def data_generator_for_CRNN(score_to_begin, score_to_win, batch_size):
datas = []
labels = []
cnt = 0
while 1:
game = Game(score_to_win = score_to_win, random = False)
agent = ExpectiMaxAgent(game)
while game.end == 0:
step = agent.step()
if game.score >= score_to_begin:
board = board2array(game)
board1 = np.swapaxes(board, 1, 2)
board2 = np.swapaxes(board1, 0, 1).reshape((16, 4, 4, 1))
datas.append(board2)
labels.append(step2array(step))
cnt += 1
game.move(step)
if cnt == batch_size:
cnt = 0
datas = np.array(datas)
labels = np.array(labels)
yield (datas, labels)
datas = []
labels = []
def testAgent(n_tests, game_size, score_to_win, model, max_iter=1000):
acc, total = 0, 0
for i in trange(n_tests):
game, n_iter = Game(game_size, score_to_win), 0
target = ExpectiMaxAgent(game)
while n_iter < max_iter and not target.game.end:
dir_ = Greedy_Action(game, model)
target_dir = target.step()
n_iter += 1
total += 1
if dir_ == target_dir:
acc += 1
target.game.move(dir_)
return acc / total
def data_generator_for_CNN(score_to_begin, score_to_win, batch_size):
datas = []
labels = []
cnt = 0
while 1:
game = Game(score_to_win = score_to_win, random = False)
agent = ExpectiMaxAgent(game)
while game.end == 0:
step = agent.step()
if game.score >= score_to_begin:
datas.append(board2array(game))
labels.append(step2array(step))
cnt += 1
game.move(step)
if cnt == batch_size:
cnt = 0
datas = np.array(datas)
labels = np.array(labels)
yield (datas, labels)
datas = []
labels = []
def data_generator(batch_size):
datas = []
labels = []
cnt = 0
while 1:
game = Game(score_to_win = 2048, random = False)
agent = ExpectiMaxAgent(game)
while game.end == 0:
step = agent.step()
board = game.board / 11
board1 = board.T
datas.append(np.hstack((board, board1)))
labels.append(step2array(step))
cnt += 1
game.move(step)
if cnt == batch_size:
cnt = 0
datas = np.array(datas)
labels = np.array(labels)
yield (datas, labels)
datas = []
labels = []
# # print ("direction: ", direction)
# for i in range(4):
# for j in range(4):
# #f.write(game.board[i,j])
# print(game.board[i, j], file = f2)
# print(direction, file = f2)
# #f.write(direction)
# game.move(direction)
# #f.write('\n')
for i in range(300):
print("i = ", i)
game = Game(size=GAME_SIZE, score_to_win=SCORE_TO_WIN0)
agent = ExpectiMaxAgent(game=game)
while True:
direction = agent.step()
if (game.end != 0):
break
# print (game.board)
# print ("direction: ", direction)
if game.board.max() < 256:
for i in range(4):
for j in range(4):
#f.write(game.board[i,j])
print(game.board[i, j], file=f1)
print(direction, file=f1)
elif game.board.max() < 512:
direction = -1
control = "USER"
if request.method == "POST":
direction = request.json
if direction == -1:
direction = agent.step()
control = 'AGENT'
game.move(direction)
return jsonify({"board": game.board.tolist(),
"score": game.score,
"end": game.end,
"direction": direction,
"control": control})
return app
if __name__ == "__main__":
GAME_SIZE = 4
SCORE_TO_WIN = 2048
APP_PORT = 5005
APP_HOST = "localhost"
from game2048.game import Game
from game2048.agents import ExpectiMaxAgent
game = Game(size=GAME_SIZE, score_to_win=SCORE_TO_WIN)
agent = ExpectiMaxAgent(game=game)
app = get_flask_app(game, agent)
app.run(port=APP_PORT, threaded=False, host=APP_HOST) # IMPORTANT: `threaded=False` to ensure correct behavior
from game2048.game import Game
from game2048.displays import Display, IPythonDisplay
from game2048.agents import Agent, RandomAgent, ExpectiMaxAgent
for i in range(0, 1):
game = Game(4, random=False, score_to_win=2048)
agent = ExpectiMaxAgent(game)
agent.play(verbose=False, save=True, savedir='data/2048.txt')
print(str(i + 1) + "th game completed")
display1 = Display()
display2 = IPythonDisplay()
def log2(board):
for i in range(16):
if board[i] != 0:
board[i] = np.log2(board[i])
return board
for i in range(0, 100):
print(i, "is running")
game = Game(4, 2048, random=False)
agent = ExpectiMaxAgent(game, display=display2)
n_iter = 0
max_iter = np.inf
data = np.zeros((0, 17), dtype=float)
while (n_iter < max_iter) and (not game.end):
arr1 = log2(np.reshape(agent.game.board, newshape=(16, )))
direction = agent.step()
arr3 = np.hstack((arr1, direction))
data = np.vstack([data, arr3])
agent.game.move(direction)
n_iter += 1
df = pd.DataFrame(data, columns=None, index=None)
df.to_csv('/home/huanning/big_project/2048-api-master/data_test.csv',
index=0,
mode='a',
header=0)
import sys
sys.path.append("..")
