MLE-Pacman/ReinforcmentLearning/game.py

302 lines
9.3 KiB
Python

import pygame
import math
import os
from ReinforcmentLearning.util import Direction, calc_current_state, calc_time_reward, epsilon_greedy, get_best_q_action, initial_q_fill
import conf
# Initialize pygame
# Define constants
SCREEN_WIDTH = 400
SCREEN_HEIGHT = 400
CELL_SIZE = 40
# Define colors
YELLOW = (255, 255, 0)
RED = (255, 0, 0)
WHITE = (255, 255, 255)
BLUE = (0, 0, 255)
BLACK = (0, 0, 0)
REWARD_ON_HALF = 50
# Labyrinth as a string
LABYRINTH_INIT = [
"##########",
"#........#",
"#.##..##.#",
"#........#",
"##########"
]
# Get labyrinth dimensions
ROWS = len(LABYRINTH_INIT)
COLS = len(LABYRINTH_INIT[0])
class Pacman:
def __init__(self, screen, x, y):
self.screen = screen
self.x = x
self.y = y
self.count = 0
def move(self, labyrinth, dx, dy):
new_x, new_y = self.x + dx, self.y + dy
if labyrinth[new_y][new_x] != "#":
self.x = new_x
self.y = new_y
def draw(self):
radius = CELL_SIZE // 2 - 4
start_angle = math.pi / 6
end_angle = -math.pi / 6
pygame.draw.circle(self.screen, YELLOW, (self.x * CELL_SIZE + CELL_SIZE // 2, self.y * CELL_SIZE + CELL_SIZE // 2), CELL_SIZE // 2 - 4)
# Calculate the points for the mouth
start_pos = (self.x* CELL_SIZE + CELL_SIZE // 2 + int(radius*1.3 * math.cos(start_angle)),
self.y* CELL_SIZE + CELL_SIZE // 2 - int(radius*1.3 * math.sin(start_angle)))
end_pos = (self.x* CELL_SIZE + CELL_SIZE // 2 + int(radius*1.3 * math.cos(end_angle)),
self.y* CELL_SIZE + CELL_SIZE // 2 - int(radius*1.3 * math.sin(end_angle)))
self.count += 1
if self.count%2==0:
# Draw the mouth by filling a polygon
pygame.draw.polygon(self.screen, BLACK, [(self.x* CELL_SIZE + CELL_SIZE // 2, self.y* CELL_SIZE + CELL_SIZE // 2), start_pos, end_pos])
class Ghost:
# Define the pixel art for the ghost using strings
ghost_pixels = [
" #### ",
"######",
"## # #",
"######",
"######",
"# # # "
]
def __init__(self, screen, x, y):
self.screen = screen
self.x = x
self.y = y
def move_towards_pacman(self, labyrinth, pacman):
if self.x < pacman.x and labyrinth[self.y][self.x + 1] != "#":
self.x += 1
elif self.x > pacman.x and labyrinth[self.y][self.x - 1] != "#":
self.x -= 1
elif self.y < pacman.y and labyrinth[self.y + 1][self.x] != "#":
self.y += 1
elif self.y > pacman.y and labyrinth[self.y - 1][self.x] != "#":
self.y -= 1
def draw(self):
pixel_size = CELL_SIZE // len(self.ghost_pixels) # Size of each pixel in the ghost art
for row_idx, row in enumerate(self.ghost_pixels):
for col_idx, pixel in enumerate(row):
if pixel == "#":
pixel_x = self.x * CELL_SIZE + col_idx * pixel_size
pixel_y = self.y * CELL_SIZE + row_idx * pixel_size
pygame.draw.rect(self.screen, RED, (pixel_x, pixel_y, pixel_size, pixel_size))
def start_try(EPSILON, ALPHA, GAMMA):
#? Learning initial
q_values = initial_q_fill()
#? Game initial
pygame.init()
screen = None
if conf.show_game:
screen = pygame.display.set_mode((COLS * CELL_SIZE, ROWS * CELL_SIZE))
pygame.display.set_caption("Micro-Pacman")
#? Start try
cookies_per_run = []
iterations = []
for x in range(conf.AMOUNT_RUNS):
if conf.show_game:
if x == conf.AMOUNT_RUNS / 4:
print("1 / 4 done")
if x == conf.AMOUNT_RUNS / 2:
print("2 / 4 done")
if x == (conf.AMOUNT_RUNS / 2) + (conf.AMOUNT_RUNS / 4):
print("3 / 4 done")
amount_cookies_ate, iterations_per_run = run_game(q_values, EPSILON, ALPHA, GAMMA, screen)
cookies_per_run.append(amount_cookies_ate)
iterations.append(iterations_per_run)
print(f"Run {x+1}: {iterations_per_run} iterations")
if conf.show_trained:
screen = pygame.display.set_mode((COLS * CELL_SIZE, ROWS * CELL_SIZE))
pygame.display.set_caption("Micro-Pacman")
while True:
print("After game")
run_game(q_values, EPSILON, ALPHA, GAMMA, screen)
pygame.quit()
return cookies_per_run, iterations
def run_game(q_values, EPSILON, ALPHA, GAMMA, screen):
