def initialize_buffer_with_single_test_tuple(size):
grid = Grid(filename="levels/" + str(size) + "x" + str(size) +
"/grid_100.txt")
result = set()
states = grid.generate_all_states()
find = False
print("Start state: ")
print(grid.start_state.spaces)
for state in states:
q_state = QState(state)
for action in range((size - 1) * 4):
color = int(action / 4 + 1)
direction = int(action % 4)
action_tu = (color, direction)
new_state, reward, terminal = q_state.step(action_tu)
if terminal:
sars = (q_state, action, reward, new_state, terminal)
print("Old state: ")
print(q_state.state.spaces)
print("New state")
print(new_state.state.spaces)
print("Action: ", action_tu)
print("Action index: ", action)
print("Old state winning: ", q_state.is_winning())
print("New state winning: ", new_state.is_winning())
result.add(sars)
find = True
break
if find:
break
return list(result)
def __init__(self, grid_size, max_generation):
self.grid_size = grid_size
self.grid_size_x, self.grid_size_y = grid_size
self.generation = 0
self.max_generation = max_generation
self.active_cells = 0
self.game_board = Grid(grid_size, Cell)
window_size = tuple([x * 5 for x in grid_size])
self.view = GameView(window_size)
self.empty_grid = Grid(self.grid_size, Cell)
def generate_random_grid(size=4, num_colors=3):
spaces, start_coords = random_start(size, num_colors)
grid = Grid.create(spaces, num_colors, start_coords, end_coords=dict())
state = grid.start_state
searching = True
won = False
while searching:
for col in range(1, num_colors + 1):
possible_actions = state.possible_actions(check_end_tips=False)
if len(possible_actions) == 0:
searching = False
break
actions_for_color = filter_actions_by_color(possible_actions, col)
if len(actions_for_color) == 0:
continue
random_action = random.choice(actions_for_color)
state = state.next_state(random_action, check_end_tips=False)
if state.is_winning(check_end_tips=False):
print("State won!")
searching = False
won = True
break
if won:
state.set_to_start_with_current_tips()
return state
else:
return None
def main():
# Grab the command line options.
options, args = get_options()
# Instantiate the FLowFree grid.
grid = Grid(filename="levels/" + options.level)
# Initialize the renderer.
renderer = GridRenderer(options.level)
print("Algorithm: ", options.algorithm)
value_function, policy = policy_iteration(grid) if options.algorithm == "policy_iteration" else value_iteration(grid)
print("Completed iteration!")
# Now let's try out our policy!
state = grid.start_state
while True:
state = state.next_state(policy[state])
# Draw the grid to the screen.
renderer.render(state)
# Break if we're in the winning state.
if state.is_winning():
break
# Close the window.
renderer.tear_down()
def load_grids(size):
print("Loading grids....")
grids = list()
for i in range(1, 900):
grid = Grid(filename="levels/" + str(size) + "x" + str(size) +
"/grid_" + str(i) + ".txt")
grids.append(grid)
return grids
def _init_grid(self):
self.grid: Grid = Grid(width=int(self.canvas_width / BOX_SIZE),
height=int(self.canvas_height / BOX_SIZE))
self.grid.init_grid()
for x in range(int(self.canvas_width / BOX_SIZE)):
for y in range(int(self.canvas_height / BOX_SIZE)):
if self.canvas_boxes.get(x) is None:
self.canvas_boxes[x] = {}
# sleep(0.1)
self.canvas_boxes[x][y] = BoxCanvas(self.grid.boxes[x][y],
self.canvas)
def __init__(self, layer_nodes: List[int],
processing_config: ProcessingConfig,
importance_data: ImportanceDataHandler = None,
processed_nn: ProcessedNNHandler = None):
logging.info("Prepare network processing for network of size: %s" % layer_nodes)
self.layer_nodes: List[int] = layer_nodes
self.layer_distance: float = processing_config["layer_distance"]
self.layer_width: float = processing_config["layer_width"]
logging.info("Create network model...")
