I'm developing an AI to play a card game with a genetic algorithm. Initially, I will evaluate it against a player that plays randomly, so there will naturally be a lot of variance in the results. I will take the mean score from X games as that agent's fitness. The actual playing of the game dominates the time to evaluate the actual genetic algorithm.
My question is: should I go for a low X, e.g. 10, so I would be able to move through generations quite fast but the fitness function would be quite inaccurate? Alternatively, I could go for a high X e.g. 100 and would move very slowly but with a more accurate function.