GA probability of selecting parents incorrect

[stefanwintersteller]

Hi,

the probability of selecting parents is calculated by (https://github.com/gkhayes/mlrose/blob/master/mlrose/opt_probs.py eval_mate_probs function):

self.mate_probs = pop_fitness/np.sum(pop_fitness)

This results in high probabilities for bad individuals and low probabilities for good individuals for minimization problems.
Example: Population with 2 individuals, individual_1 fitness -1 and individual_2 fitness -2. The probability of selecting individual_1 (the good one) is 1/3 and probability of selecting individual_2 (the bad one) is 2/3. It is more likely to select individuals with poor fitness.

A quick fix could be the use of the reciprocal value for minimization problems:

if self.maximize == 1:
self.mate_probs = pop_fitness / np.sum(pop_fitness)
else:
self.mate_probs = 1/pop_fitness / np.sum(1/pop_fitness)