Hi,

I encountered a case where, if I'm using roulette wheel selection or stochastic universal selection and if the fitness sum is 0, there will be a division by 0. I know that probably it's very little probability to encounter such a case, but my suggestion would be to instead of having probs = fitness / fitness_sum
to have:

if fitness_sum !=0:
   probs = fitness / fitness_sum
else:
   probs = 0

or instead of probs=0, to divide fitness by a very large number.

Best Regards!