I just completed the Python Rock Paper Scissors Machine Learning project… but I did not find my solution satisfying. I thought there was a way for a certain piece of code to look back at the opponents history and then through pattern recognition clear the challenges.
I was looking at a Markov Chain approach and Bayesian Statistics, however… the hint stated that I needed to change strategies and therefore I hardcoded a way to counter the strategies given no matter the order.
Was hardcoding the way to solve this challenge? Or did I miss something?
I did create another list of my own history , and another for winrate to switch strategies… However, that is not a machine learning a pattern on its own.
I found this code has high win-rate, created by eisbilir
# 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 numpy as np
ideal_response = {"P": "S", "R": "P", "S": "R"}
my_moves = ["R"]
opponent_history = []
strategy = [0, 0, 0, 0]
opponent_guess = ["", "", "", ""]
strategy_guess = ["", "", "", ""]
opponent_play_order = {}
my_play_order = {}
def player(prev_play):
if prev_play in ["R", "P", "S"]:
opponent_history.append(prev_play)
for i in range(0, 4):
if opponent_guess[i] == prev_play:
strategy[i] += 1
else:
reset()
my_last_ten = my_moves[-10:]
if len(my_last_ten) > 0:
my_most_frequent_move = max(set(my_last_ten), key=my_last_ten.count)
opponent_guess[0] = ideal_response[my_most_frequent_move]
strategy_guess[0] = ideal_response[opponent_guess[0]]
if len(my_moves) > 0:
my_last_play = my_moves[-1]
opponent_guess[1] = ideal_response[my_last_play]
strategy_guess[1] = ideal_response[opponent_guess[1]]
if len(opponent_history) >= 3:
opponent_guess[2] = predict_move(opponent_history, 3, opponent_play_order)
strategy_guess[2] = ideal_response[opponent_guess[2]]
if len(my_moves) >= 2:
opponent_guess[3] = ideal_response[predict_move(my_moves, 2, my_play_order)]
strategy_guess[3] = ideal_response[opponent_guess[3]]
best_strategy = np.argmax(strategy)
guess = strategy_guess[best_strategy]
if guess == "":
guess = "S"
my_moves.append(guess)
return guess
def predict_move(history, n, play_order):
if "".join(history[-n:]) in play_order.keys():
play_order["".join(history[-n:])] += 1
else:
play_order["".join(history[-n:])] = 1
possible = ["".join(history[-(n - 1) :]) + k for k in ["R", "P", "S"]]
for pm in possible:
if not pm in play_order.keys():
play_order[pm] = 0
predict = max(possible, key=lambda key: play_order[key])
return predict[-1]
def reset():
global my_moves, opponent_history, strategy, opponent_guess, strategy_guess, opponent_play_order, my_play_order
my_moves = ["R"]
opponent_history.clear()
strategy = [0, 0, 0, 0]
opponent_guess = ["", "", "", ""]
strategy_guess = ["", "", "", ""]
opponent_play_order = {}
my_play_order = {}
It is great that you solved the challenge, but instead of posting your full working solution, it is best to stay focused on answering the original poster’s question(s) and help guide them with hints and suggestions to solve their own issues with the challenge.
We are trying to cut back on the number of spoiler solutions found on the forum and instead focus on helping other campers with their questions and definitely not posting full working solutions.
You can post solutions that invite discussion (like asking how the solution works, or asking about certain parts of the solution). But please don’t just post your solution for the sake of sharing it.
If you post a full passing solution to a challenge and have questions about it, please surround it with [spoiler] and [/spoiler] tags on the line above and below your solution code.