from game2048.expectimax import board_to_move
from game2048.game import Game
from game2048.agents import ExpectiMaxAgent
from game2048.displays import Display
import csv, os
GAME_SIZE = 4
SCORE_TO_WIN = 2048
iter_num = 300
game = Game(GAME_SIZE, SCORE_TO_WIN)
board = game.board
agent = ExpectiMaxAgent(game, Display())
direction = agent.step()
board = game.move(direction)
i = 0
dic = {}
idx = 0
# ------------------------------------------------------
# save each board and its direction to a dict
# -------------------------------------------------------
filename = '/home/zhouykai/Workspace/MachinceLearning/Dataset_2048/Train.csv'
# filename = './Dataset/Train.csv'
if os.path.exists(filename):
head = True
else:
model_256 = load_model('myAgent_256.h5')
model_512 = load_model('myAgent_512.h5')
model_1024 = load_model('myAgent_1024.h5')
boards_256 = []
boards_512 = []
boards_1024 = []
directions_256 = []
directions_512 = []
directions_1024 = []
for i in range(30000):
print('i = ', i)
game = Game(size=4)
expectiMaxAgent = ExpectiMaxAgent(game=game)
while True: # one turn
rightDirection = expectiMaxAgent.step()
if (game.end == True): # game over
break
maxNum = 0
for p in range(4):
for q in range(4):
if game.board[p, q] > maxNum:
maxNum = game.board[p, q]
if maxNum == 2048: # start the next turn
break
inputboard = np.zeros((1, 4, 4, 16))
for p in range(4):
for q in range(4):
num = game.board[p, q]
import numpy as np
from sklearn.model_selection import train_test_split
BATCH_SIZE = 128
NUM_CLASSES = 4
NUM_EPOCHS = 20
display1 = Display()
display2 = IPythonDisplay()
model = keras.models.load_model('model.h5')
image = []
label = []
for i in range(0, 10):
game = Game(4, score_to_win=2048, random=False)
agent = ExpectiMaxAgent(game, display=display1)
while game.end == False:
direction = agent.step()
image.append(game.board)
label.append(direction)
game.move(direction)
display1.display(game)
#运行10次游戏并记录棋盘和方向
x_train = np.array(image)
y_train = np.array(label)
x_train = np.log2(x_train + 1)
x_train = np.trunc(x_train)
def board_log2(board):
for i in range(16):
if board[i] != 0:
board[i] = np.log2(board[i])
return board
# Set the needed number of games
# Limit the max number of steps given by Expectimax each game
NUM_GAMES = 10000
MAX_ITER = 1e3
# This loop takes really a long time
# Use trange to monitor the execution status
for i in trange(NUM_GAMES):
game = Game(size=4, score_to_win=2048)
target = ExpectiMaxAgent(game)
n_iter = 0
data = np.zeros((0, 17), dtype=float)
# To avoid frequent I/O operation,
# data is written into .csv after an entire game rather than one step
while n_iter < MAX_ITER and not target.game.end:
dir_ = target.step()
x = board_log2(np.reshape(target.game.board, newshape=(16, )))
item = np.hstack((x, dir_))
data = np.vstack([data, item])
target.game.move(dir_)
n_iter += 1
df = pd.DataFrame(data, columns=None, index=None)
df.to_csv('./Data_Compressed.csv', mode='a', index=False, header=False)
# print(df)
from game2048.game import Game
from game2048.displays import Display, IPythonDisplay
from game2048.agents import Agent, RandomAgent, ExpectiMaxAgent,YourOwnAgent,SimpleNet
import numpy as np
import pandas as pd
display1 = Display()
display2 = IPythonDisplay()
%%time
for i in range(0,3000):
print(i)
game = Game(4, score_to_win=2048, random=False)
#display2.display(game)
agent = ExpectiMaxAgent(game, display=display2)
max_iter=np.inf
n_iter = 0
while (n_iter < max_iter) and (not game.end):
tmp1 = game.board.reshape(1,16)
direction = np.array(agent.step())
tmp3 = direction.reshape(1,1)
a = np.hstack((tmp1,tmp3))
n_iter += 1
agent.game.move(direction)
df1 = pd.DataFrame((a))
df1.to_csv('lastdata.xls',index=0,mode='a',header=0)
weak_agent.game.reset()
# record board between min and max score
# maybe cannot gennerate min score, lower it by mul 0.6
while (not weak_agent.game.end) and (weak_agent.game.score <=
board_max):
if weak_agent.game.score > board_min * 0.6:
