In attempting to use BayesSearchCV from the skopt library, I have two feature distributions that are dependent on one another, such that par_B must be > par_A
Is there an efficient way to do this within a Real search space?
I have tried the following,
par_A_search_space = Real(0.01, 0.9)
BayesSearchCV(
estimator=pipeline,
search_spaces={
'pipeline__par_A': par_A_search_space,
'pipeline__par_B': Real(par_A_search_space, 1)
}
)
which fails with the following traceback:
TypeError: '<=' not supported between instances of 'float' and 'Real'
Changing the upper bound float to a Real search space just returns the same error, that <= is not supported between 'Real' and 'Real'
Note: This question is similar, but on explicit, categorical distributions rather than continous