def reset(self):
# Reset the state of the environment to an initial state
self.game = Game(self.board_width, self.board_height)
self.steps = 0
if (self.game.current_player==1): # so ist immer der RL Agent als erstes im step dran und kann keinen illegalen zug machen wenn er als action das selbe feld wählt wie der Gegner
self.game.play(find_best_move(minmax(self.game,4)), self.game.current_player)
return np.array(self.game.board)
class C4Env(gym.Env):
"""Custom Environment that follows gym interface"""
metadata = {'render.modes': ['human']}
def __init__(self,board_width=7,board_height=6):
super(C4Env, self).__init__()
self.board_width = board_width
self.board_height = board_height
self.reset()
# The player has the option to
self.action_space = spaces.Discrete(self.board_width)
# Example for using image as input:
self.observation_space = spaces.Box(low=-1, high=1, shape=(self.board_width, self.board_height), dtype=np.int8)
def step(self, action):
# Execute one time step within the environment (each player makes a move)
self.game.play(action, self.game.current_player)
if (not self.game.get_status()):
self.game.play(find_best_move(minmax(self.game,4)), self.game.current_player)
self.game.get_status()
self.steps += 1
reward = float(-self.game.winner) / self.steps
done = self.game.get_status()
obs = self.game.board
return obs, reward, done, {} #TODO was ist {}
def reset(self):
# Reset the state of the environment to an initial state
self.game = Game(self.board_width, self.board_height)
self.steps = 0
if (self.game.current_player==1): # so ist immer der RL Agent als erstes im step dran und kann keinen illegalen zug machen wenn er als action das selbe feld wählt wie der Gegner
self.game.play(find_best_move(minmax(self.game,4)), self.game.current_player)
return np.array(self.game.board)
def render(self, mode='human', close=False):
# Render the environment to the screen
input = np.fliplr(np.rot90(self.game.board, axes=(1,0)))
print(Back.BLUE)
for i in range(len(input)):
for j in range(len(input[i])):
if input[i][j]==-1:
print('\033[31m' + "\u25CF" , end =" ")
elif input[i][j]==1:
print('\033[33m' + "\u25CF" , end =" ")
else:
print('\033[30m' + "\u25CF" , end =" ")
print(Style.RESET_ALL + '\x1b[K')
print(Back.BLUE, end ="")
print(Style.RESET_ALL+'\x1b[K')
def get_legal_moves(self):
return self.game.get_legal_moves()
def test_next_move_gives_the_best_defence_move(self):
minimax = Minimax(1)
game = Game(4, 4)
game.board = [
[0, 0, 0, 0],
[0, 0, 0, 0],
[0, -1, -1, -1],
[0, 0, 1, 1],
]
self.assertEqual(minimax.next_move(game), 2)
def test_next_move_gives_the_best_move_with_diagonals(self):
minimax = Minimax(1)
game = Game(4, 4)
game.board = [
[0, -1, 1, -1],
[0, 1, -1, -1],
[0, -1, 1, -1],
[0, 0, 0, 1],
]
self.assertEqual(minimax.next_move(game), 0)
class RandomEnv(C4Env):
def step(self, action):
# Execute one time step within the environment (each player makes a move)
self.game.play(action, self.game.current_player)
if (not self.game.get_status()):
self.game.play(self.game.random_action(), self.game.current_player)
self.game.get_status()
self.steps += 1
reward = self.boardchecker()
#reward = float(-self.game.winner) / self.steps
done = self.game.get_status()
obs = self.game.board
return obs, reward, done, {} #TODO was ist {}
def reset(self):
# Reset the state of the environment to an initial state
self.game = Game(self.board_width, self.board_height)
self.steps = 0
if (
self.game.current_player == 1
): # so ist immer der RL Agent als erstes im step dran und kann keinen illegalen zug machen wenn er als action das selbe feld wählt wie der Gegner
