@floop for _λ in Iterators.product(1:5,1:5,1:5)
let λ = [ 1, _λ... ] # TODO it would be nice if we could have one per thread
λ .= λ ./ sqrt( norm(λ) )
m = f(λ)
@reduce() do (_min = Inf; m), (λmin = zeros(4); λ)
if gm < _min
_min = gm
λmin .= λ
end
end
end
end
where instead of the let λ I have one λ per thread to minimize garbage collection when the product-space size gets really large.
Is this possible with the implementation of the multithreading?
Hi, I'd like to implement something like
where instead of the
let λI have one λ per thread to minimize garbage collection when the product-space size gets really large.Is this possible with the implementation of the multithreading?