def generate_statistics(self, filename=TRAINING_DATA_FILENAME):
"""
Run the module. Play 2048 50 times using random player, then 50 times using the neural network. Report results.
:return: Results from all 100 playings, in two lists. One list for random, and one for intelligent
"""
results = []
try:
for _ in range(50):
# Training the neural network
self.ann = Ann(self.hidden_layer_sizes,
self.activation_functions, 16, 4,
self.learning_rate, self.error_limit,
self.batch_size, self.max_epochs)
self.train(filename)
# Playing using the neural network
results.append(self.play_intelligently())
except KeyboardInterrupt:
pass
logging.info(
'Completed playing the intelligent games. Mean highest value: %f' %
numpy.mean(results))
return results
def __init__(self,
hidden_layer_sizes,
activation_functions,
max_epochs,
learning_rate,
error_limit,
batch_size=100):
self.hidden_layer_sizes = hidden_layer_sizes
self.activation_functions = activation_functions
self.max_epochs = max_epochs
self.learning_rate = learning_rate
self.error_limit = error_limit
self.batch_size = batch_size
self.ann = Ann(hidden_layer_sizes, activation_functions, 16, 4,
learning_rate, error_limit, batch_size, max_epochs)
class ModuleFive:
def __init__(self, hidden_layer_sizes, activation_functions, learning_rate=0.1, error_limit=1e-4, max_epochs=50):
self.ann = Ann(hidden_layer_sizes, activation_functions, learning_rate=learning_rate,
error_limit=error_limit, max_epochs=max_epochs)
# Trains the network, before calling the major_demo function
def run(self, r=0):
epochs, accuracy = self.ann.train_mnist()
return mnistdemo.major_demo(self.ann, r, mnist_basics.__mnist_path__), epochs, accuracy
def __init__(self, hidden_layer_sizes, activation_functions, max_epochs, learning_rate, error_limit,
batch_size=100):
self.hidden_layer_sizes = hidden_layer_sizes
self.activation_functions = activation_functions
self.max_epochs = max_epochs
self.learning_rate = learning_rate
self.error_limit = error_limit
self.batch_size = batch_size
self.ann = Ann(hidden_layer_sizes, activation_functions, 16, 4, learning_rate, error_limit,
batch_size, max_epochs)
def generate_statistics(self, filename=TRAINING_DATA_FILENAME):
"""
Run the module. Play 2048 50 times using random player, then 50 times using the neural network. Report results.
:return: Results from all 100 playings, in two lists. One list for random, and one for intelligent
"""
results = []
try:
for _ in range(50):
# Training the neural network
self.ann = Ann(self.hidden_layer_sizes, self.activation_functions, 16, 4, self.learning_rate,
self.error_limit, self.batch_size, self.max_epochs)
self.train(filename)
# Playing using the neural network
results.append(self.play_intelligently())
except KeyboardInterrupt:
pass
logging.info('Completed playing the intelligent games. Mean highest value: %f' % numpy.mean(results))
return results
class Ann2048:
def __init__(self,
hidden_layer_sizes,
activation_functions,
max_epochs,
learning_rate,
error_limit,
batch_size=100):
self.hidden_layer_sizes = hidden_layer_sizes
self.activation_functions = activation_functions
self.max_epochs = max_epochs
self.learning_rate = learning_rate
self.error_limit = error_limit
self.batch_size = batch_size
self.ann = Ann(hidden_layer_sizes, activation_functions, 16, 4,
learning_rate, error_limit, batch_size, max_epochs)
@staticmethod
def nest_list(flat_list):
"""
Nest list. Return a nested 4x4 version of the flat lits supplied
:param flat_list: List to nest
:return: Nested list
"""
nested_list = numpy.array(flat_list)
nested_list = numpy.reshape(nested_list, (4, 4))
return list(nested_list.tolist())
@staticmethod
def flatten_list(nested_list):
"""
Flatten list. Return a flat version of the supplied nested list
:param nested_list: Nested list to flatten
:return: Flattened list
"""
def kd_reduce(func, seq):
res = seq[0]
for item in seq[1:]:
res = func(res, item)
return res
def flatten(a, b):
return a + b
return kd_reduce(flatten, nested_list)
def pre_process(self, feature_sets):
for feature_set in feature_sets:
for i in range(len(feature_set)):
feature_set[i] *= feature_set[i]
return feature_sets
def train(self, filename):
"""
Train the network using supplied file
:param filename: Filename containing training data
"""
feature_sets, correct_labels = self.ann.load_data(filename)
feature_sets = self.pre_process(feature_sets)
self.ann.start_training(feature_sets, correct_labels)
@staticmethod
def list_from_integer(integer):
return list(map(lambda x: (integer >> x * 4) & 0xf, range(16)))
def load_integer_formatted_data(self, filename):
"""
Load data from a file, where the states are formatted as integers
:param filename: Name of file containing data to load
:return: Feature sets and labels from the supplied data
"""
feature_sets = []
correct_labels = []
with open(filename, 'r') as f:
for line in f.readlines():
line = line.strip()
