Can MultiStart be vectorized ?

Can MultiStart be modified to be vectorized like GeneticAlgorithm?

Answer 1

[Walter Roberson]

No, but you can add the 'UseParallel', true option, which it will pass on the the solver. Some solvers can make use of that; for example, fmincon can do the finite differencing in parallel (but not much else)

https://www.mathworks.com/help/optim/examples/minimizing-an-expensive-optimization-problem-using-parallel-computing-toolbox.html

[mahmoud tarek]

Thank you for your answer

[Walter Roberson]

I see that you unaccepted my answer on the grounds that it does not work. However I correctly identified the only available multiple computation option, https://www.mathworks.com/help/gads/example-parallel-multistart.html and I also correctly identified the limitations that the effects are determined by the individual solvers which will not necessarily do nearly as much as as one might hope with the option.

Could you explain further what more you need from the answer?

By the way, Mathworks staff are on record as indicating that parallel processing was partly written for ga() but was not completed and does not currently do anything. I was a bit surprised to see that as it looked pretty complete when I read through the code.

[mahmoud tarek]

I wanted to know if there is any possible way to make multistart vectorized. A change in multistart code itself.is that possible?

[Walter Roberson]

The running of the solvers is done by constructing a number of starting points, and then calling fmultistart() passing in the problem description and cell arrays of starting points.

Unfortunately, fmultistart is toolbox/globaloptim/globaloptim/private/fmultistart.p -- that is, code is not available for it, so it cannot be modified.

Answer 2

[mahmoud tarek]

Thank you