self.game.play(self.game.random_action(), self.game.current_player)
return np.array(self.game.board)
def boardchecker(self):
#Anzahl der 2er,3er und 4er Ketten der Spieler werden verrechnet
return (self.game.check_for_streak(-1, 2) +
(self.game.check_for_streak(-1, 3) * 10) +
(self.game.check_for_streak(-1, 4) * 1000)) - (
self.game.check_for_streak(1, 2) +
(self.game.check_for_streak(1, 3) * 10) +
(self.game.check_for_streak(1, 4) * 1000))
def main(argv):
first_player = -1
second_player = 1
p1_type = 'neural_network'
p2_type = 'neural_network'
number_of_games = 1
randomness = 0.25
board_width = 4
board_height = 4
try:
opts, args = getopt.getopt(argv, "hv", [
"p1=", "p2=", "board_width=", "board_height=", "iterations=",
"randomness="
])
except getopt.GetoptError:
print 'train.py --p1 <1st player type> --p2 <2nd player type>'
sys.exit(2)
for opt, arg in opts:
if opt == '-h':
print 'train.py --p1 <1st player type> --p2 <2nd player type>'
sys.exit()
if opt == '-v':
global verbose
verbose = True
elif opt in ("--p1"):
p1_type = arg
elif opt in ("--p2"):
p2_type = arg
elif opt in ("--board_width"):
board_width = string.atoi(arg)
elif opt in ("--board_height"):
board_height = string.atoi(arg)
elif opt in ("--iterations"):
number_of_games = string.atoi(arg)
elif opt in ("--randomness"):
randomness = string.atof(arg)
game = Game(board_width, board_height)
batch_statuses = {'draw': 0, 'won_p1': 0, 'won_p2': 0}
game_statuses = {'draw': 0, 'won_p1': 0, 'won_p2': 0}
batch_number = 1
with ai.factory.create(p1_type, first_player) as p1_ai:
with ai.factory.create(p2_type, second_player) as p2_ai:
for game_number in range(number_of_games):
status = game.get_status()
while status == GAME_STATUS['PLAYING']:
current_player = game.current_player
current_ai = p1_ai if current_player == first_player \
else p2_ai
action = current_ai.next_move(game)
if random(
) < randomness and current_player == first_player:
printv('random triggered')
action = game.random_action()
printv(str(current_player) + ' plays ' + str(action))
game.play(action, current_player)
printv(np.matrix(game.board).transpose())
status = game.get_status()
p1_ai.turn_feedback(current_player, action)
p2_ai.turn_feedback(current_player, action)
p1_ai.game_feedback(game, status, game.winner)
p2_ai.game_feedback(game, status, game.winner)
if (game.winner == first_player):
batch_statuses['won_p1'] += 1
elif game.winner == second_player:
batch_statuses['won_p2'] += 1
else:
batch_statuses['draw'] += 1
game.reset()
if game_number > 1 and game_number % 100 == 0:
print("### BATCH N " + str(batch_number) + " ###")
batch_number += 1
handle_stats(game_statuses, batch_statuses, p1_type,
p2_type)
def test_change_turn(self):
game = Game(self.player_one, self.player_two, self.height, self.width)
assert game.current_player == self.player_one
game.change_turn()
assert game.current_player == self.player_two
from connectfour.game import Game, GAME_STATUS from minmax import minmax,find_best_move import numpy as np from colorama import init, Fore , Back , Style import datetime BOARD_WIDTH = 7 BOARD_HIGHT = 6 game = Game(BOARD_WIDTH, BOARD_HIGHT) print(game.board) # game.current_player= 1 # game.play(2, 1) # game.play(2, -1) # game.play(4, 1) # game.play(4, -1) # before = datetime.datetime.now() # minmax(game,7) # after = datetime.datetime.now() # print(after-before) print(find_best_move(minmax(game,4))) input = np.fliplr(np.rot90(game.board, axes=(1,0))) print(Back.BLUE, end ="")