content = line.split(',')
feature_sets.append(self.list_from_integer(int(content[0])))
correct_labels.append(int(content[1]))
return feature_sets, correct_labels
def play_intelligently(self):
"""
Play 2048 using the trained neural network
:return: Highest tile achieved
"""
logging.info('Playing using the neural network')
ui = Ui()
twenty_forty_eight = TwentyFortyEight(ui=ui)
is_game_over = not twenty_forty_eight.make_computer_move()
# Makes moves as long as the game isn't lost yet
while not is_game_over:
# Makes the move
board_state = self.flatten_list(
twenty_forty_eight.game_board.get_cell_values())
next_moves = self.ann.testing_function_list([board_state])[0]
moved = False
illegal_moves = 0
while not moved:
next_move_index = next_moves.argmax()
moved = twenty_forty_eight.game_board.make_player_move(
next_move_index)
if not moved:
next_moves[next_move_index] = -1
illegal_moves += 1
if illegal_moves == 4:
break
# Updates UI, if any
if not moved and ui:
ui.update_ui(twenty_forty_eight.game_board.state)
# Spawns new value
is_game_over = not twenty_forty_eight.make_computer_move()
result = 2**twenty_forty_eight.game_board.get_max_value()
logging.info('Result: %d' % result)
# Returns final score
return result
@staticmethod
def play_randomly():
"""
Play 2048 using only random moves
:return: Highest tile achieved
"""
twenty_forty_eight = TwentyFortyEight()
logging.info('Playing using random moves')
result = twenty_forty_eight.run(twenty_forty_eight.make_random_move)
logging.info('Result: %d' % result)
return result
def run(self, filename=TRAINING_DATA_FILENAME):
"""
Run the module. Play 2048 50 times using random player, then 50 times using the neural network. Report results.
:return: Results from all 100 playings, in two lists. One list for random, and one for intelligent
"""
results = [
[], # Random
[] # Intelligently
]
# Playing using random player
for i in range(50):
results[0].append(self.play_randomly())
logging.info(
'Completed playing the random games. Mean highest value: %f' %
numpy.mean(results[0]))
# Training the neural network
self.train(filename)
# Playing using the neural network
for i in range(50):
results[1].append(self.play_intelligently())
logging.info(
'Completed playing the intelligent games. Mean highest value: %f' %
numpy.mean(results[1]))
return results
def generate_statistics(self, filename=TRAINING_DATA_FILENAME):
"""
Run the module. Play 2048 50 times using random player, then 50 times using the neural network. Report results.
:return: Results from all 100 playings, in two lists. One list for random, and one for intelligent
"""
results = []
try:
for _ in range(50):
# Training the neural network
self.ann = Ann(self.hidden_layer_sizes,
self.activation_functions, 16, 4,
self.learning_rate, self.error_limit,
self.batch_size, self.max_epochs)
self.train(filename)
# Playing using the neural network
results.append(self.play_intelligently())
except KeyboardInterrupt:
pass
logging.info(
'Completed playing the intelligent games. Mean highest value: %f' %
numpy.mean(results))
return results
def generate_training_data(self, filename=TRAINING_DATA_FILENAME):
"""
Generate training data sets. Play the 2048 game using A*-GAC, and write every move made to supplied file
:param filename: Filename to save the results to
"""
while True:
logging.info('Starting AI run')
twenty_forty_eight = TwentyFortyEight()
moves = twenty_forty_eight.run(twenty_forty_eight.make_player_move,
True)
logging.info('AI run completed. Saving to file')
with open(filename, 'a') as f:
for move in moves:
game_board = self.flatten_list(move[0])
move_made = move[1]
f.write('[%s] %d\n' %
(','.join([str(element)
for element in game_board]), move_made))
logging.info('Moves successfully saved to file %s' % filename)
def __init__(self, hidden_layer_sizes, activation_functions, learning_rate=0.1, error_limit=1e-4, max_epochs=50):
self.ann = Ann(hidden_layer_sizes, activation_functions, learning_rate=learning_rate,
error_limit=error_limit, max_epochs=max_epochs)
class Ann2048:
def __init__(self, hidden_layer_sizes, activation_functions, max_epochs, learning_rate, error_limit,
batch_size=100):
self.hidden_layer_sizes = hidden_layer_sizes
self.activation_functions = activation_functions
self.max_epochs = max_epochs
self.learning_rate = learning_rate
self.error_limit = error_limit
self.batch_size = batch_size
self.ann = Ann(hidden_layer_sizes, activation_functions, 16, 4, learning_rate, error_limit,
batch_size, max_epochs)
@staticmethod
def nest_list(flat_list):
"""
Nest list. Return a nested 4x4 version of the flat lits supplied
:param flat_list: List to nest
:return: Nested list
"""
nested_list = numpy.array(flat_list)
nested_list = numpy.reshape(nested_list, (4, 4))
return list(nested_list.tolist())
@staticmethod
def flatten_list(nested_list):
"""
Flatten list. Return a flat version of the supplied nested list
