Can't beat abbey Rock Paper Scissors project

marius-mm · February 20, 2021, 8:53pm

Tell us what’s happening:
I’m using a markov model.
I just take the state with the highes probability and return the counter as my guess.
With a memory of 800previous plays i can easily beat 3/4.
Any idea how i can adapt my model to beat her?
we even out around 40-50% no matter how long the memory.

Challenge: Rock Paper Scissors

Link to the challenge:

marius-mm · February 20, 2021, 10:00pm

github.com

marius-mm/freeCodeCampProjects/blob/main/Machine Learning with Python/RockPaperScissors/RPS.py

# The example function below keeps track of the opponent's history and plays whatever the opponent played two plays ago. It is not a very good player so you will need to change the code to pass the challenge.

import random
import mchmm as mc
import numpy as np

winDict = {"R": "P", "S": "R", "P": "S"}

strategy = 1


def player(prev_play, opponent_history=[]):
    global strategy
    # firstCall
    if len(opponent_history) <= 0:
        opponent_history.append("R")
        opponent_history.append("S")
    if len(prev_play) <= 0:
        prev_play = "P"
    # /firstCall

This file has been truncated. show original

jeremy.a.gray · February 20, 2021, 11:09pm

Abbey is a Markov chain player. You can check the source to see her implementation and chain length. I’ve never used mchmm before, but I think to use it for this project, you will need to read starting at the Calculating Nth order transition probability matrix: section and onward because this is really a Markov chain problem. I don’t think memory will be an issue here since for the 800 moves you are using, that can only affect the last 200 moves of the trial.

There are at least two ways I’ve found to beat Abbey. One, you can use a long enough Markov chain (longer than hers) and win (enough to pass the test, but not convincingly). Two, you can implement a specific anti-Abbey strategy and win at almost 90% but to do this you’ll have to understand exactly how she is playing.

There are undoubtedly other ways, but I haven’t bothered to research or implement some of the more advanced ones I’ve seen mentioned.

marius-mm · February 20, 2021, 11:27pm

i meant maxlength of 800 sorry,
so at game 801 I shift my List by deleting the first value in my memory.

jeremy.a.gray · February 21, 2021, 12:27am

I understand. But since a trial is only a 1000 moves long, you’re not forgetting anything until there are only 200 moves left. If you’re not resetting after every opponent, then that means you’ll be predicting moves off the previous opponent’s move for a large part of the new trial. Either way, it really won’t help, and in the latter case, could hurt your ability to win.

As I said, she is beatable with a Markov chain algorithm of sufficient length, but without familiarity with mchmm, I can’t tell from your code what chain length you are using and if it would beat Abbey, and hence, I can’t test it with my Markov chain algorithm with which I can set the chain length. So I suggest you try to determine how long a chain Abbey is using and then experiment with different chain lengths until you can find one that can defeat her.

marius-mm · February 21, 2021, 9:40pm

Ah didn‘t think about that the chain won‘t get reset after each enemy.
But I mean i can‘t beat abbey with every setting
do you mean with chain length like different states? RS RR RP SR … instead of S R P ?

jeremy.a.gray · February 22, 2021, 12:54am

That’s exactly what I mean. That’s also what Abbey is doing, so you will have to use a longer chain than she does or use her chain against her to win.