self.network: NetworkModel = NetworkModel(self.layer_nodes, self.layer_width, self.layer_distance,
importance_data, processed_nn, processing_config["prune_percentage"])
self.sample_length: float = self.network.layer_width / processing_config["sampling_rate"]
self.grid_cell_size: float = self.sample_length / 3.0
self.sample_radius: float = self.sample_length * 2.0
RenderShaderHandler().set_classification_number(self.network.num_classes)
ComputeShaderHandler().set_classification_number(self.network.num_classes)
self.node_advection_status: AdvectionProgress = AdvectionProgress(self.network.average_node_distance,
processing_config["node_bandwidth_reduction"],
self.grid_cell_size * 2.0)
self.edge_advection_status: AdvectionProgress = AdvectionProgress(self.network.average_edge_distance,
processing_config["edge_bandwidth_reduction"],
self.grid_cell_size * 2.0)
self.edge_importance_type: int = processing_config["edge_importance_type"]
logging.info("Create grid...")
self.grid: Grid = Grid(Vector3([self.grid_cell_size, self.grid_cell_size, self.grid_cell_size]),
self.network.bounding_volume, self.layer_distance)
logging.info("Prepare node processing...")
self.node_processor: NodeProcessor = NodeProcessor(self.network)
self.node_renderer: NodeRenderer = NodeRenderer(self.node_processor, self.grid)
logging.info("Prepare edge processing...")
self.edge_processor: EdgeProcessor = EdgeProcessor(self.sample_length,
edge_importance_type=self.edge_importance_type)
self.edge_processor.set_data(self.network)
if not self.edge_processor.sampled:
self.edge_processor.init_sample_edge()
self.edge_renderer: EdgeRenderer = EdgeRenderer(self.edge_processor, self.grid)
logging.info("Prepare grid processing...")
self.grid_processor: GridProcessor = GridProcessor(self.grid, self.node_processor, self.edge_processor, 10000.0)
self.grid_processor.calculate_position()
self.grid_renderer: GridRenderer = GridRenderer(self.grid_processor)
self.action_finished: bool = False
self.last_action_mode: NetworkProcess = NetworkProcess.RESET
self.edge_smoothing: bool = processing_config["smoothing"]
self.edge_smoothing_iterations: int = processing_config["smoothing_iterations"]
self.bar: ProgressBar or None = None
def main():
# Grab the command line options.
options, args = get_options()
# Train by iterating over all grids in the training set, while using the
# previous iteration's value function as the starting point for the next
# iteration's training. The idea is that different boards of the same size
# would be able to share states and ths would thus allow us to generalize
# better to unseen boards.
if options.mode == "train":
algorithm = policy_iteration if options.algorithm == "policy_iteration" else value_iteration
value = dict()
policy = dict()
for i in range(1,900):
print("Training iteration " + str(i))
grid = Grid(filename="levels/4x4/grid_" + str(i) + ".txt")
new_value, policy = algorithm(grid, policy, value)
value = new_value.copy()
# Save the result of training to a pickle file.
with open("policies/" + str(options.file) + ".pickle", 'wb') as handle:
pickle.dump(policy, handle, protocol=pickle.HIGHEST_PROTOCOL)
# If we're in test mode, then we load up an existing policy, and have it play
# boards numbered 16-25. We count up how many of those 10 boards it wins.
elif options.mode == "test":
# Load up the pickle file we saved to during training.
with open("policies/" + str(options.file) + ".pickle", 'rb') as handle:
policy = pickle.load(handle)
print("Num keys: ", len(policy.keys()))
num_wins = 0
for i in range(900, 1000):
print("Playing game " + str(i))
grid = Grid(filename="levels/4x4/grid_" + str(i) + ".txt")
num_wins += play_game(grid, policy)
print("Testing won " + str(num_wins) + "out of 100 games!")
def main():
# Grab the command line options.
options, args = get_options()
# Instantiate the FLowFree grid.
grid = Grid(filename="levels/" + options.level)
# Initialize the renderer.
renderer = GridRenderer(options.level)
# Draw the grid to the screen.
renderer.render(grid.start_state)
# Close the window.
renderer.tear_down()
def initialize_replay_buffer_with_single_grid(size, mlp):
print("Initialize replay buffer with single board")
result = set()
grid = Grid(filename="levels/" + str(size) + "x" + str(size) +
"/grid_100.txt")
state = QState(grid.start_state)
for i in range(starting_items_in_replay):
action, q_sa = mlp.get_next_action(state.get_feature_vector(),
grad=False,
exploration_rate=1.0)
new_state, reward, terminal = state.step(action)
sars = (state, action, reward, q_sa, new_state, terminal)
result.add(sars)
state = new_state if not terminal else QState(grid.start_state)
return result
def initialize_buffer_with_all_tuples(size, mlp):
grid = Grid(filename="levels/" + str(size) + "x" + str(size) +
"/grid_1.txt")
result = set()
states = grid.generate_all_states()
for state in states:
q_state = QState(state)
# Don't store the winning state in the replay buffer.
if q_state.is_winning():
print("Don't include me!")
