Parallel optimisation with constraints

[enajx]

Being able to define constraints and doing parallelization over many processes would be very useful. As I understand it, right now only fmin_con allows for constrained optimisation but not parallelization. What could be a good way of implement this?

[nikohansen]

I can't say out of my head. Maybe it is possible to adapt EvalParallel2 accordingly. I am currently implementing a ConstrainedFitness callable class as a simpler interface to constrained optimization which might work out-of-the-box with the EvalParallel2 class. I'll keep this in mind to check.

[Firimo]

Are bounds – as set by the option { 'bounds' : [lb,ub] } – effected by this issue? Will my bounded optimization, barring other polluting effects, run smoothly using EvalParallel2?

[nikohansen]

@dornhe bounds are unaffected and should work as expected.

[Firimo]

thank you for clearing up this small confusion.

[nikohansen]

Unfortunately, parallelization does not work out of the box

    >>> import cma
    >>> def constraints(x):  # define the constraint
    ...     return [x[0] + 1, x[1]]  # shall be <= 0
    >>> cfun = cma.ConstrainedFitnessAL(cma.ff.sphere, constraints,
    ...                                 find_feasible_first=True)
    >>> es = cma.CMAEvolutionStrategy(3 * [1.1], 0.1,
    ...                   {'tolstagnation': 0, 'verbose':9})  # verbosity for doctest only
    >>> with cma.optimization_tools.EvalParallel2(cfun,
    ...                    number_of_processes=2) as eval_all:
    ...    while not es.stop():
    ...        X = es.ask()
    ...        es.tell(X, eval_all(X))
    ...        es.disp()
    ...        cfun.update(es)  # eval_all.fitness_function.update(es)  # 
    >>> x = es.result.xfavorite
    >>> assert sum((x - [-1, 0, 0])**2) < 1e-9, x
    >>> assert es.countevals < 2200, es.countevals
    >>> assert cfun.best_feas.f < 10, str(cfun.best_feas)

---------------------------------------------------------------------------
AssertionError                            Traceback (most recent call last)
/var/folders/0_/q0twpqj16xv3khqtjp8l9c4c0000gn/T/ipykernel_98238/12915452.py in <module>
     14        cfun.update(es)  # eval_all.fitness_function.update(es)  #
     15 x = es.result.xfavorite
---> 16 assert sum((x - [-1, 0, 0])**2) < 1e-9, x
     17 assert es.countevals < 2200, es.countevals
     18 assert cfun.best_feas.f < 10, str(cfun.best_feas)

AssertionError: [-1.08132536e+00  4.60250264e-08  1.55491015e+00]

[nikohansen]

Addressing issue #258 will allow this by design.