buffer.append(weak_agent.game.board)
direction = weak_agent.step()
weak_agent.game.move(direction)
# return a batch data
while len(buffer) > batch_size:
yield (buffer[:batch_size])
buffer = buffer[batch_size:]
if __name__ == "__main__":
tmpGame = Game(size=4, enable_rewrite_board=True)
expertAgent = ExpectiMaxAgent(tmpGame)
myAgent = myOwnAgent(game=tmpGame,
maxPower=16,
modelLevel=[256, 512, 900],
firstTime=False)
myAgent.train(expertAgent,
train_batch_size=32,
batches_round=1000,
epoch_per_train=10,
max_loop=30)
myAgent.save_model()
boards = []
directions = []
while count < 98:
score_train = []
score_test = []
boards = []
directions = []
count = 0
print('array deleted')
for i in range(30):
game = Game(4, score_to_win=2048, random=False)
agent1 = ExpectiMaxAgent(game, display=display1)
while game.end == False:
if np.sum(game.board) > 384:
break
a = np.array(game.board)
a = np.log2(a + 1)
a = np.trunc(a)
a = keras.utils.to_categorical(a, board_class)
a = a.reshape(1, 4, 4, board_class)
prediction = model.predict(a, batch_size=128)
b = prediction[0]
b = b.tolist()
direction2 = b.index(max(b))
direction1 = agent1.step()
from keras.optimizers import Adam
import numpy as np
from sklearn.model_selection import train_test_split
display1 = Display()
display2 = IPythonDisplay()
model = load_model('2048_new2_2048.h5')
image = []
label = []
#清空棋盘和方向的矩阵
count1 = 0
while count1 < 1:
count = 0
for j in range(0, 50):
game = Game(4, score_to_win=1024, random=False)
agent1 = ExpectiMaxAgent(game)
#agent2 = MyOwnAgent(game, display=display1)
while game.end == False:
direction1 = agent1.step() #用强Agent判断方向
x = np.array(game.board)
x = np.log2(x + 1)
x = np.trunc(x)
x = keras.utils.to_categorical(x, 12)
x = x.reshape(1, 4, 4, 12)
pred = model.predict(x, batch_size=128)
r = pred[0]
r1 = r.tolist()
direction2 = r1.index(max(r1))
#用新跑出来的模型判断方向
from keras.optimizers import RMSprop
import numpy as np
from sklearn.model_selection import train_test_split
image=[]
label=[]
display1 = Display()
display2 = Display()
stop_number = 2048
size = int(np.log2(stop_number)) +1 #跑到stop number时所需的one-hot编码位数
for i in range(0,500): #跑500次棋盘，跑到stop_number停止
game = Game(4, score_to_win=2048, random=False)
agent = ExpectiMaxAgent(game, display=display1) #使用强Agent
while game.end==False:
a=np.array(game.board)
direction=agent.step()
image.append(game.board)
label.append(direction)
game.move(direction)
if np.amax(a)==stop_number:
break
display1.display(game)
image=np.array(image) #将得到的数据和标签转换为numpy数组
label=np.array(label)
import numpy as np
from game2048.game import Game
from game2048.agents import ExpectiMaxAgent
from game2048.agents import MyAgent
from game2048.displays import Display
import csv
import os
game_size = 4
score_to_win = 2048
iter_num = 3000
game = Game(game_size, score_to_win)
board = game.board
agenta = ExpectiMaxAgent(game, Display())
agentb = MyAgent(game, Display())
directiona = agenta.step()
directionb = agentb.step()
board = game.move(directionb)
i = 0
dic = {}
idx = 0
# save file
filename = '/home/olivia/PycharmProjects/2048/game2048/data/traindata10.csv'
if os.path.exists(filename):
start = True
else:
start = False
os.mknod(filename)
from game2048.game import Game
from game2048.displays import Display, IPythonDisplay
from game2048.agents import Agent, RandomAgent, ExpectiMaxAgent, MyAgent
import numpy as np
import pandas as pd
num = 10000
i = 0
results = []
direction = []
while i < num:
game = Game(4, score_to_win=2048, random=False)
agent_exp = ExpectiMaxAgent(game)
agent = MyAgent(game)
while (game.score <= 1024) and (not game.end):
A = game.board
A[A == 0] = 1
A = np.log2(A)
A = np.int32(A)
A = A.reshape(16)
dir = agent.step()
# you can change the condition to get different data
if game.score >= 512:
dir_exp = agent_exp.step()
results.append(A)
direction.append(dir_exp)
game.move(dir)
if 0 == i % 100:
# save the result every 100 games
results = np.array(results)
direction = np.array(direction)