continue
for action in range((size - 1) * 4):
color = int(action / 4 + 1)
direction = int(action % 4)
action_tu = (color, direction)
new_state, reward, terminal = q_state.step(action_tu)
sars = (q_state, action, reward, new_state, terminal)
result.add(sars)
print("Initial replay buffer size: ", len(result))
return list(result)
def test_upward_diagonal_win(self):
o_token = OToken()
grid = Grid(3, 3)
grid_controller = GridController()
grid_controller.assign_grid(grid)
grid.add_token_position(o_token, 0, 2)
grid.add_token_position(o_token, 1, 1)
grid.add_token_position(o_token, 2, 0)
winner = grid_controller.check_for_win()
assert winner == o_token
def test_horizontal_win(self):
x_token = XToken()
grid = Grid(3, 3)
grid_controller = GridController()
grid_controller.assign_grid(grid)
grid.add_token_position(x_token, 0, 0)
grid.add_token_position(x_token, 0, 1)
grid.add_token_position(x_token, 0, 2)
winner = grid_controller.check_for_win()
assert winner == x_token
def test_downward_diagonal_win(self):
x_token = XToken()
grid = Grid(3, 3)
grid_controller = GridController()
grid_controller.assign_grid(grid)
grid.add_token_position(x_token, 0, 0)
grid.add_token_position(x_token, 1, 1)
grid.add_token_position(x_token, 2, 2)
winner = grid_controller.check_for_win()
assert winner == x_token
def play(file, Q, size):
grid = Grid(filename=file)
print("Playing ", file)
epsilon = 0.001
action_size = 4 * (size-1) # number of colors is (size-1), number of directions is 4.
state = grid.start_state
won = False
turns = 0
while turns < 100000:
if random.uniform(0, 1) < epsilon or not state in Q:
action = random.randint(0,action_size-1)
else:
action = np.argmax(Q[state])
color = int(action / 4 + 1)
direction = int(action % 4)
action_tu = (color, direction)
turns += 1
if not state.is_viable_action(action_tu):
continue
# Advance to the next state.
state = state.next_state(action_tu)
# Break if we're in the winning state.
if state.is_winning():
won = True
break
renderer = Renderer("Play")
renderer.render(state)
renderer.tear_down()
print ("Took " + str(turns) + " turns!")
return won
def play(mlp, size=4, index=1):
grid = Grid(filename="levels/" + str(size) + "x" + str(size) + "/grid_" +
str(index) + ".txt")
# Wrap the states as QStates to get functionality
# specifically needed for Q-learning.
state = QState(grid.start_state)
won = False
turns = 0
while turns < 1000:
# Grab the feature vector for the given QState.
features = state.get_feature_vector()
# Get best action from the MLP.
action, _ = mlp.get_next_action(features,
grad=False,
exploration_rate=0.05)
#print("Take action: ", action)
turns += 1
if not state.is_viable_action(action):
continue
# Advance to the next state.
state = state.next_state(action)
