COBYLA minimizer keeps searching outside the bounds/constraints

Question

I have troubles using shgo.shgo (I would say, just the same as scipy.optimize.shgo except for an annoying bug in the latter).

Now, my optimization problem behaves badly when using SLSQP as the local minimizer, and I found various suggestions for switching to COBYLA. It looks like there are some known issues in passing bounds to COBYLA, and I am addressing them as specified here and here.

Still, the algorithm keeps searching outside some parameters' boundaries, causing the code to stop running. I could handle this behavior "manually" by imposing an np.inf error whenever the optimizer tries to use out-of-boundaries values, but I am not sure this is the clever way to do things.

Currently, I am running something like this:

from shgo import shgo

parameter_bounds = [ (1, 10), (1, 10), (1, 10), (1, 10), (0.05, 1.1), (0.1, 1), (0.01421, 0.01431), (0.3, 1.7),]

cons = []
for factor in range(len(parbounds)):
   lower, upper = parbounds[factor]
   l = {"type": "ineq", "fun": lambda x, lb=lower, i=factor: x[i] - lb}
   u = {"type": "ineq", "fun": lambda x, ub=upper, i=factor: ub - x[i]}
   cons.append(l)
   cons.append(u)

result = shgo(
func=objfun,
args=(data_moments, W, rng),
bounds=parameter_bounds, # SHGO *needs* bounds
minimizer_kwargs={'method': 'COBYLA', 'constraints': cons},
options={'disp': True},
)

Any suggestion would be greatly appreciated.

COBYLA minimizer keeps searching outside the bounds/constraints

0 Answers0