From my understanding of genetic algorithms, the mating pool size comes from a randomly selected sample of the population (Assuming we are using tournament selection).
We then sort the mating pool by each of the individuals' fitness, and select an elite size (number of best individuals that moves directly into the children population).
Using the mating pool, we now begin to cross breed randomly selected individuals from the mating pool using 1 or 2 point crossover. We repeat this process until we have enough children for the next population.

Finally, we set a chance for mutation for each individual in the children and repeat the process for n many generations.

I'm having trouble determining the best combination of population size, elite size, and mating pool size to solve the problem. 
Would it be better to select a mating pool size around 90% the size of the population or lower? Would it be better to generally just have larger elite/mating pool/population sizes in general?

How long should it take for a GA for, let's say.. a 5x5 Nurikabe take to solve?

After attempting some GA problems, I'm starting to feel like calculating Fitness for some problems was like writing heuristics for Pac-man. 
Currently, calculating Fitness for Nurikabe has been a big issue for me - and I'm starting to believe that determining Fitness is one of the biggest challenges for GA.

Thanks & Happy Holidays!

To ensure diversity, the Population selects its next generation by selecting a certain percentage the usual way, e.g., tournament selection, and the rest as random new individuals. You can specify what percentage will be random. See one_max_2D.  If you make select_pct_random=0.9 (as it's currently set), evolution almost always find a solution. If you set it to  select_pct_random=0.1 it often fails to find a solution.