# Break if we're in the winning state.
if state.is_winning():
won = True
break
# if won:
# renderer = GridRenderer("Q-Learning")
# renderer.render(state.state)
# renderer.tear_down()
return won
def test_entire_game_with_two_AI(self):
new_game = Game()
new_game.game_mode = 2
token_1 = XToken()
token_2 = OToken()
tokens = (token_1, token_2)
new_game.add_token_options(tokens)
game_controller = GameController()
game_controller.assign_game(new_game)
player_one = Player()
player_two = Player()
players = [player_one, player_two]
game_controller.add_players_to_game(players)
player_one_controller = AIPlayerController()
player_two_controller = AIPlayerController()
player_one_controller.assign_player(player_one)
player_two_controller.assign_player(player_two)
player_controllers = [player_one_controller, player_two_controller]
player_one_controller.determine_name()
player_two_controller.determine_name()
random.shuffle(player_controllers)
game_controller.assign_tokens(player_controllers)
new_game.print_players()
grid = Grid(3, 3)
grid_controller = GridController()
grid_controller.assign_grid(grid)
while not new_game.game_over:
game_controller.play_round(player_controllers, grid_controller)
def test_no_winner_board_full(self):
x_token = XToken()
o_token = OToken()
grid = Grid(3, 3)
grid_controller = GridController()
grid_controller.assign_grid(grid)
grid.add_token_position(x_token, 0, 0)
grid.add_token_position(x_token, 0, 1)
grid.add_token_position(x_token, 1, 2)
grid.add_token_position(x_token, 2, 0)
grid.add_token_position(x_token, 2, 1)
grid.add_token_position(o_token, 0, 2)
grid.add_token_position(o_token, 1, 0)
grid.add_token_position(o_token, 1, 1)
grid.add_token_position(o_token, 2, 2)
winner = grid_controller.check_for_win()
assert winner is None
assert grid_controller.grid_is_full()
class Game():
def __init__(self, grid_size, max_generation):
self.grid_size = grid_size
self.grid_size_x, self.grid_size_y = grid_size
self.generation = 0
self.max_generation = max_generation
self.active_cells = 0
self.game_board = Grid(grid_size, Cell)
window_size = tuple([x * 5 for x in grid_size])
self.view = GameView(window_size)
self.empty_grid = Grid(self.grid_size, Cell)
def run_game(self, frames_per_second=100, display_after_end_of_game=True):
"""
Begins running the game at an optional delay and persists the window unless display_after_end_of_game is False
"""
keep_displaying_window = True
while (self.generation <
self.max_generation) and keep_displaying_window:
self.update_generation()
keep_displaying_window = self.view.update_screen(self.generation)
time.sleep(1 / frames_per_second)
while (keep_displaying_window and display_after_end_of_game):
keep_displaying_window = self.view.update_screen(self.generation)
sys.exit()
def update_generation(self):
"""Checks each cell to see if it survives to the next generation"""
next_generation = copy.deepcopy(
self.empty_grid
) # create new "dead" grid using deepcopy to avoid added iteration in the grid.__init__()
self.view.redraw_screen() # reset screen
row_number = 0
for row in self.game_board:
cell_number = 0
for cell in row:
number_neighbors = self.check_neighboring_living_cells(
(row_number, cell_number))
if cell.is_living():
if number_neighbors == 2 or number_neighbors == 3:
next_generation.getGridItem(
row_number, cell_number).toggle_living(
) # carry over living cell to next generation
self.view.draw_cell(
(row_number, cell_number)) # draw cell
else:
if number_neighbors == 3:
next_generation.getGridItem(
row_number, cell_number
).toggle_living(
) # non-living cell becomes living in the next generation
self.view.draw_cell(
(row_number, cell_number)) # draw cell
cell_number += 1
row_number += 1
self.game_board = next_generation #This generation has ended, copy next generation's grid into current grid
self.generation += 1
def check_neighboring_living_cells(self, coordinates):
"""Check neighboring cells to see if they are living and return the number of living cells"""
x, y = coordinates
num_neighbors = 0
for neighbor_coordinates in NEIGHBORING_CELLS:
neighbor_x, neighbor_y = neighbor_coordinates
neighbor_x = (neighbor_x + x) % self.grid_size_x
neighbor_y = (neighbor_y + y) % self.grid_size_y
if self.game_board.getGridItem(neighbor_x, neighbor_y).is_living():
num_neighbors += 1
return num_neighbors
def load_coordinates_into_grid(self, coordinate_list):
"""Takes a list of coordinates, a tuple(int x, int y), and sets the corresponding cell to living
Raises an IndexError if the coordinates are not
"""
for coordinate in coordinate_list:
x, y = coordinate
if (x < self.grid_size_x and x >= 0 and y < self.grid_size_y
and y >= 0):
self.game_board.getGridItem(x, y).toggle_living()
else:
raise (IndexError)
def get_game_board(self):
"""Returns the current generation's game board"""
return self.game_board
def __init__(self, state):
self.grid = Grid(state)
def train(file, size, Q=dict(), gamma=0.9, num_epochs=3):
print("Train ", file)
print("Epochs: ", num_epochs)
grid = Grid(filename=file)
epsilon = 1.0
print("Generate all states!")
