Describe the bug
There is no NaN/infinity verification in the SBX Crossover implementation. This makes it possible for the crossover to assign NaN/infinity values to genes.
An example of how this seems to happen is when the population is set with genes that don't respect their bounds in NSGA-II (and maybe other genetic algorithm implementation). There should be a verification beforehand that all those genes respect the bounds set.
Expected behavior
For NSGA-II
- An error message saying that the initial population have some individuals not respecting bounds OR those said individuals are bounded before executing the evolution.
For sbx crossover impl
- A bounding step at the end of the crossover, making sure neither C1 or C2 are NaN/infinity values and then assigned to genes
Environment (please complete the following information):
- OS: Ubuntu 20.04
- Installation method: source
- Version: branch master
Additional context
I have not verified if other genetic algorithms have this problem nor if other crossover operators can assign NaN/infinity values. But this should be confirmed into the same pull request in my opinion.
Describe the bug
There is no NaN/infinity verification in the SBX Crossover implementation. This makes it possible for the crossover to assign NaN/infinity values to genes.
An example of how this seems to happen is when the population is set with genes that don't respect their bounds in NSGA-II (and maybe other genetic algorithm implementation). There should be a verification beforehand that all those genes respect the bounds set.
Expected behavior
For NSGA-II
For sbx crossover impl
Environment (please complete the following information):
Additional context
I have not verified if other genetic algorithms have this problem nor if other crossover operators can assign NaN/infinity values. But this should be confirmed into the same pull request in my opinion.