Consider a game like Pig (https://en.wikipedia.org/wiki/Pig_(dice_game)), but with a few additions: namely functions of both player's score and turn number that have unique impacts on scoring.
What machine learning model should I use to try and get the optimal number of dice roles per turn (say number of dice roles are bounded between 1 and 10)?
I was reading this tutorial: https://towardsdatascience.com/playing-cards-with-reinforcement-learning-1-3-c2dbabcf1df0, and they suggested reinforcement learning with a Q value function. I don't know how this work though, because turn number isn't bounded, but also needs to be a parameter to the Q value function. Multiplying the range of all parameters suggest this Q value function needs 2,000,000 states. Is this too many? - I have no idea how to judge this.
Is there a better model I should use to try and solve this problem which at its core takes the parameters of (my_score, opponent_score, turn_number) and should return a number 0 - 10 representing how many dice to roll.