all_states = grid.generate_all_states()
state_size = len(all_states)
print("All states: ", state_size)
lr = 0.5
gamma = 0.9
winning_states = 0
action_size = 4 * (size-1) # number of colors is 4 in 5*5 grid
iter = 0
for epoch in range(num_epochs):
print("Epoch: ", epoch)
for state in all_states:
for action in range(action_size):
if iter % 1000 == 0:
print("iteration ", iter)
if not state in Q:
Q[state] = np.zeros((action_size))
color = int(action /4 + 1)
direction = int(action % 4)
action_tu = (color, direction)
def get_next_tuple():
if state.is_viable_action(action_tu):
new_state = state.next_state(action_tu)
if new_state.is_winning():
reward = 1000000000
return new_state, reward
else:
flows = new_state.completed_flow_count()
zeroes = new_state.num_zeroes_remaining()
reward = -5 * zeroes
for f in range(flows):
reward += 1000
return new_state, reward
else:
reward = -1000000
new_state = state
return new_state, reward
new_state, reward = get_next_tuple()
if not new_state in Q:
Q[new_state] = np.zeros((action_size))
Q[state][action] = Q[state][action] + lr * (reward + gamma * np.max(Q[new_state]) - Q[state][action])
if new_state.is_winning():
print("Winning State!")
winning_states += 1
iter+=1
print("Number of winning states: ", winning_states)
return Q
player_two_controller = AIPlayerController()
else:
player_one_controller = HumanPlayerController()
player_two_controller = HumanPlayerController()
player_one_controller.assign_player(player_one)
player_two_controller.assign_player(player_two)
player_controllers = [player_one_controller, player_two_controller]
print(data.PLAYER_ONE_FIRST_MESSAGE)
player_one_controller.determine_name()
print(data.PLAYER_TWO_NEXT_MESSAGE)
player_two_controller.determine_name()
print(data.RANDOMLY_SELECT_FIRST_PLAYER)
random.shuffle(player_controllers)
game_controller.assign_tokens(player_controllers)
new_game.print_players()
grid = Grid(data.GRID_ROWS, data.GRID_COLUMNS)
grid_controller = GridController()
grid_controller.assign_grid(grid)
while not new_game.game_over:
game_controller.play_round(player_controllers, grid_controller)
play_again = input(data.PLAY_AGAIN_MESSAGE)
if str.lower(play_again) not in ["yes", "y"]:
break
snake_pos_x = 10
snake_pos_y = 10
snake_dir = 'east'
# Game variables
gameOver = False
gameStarted = False
score = 0
highscore = 0
pygame.init()
pygame.display.set_caption("Snake Game")
window = pygame.display.set_mode((SCREEN_WIDTH, SCREEN_HEIGHT))
grid = Grid(window, SCREEN_WIDTH, SCREEN_HEIGHT, SQUARE_SIZE)
snake = Snake(snake_pos_x, snake_pos_y, snake_dir, SQUARE_SIZE)
food = Food(15, 15, SCREEN_WIDTH / SQUARE_SIZE, SCREEN_HEIGHT / SQUARE_SIZE,
SQUARE_SIZE)
menu = Menu(window, SCREEN_WIDTH, SCREEN_HEIGHT)
clock = pygame.time.Clock()
while True:
pygame.time.delay(40)
clock.tick(10)
if gameStarted == False:
menu.draw(score, highscore)
if gameOver == False and gameStarted:
def main():
win = pygame.display.set_mode((540, 600))
pygame.display.set_caption("Sudoku")
board = Grid(9, 9, 540, 540, win)
key = None
run = True
start = time.time()
strikes = 0
while run:
play_time = round(time.time() - start)
for event in pygame.event.get():
if event.type == pygame.QUIT:
run = False
if event.type == pygame.KEYDOWN:
if event.key == pygame.K_1:
key = 1
if event.key == pygame.K_2:
key = 2
if event.key == pygame.K_3:
key = 3
if event.key == pygame.K_4:
key = 4
if event.key == pygame.K_5:
key = 5
if event.key == pygame.K_6:
key = 6
if event.key == pygame.K_7:
key = 7
if event.key == pygame.K_8:
key = 8
if event.key == pygame.K_9:
key = 9
if event.key == pygame.K_DELETE:
board.clear()
key = None
if event.key == pygame.K_SPACE:
board.solve_gui()
if event.key == pygame.K_RETURN:
i, j = board.selected
if board.cubes[i][j].temp != 0:
if board.place(board.cubes[i][j].temp):
print("Success")
else:
print("Wrong")
strikes += 1
key = None
if board.is_finished():
print("Game over")
if event.type == pygame.MOUSEBUTTONDOWN:
pos = pygame.mouse.get_pos()
clicked = board.click(pos)
if clicked:
board.select(clicked[0], clicked[1])
key = None
if board.selected and key != None:
board.sketch(key)
redraw_window(win, board, play_time, strikes)
pygame.display.update()
