In game theory, a strategy set is normally assumed. This strategy set is fixed and constitutes the only possible actions the player can choose from. In a repeated game, a game is repeated multiple times until convergence or a termination criteria is met. Given a computer memory limitation, 5 strategies out of the millions of possible strategies for each player can only be accomodated. Is there a paper or theory that allows for the replacement of "weak" strategies or does game theory cover this in the model itself.
An example algorithm would be:
1) Randomly select 5 strategies from the strategy set (generated by a population-based search algorithm such as the genetic algorithm).
2) Get the player's payoff for each action using the Method of Successive Averages (MSA).
3) Remove the strategy that gives the lowest payoff and replace it with a new strategy from the strategy set (not currently in memory and possibly generated by the mutation or crossover operator of a genetic algorithm).
4) Check termination criteria. If satisfied, terminate. If not satisfied, Repeat steps 2-4.