[Suggestion] Custom duplicate detection #2
Description
Hello, thanks for your work on this library!
In the problem I need to solve, some parts of individuals (the last 2 values for example) can be permuted and still represent the same solution, so [1, 2, 3] and [1, 3, 2] have to be treated as duplicates.
Currently, GeneticalAlgorithm2's without_dup() is an inner function of a method, so it can't be changed easily as it would have if it could be inherited.
My workaround for now is copy-pasting the whole run() function into a new subclass of GeneticAlgorithm2, which is far from ideal, then redefine without_dup as:
def without_dup(pop: np.ndarray, scores: np.ndarray):
# individuals are duplicates if the only differences are permutations within the first 6 or the next 2 values
_, index_of_dups = np.unique(np.concatenate((
np.sort(pop[:, :6]),
np.sort(pop[:, 6:8]),
pop[:, 8:]),
axis=1), axis=0, return_index=True)
return pop[index_of_dups], scores[index_of_dups], scores.size - index_of_dups.size
or
def without_dup(pop: np.ndarray, scores: np.ndarray):
# it's almost impossible in my problem for 2 different solutions to have the same score, so we get the indices of the unique scores instead
_ , index_of_dups = np.unique(scores, return_index=True)
return pop[index_of_dups], scores[index_of_dups], scores.size - index_of_dups.size
It would be cleaner to have without_dup() be a class function (that can be redefined by inheritance), or an optional parameter for run().