimizeResult should include a new attribute that memorizes the last population.

Describe alternatives you’ve considered.

Right now I have downloaded the source code to work with the class DifferentialEvolutionSolver directly. So it is possible to make this class available to be imported i.e: from scipy.optimize import DifferentialEvolutionSolver

Additional context (e.g. screenshots, GIFs)

No response

[Abduhu]

Just as a matter of interest, is there a reason you’d like to see the final population (or even intermediate populations)? Note that the final population, if left to converge, is unlikely to have much variation in it. It would be an easy thing to return in the OptimizeResult, but I do wonder about the utility.

In some cases, users want to study the evolution of a specific optimization problem. Returning back the last population is enough since we can always continue evolution from it.

[andyfaff]

It’d require a bit of thought. Each minimizer is going to want to fill out slightly different attributes of an intermediate OptimizeResult. There would have to be a wrapper that could distinguish between the old/new signatures (e.g. state present as a keyword argument), but I’m not sure if (how much) the wrapper would have to be specialised for each method. It might require that each minimizer method has to be altered, which is not trivial.

I think it’d be worth discussing a design before a lot of code is written.