:param nested_list: Nested list to flatten
:return: Flattened list
"""
def kd_reduce(func, seq):
res = seq[0]
for item in seq[1:]:
res = func(res, item)
return res
def flatten(a, b):
return a + b
return kd_reduce(flatten, nested_list)
def pre_process(self, feature_sets):
for feature_set in feature_sets:
for i in range(len(feature_set)):
feature_set[i] *= feature_set[i]
return feature_sets
def train(self, filename):
"""
Train the network using supplied file
:param filename: Filename containing training data
"""
feature_sets, correct_labels = self.ann.load_data(filename)
feature_sets = self.pre_process(feature_sets)
self.ann.start_training(feature_sets, correct_labels)
@staticmethod
def list_from_integer(integer):
return list(map(lambda x: (integer >> x*4) & 0xf, range(16)))
def load_integer_formatted_data(self, filename):
"""
Load data from a file, where the states are formatted as integers
:param filename: Name of file containing data to load
:return: Feature sets and labels from the supplied data
"""
feature_sets = []
correct_labels = []
with open(filename, 'r') as f:
for line in f.readlines():
line = line.strip()
content = line.split(',')
feature_sets.append(self.list_from_integer(int(content[0])))
correct_labels.append(int(content[1]))
return feature_sets, correct_labels
def play_intelligently(self):
"""
Play 2048 using the trained neural network
:return: Highest tile achieved
"""
logging.info('Playing using the neural network')
ui = Ui()
twenty_forty_eight = TwentyFortyEight(ui=ui)
is_game_over = not twenty_forty_eight.make_computer_move()
# Makes moves as long as the game isn't lost yet
while not is_game_over:
# Makes the move
board_state = self.flatten_list(twenty_forty_eight.game_board.get_cell_values())
next_moves = self.ann.testing_function_list([board_state])[0]
moved = False
illegal_moves = 0
while not moved:
next_move_index = next_moves.argmax()
moved = twenty_forty_eight.game_board.make_player_move(next_move_index)
if not moved:
next_moves[next_move_index] = -1
illegal_moves += 1
if illegal_moves == 4:
break
# Updates UI, if any
if not moved and ui:
ui.update_ui(twenty_forty_eight.game_board.state)
# Spawns new value
is_game_over = not twenty_forty_eight.make_computer_move()
result = 2 ** twenty_forty_eight.game_board.get_max_value()
logging.info('Result: %d' % result)
# Returns final score
return result
@staticmethod
def play_randomly():
"""
Play 2048 using only random moves
:return: Highest tile achieved
"""
twenty_forty_eight = TwentyFortyEight()
logging.info('Playing using random moves')
result = twenty_forty_eight.run(twenty_forty_eight.make_random_move)
logging.info('Result: %d' % result)
return result
def run(self, filename=TRAINING_DATA_FILENAME):
"""
Run the module. Play 2048 50 times using random player, then 50 times using the neural network. Report results.
:return: Results from all 100 playings, in two lists. One list for random, and one for intelligent
"""
results = [
[], # Random
[] # Intelligently
]
# Playing using random player
for i in range(50):
results[0].append(self.play_randomly())
logging.info('Completed playing the random games. Mean highest value: %f' % numpy.mean(results[0]))
# Training the neural network
self.train(filename)
# Playing using the neural network
for i in range(50):
results[1].append(self.play_intelligently())
logging.info('Completed playing the intelligent games. Mean highest value: %f' % numpy.mean(results[1]))
return results
def generate_statistics(self, filename=TRAINING_DATA_FILENAME):
"""
Run the module. Play 2048 50 times using random player, then 50 times using the neural network. Report results.
:return: Results from all 100 playings, in two lists. One list for random, and one for intelligent
"""
results = []
try:
for _ in range(50):
# Training the neural network
self.ann = Ann(self.hidden_layer_sizes, self.activation_functions, 16, 4, self.learning_rate,
self.error_limit, self.batch_size, self.max_epochs)
self.train(filename)
# Playing using the neural network
results.append(self.play_intelligently())
except KeyboardInterrupt:
pass
logging.info('Completed playing the intelligent games. Mean highest value: %f' % numpy.mean(results))
return results
def generate_training_data(self, filename=TRAINING_DATA_FILENAME):
"""
Generate training data sets. Play the 2048 game using A*-GAC, and write every move made to supplied file
:param filename: Filename to save the results to
"""
while True:
logging.info('Starting AI run')
twenty_forty_eight = TwentyFortyEight()
moves = twenty_forty_eight.run(twenty_forty_eight.make_player_move, True)
logging.info('AI run completed. Saving to file')
with open(filename, 'a') as f:
for move in moves:
game_board = self.flatten_list(move[0])
move_made = move[1]
f.write('[%s] %d\n' % (','.join([str(element) for element in game_board]), move_made))
logging.info('Moves successfully saved to file %s' % filename)