clock = pygame.time.Clock()
labyrinth = LABYRINTH_INIT.copy()
# Initialize Pacman and Ghost positions
pacman = Pacman(screen, 1, 1)
ghost = Ghost(screen, COLS - 2, ROWS - 2)
#? -------------------------MY CODE-----------------------------------
state = calc_current_state(labyrinth, pacman.x, pacman.y, ghost.x, ghost.y)
#? -------------------------MY CODE-----------------------------------
#? GAME LOOP
running = True
iter = 0
while running:
#? -------------------------MY CODE-----------------------------------
reward = 0
#? -------------------------MY CODE-----------------------------------
# Handle events
for event in pygame.event.get():
if event.type == pygame.QUIT:
running = False
if conf.show_game:
screen.fill(BLACK)
iter = iter + 1
# Handle Pacman movement
keys = pygame.key.get_pressed()
if keys[pygame.K_LEFT]:
pacman.move(-1, 0)
if keys[pygame.K_RIGHT]:
pacman.move(1, 0)
if keys[pygame.K_UP]:
pacman.move(0, -1)
if keys[pygame.K_DOWN]:
pacman.move(0, 1)
#? -------------------------MY CODE-----------------------------------
action = epsilon_greedy(q_values, state, EPSILON)
if action == Direction.LEFT:
pacman.move(labyrinth, -1, 0)
if action == Direction.RIGHT:
pacman.move(labyrinth, 1, 0)
if action == Direction.UP:
pacman.move(labyrinth, 0, -1)
if action == Direction.DOWN:
pacman.move(labyrinth, 0, 1)
#? -------------------------MY CODE-----------------------------------
if iter%3==0:
ghost.move_towards_pacman(labyrinth, pacman)
if pacman.x == ghost.x and pacman.y == ghost.y:
if conf.show_game:
print("Game Over! The ghost caught Pacman.")
running = False
reward = conf.REWARD_ON_LOSE
# Eat cookies
if labyrinth[pacman.y][pacman.x] == ".":
labyrinth[pacman.y] = labyrinth[pacman.y][:pacman.x] + " " + labyrinth[pacman.y][pacman.x+1:]
#? -------------------------MY CODE-----------------------------------
#? half reward
# cookie_counter = 0
# for y, row in enumerate(labyrinth):
# for x, cell in enumerate(row):
# if cell == ".":
# cookie_counter += 1
# if cookie_counter == 10:
# # reward = REWARD_ON_HALF
# if show_game:
# print("Got half reward")
#? -------------------------MY CODE-----------------------------------
# Check if all cookies are eaten (game over)
if all("." not in row for row in labyrinth):
# time_reward = calc_time_reward(iter)
# reward = REWARD_ON_WIN * time_reward
reward = conf.REWARD_ON_WIN
running = False
if conf.show_game:
# print(f"You Win! Took {iter} iterations, reward: {time_reward}")
print(f"You Win! Took {iter} iterations")
#? -------------------------MY CODE-----------------------------------
if not running:
new_state = state
else:
new_state = calc_current_state(labyrinth, pacman.x, pacman.y, ghost.x, ghost.y)
best_action_new_state, _ = get_best_q_action(q_values, new_state)
best_value_new_state = q_values[(new_state, best_action_new_state)]
current_value = q_values.get((state, action))
adjusted_value = ALPHA * (reward + GAMMA * best_value_new_state - current_value)
q_values[(state, action)] = current_value + adjusted_value
state = new_state
if not running:
counter = 0
for y, row in enumerate(labyrinth):
for x, cell in enumerate(row):
if cell == ".":
counter += 1
return 20-counter, iter
#? -------------------------MY CODE-----------------------------------
# Draw the labyrinth, pacman, and ghost
if conf.show_game:
draw_labyrinth(screen, labyrinth)
pacman.draw()
ghost.draw()
# Update display
pygame.display.flip()
# Cap the frame rate
clock.tick(40)
def draw_labyrinth(screen, labyrinth):
for y, row in enumerate(labyrinth):
for x, cell in enumerate(row):
if cell == "#":
pygame.draw.rect(screen, BLUE, (x * CELL_SIZE, y * CELL_SIZE, CELL_SIZE, CELL_SIZE))
elif cell == ".":
pygame.draw.circle(screen, WHITE, (x * CELL_SIZE + CELL_SIZE // 2, y * CELL_SIZE + CELL_SIZE // 2), 5)
if __name__ == "__main__":
run_game()