RAM issue when running large problems about metaheuristics.jl HOT 5 CLOSED

I already implemented that feature but is still at develop branch
It will be merged to main branch ASAP.

from metaheuristics.jl.

You can do it as follows:

f, bounds, optimums = Metaheuristics.TestProblems.get_problem(:sphere)
D = size(bounds,2)

# known solution
x_known = 0.6ones(D)

# random solutions (uniform distribution)
X = [ bounds[1,:] + rand(D).* ( bounds[2,:] -  bounds[1,:]) for i in 1:19  ]

push!(X, x_known) # save an interest solution

# generate the population with 19+1 solutions
population = [ Metaheuristics.create_child(x, f(x)) for x in X ]

# prior state
prev_status = State(Metaheuristics.get_best(population), population)

# say to ECA that you have generated a population
method = ECA(N = length(population))
method.status = prev_status

# optimize
optimize(f, bounds, method)

If you also know x_known and fx = f(x_known), then instead of pushing x into X,
you may do push!(population, Metaheuristics.create_child(x_known, fx)).

from metaheuristics.jl.

By default the population size in ECA is N = K * D where K = 7 and D is the dimension number (parameters), maybe you should change the value N (say N=100). Also limit the number of function evaluations.

Example:

options = Options(f_calls_limit=10_000)
method = ECA(N = 100, options = options)
optimize(f, bounds, method)

Note that ECA and the other metaheuristics implemented in this Package work well for up to 100 parameters. Probably later some methods for large-scale problems (>100 parameters) will be implemented.

from metaheuristics.jl.

Thanks now it worked!, I limited the population size. Do you now if an initial population can be given as hot start solution?.

from metaheuristics.jl.

I already implemented that feature but is still at develop branch
It will be merged to main branch ASAP.

How would you declare an starting point from the beginning?, since the example provided assume that you have a previous population and a result from a previous optimization process..

from metaheuristics.jl.