RM6.py

class Player:
    # ....
    def update_strategy(self):
        """
        Set the preference (strategy) of choosing an action to be proportional to positive regrets. e.g, a strategy that prefers PAPER can be [0.2, 0.6, 0.2]
        """
        self.strategy = np.copy(self.regret_sum)
        self.strategy[self.strategy < 0] = 0  # reset negative regrets to zero

        summation = sum(self.strategy)
        if summation > 0:
            # normalise
            self.strategy /= summation
        else:
            # uniform distribution to reduce exploitability
            self.strategy = np.repeat(1 / RPS.n_actions, RPS.n_actions)

        self.strategy_sum += self.strategy

    def learn_avg_strategy(self):
        # averaged strategy converges to Nash Equilibrium
        summation = sum(self.strategy_sum)
        if summation > 0:
            self.avg_strategy = self.strategy_sum / summation
        else:
            self.avg_strategy = np.repeat(1/RPS.n_actions, RPS.n_actions)