from tensorflow.keras import callbacks
# import xdot
EPOCHS = 20
EPISODES = 50
BOARDSIZE = 9
LEARNINGRATE = 0.0002
MAXTURNS = (BOARDSIZE * BOARDSIZE) - BOARDSIZE
# logdir= "testlog"
# tensorboard_callback = callbacks.TensorBoard(logdir, histogram_freq=1)
# callbacks = [tensorboard_callback]
callbacks = None
env = GoGame(BOARDSIZE, maxTurns=MAXTURNS)
model = Model(num_actions=env.action_space.n, callbacks=callbacks)
model.action_value(env.reset()[None, :]) #build the model
model.summary()
agent = A2CAgent(model, lr=LEARNINGRATE)
for epoch in range(EPOCHS):
# gc.collect()
backend.clear_session()
# objgraph.show_growth()
# print("--------------------------------")
time_start = time.time()
rewards_history, episode_wins, losses = agent.train(env,
max_steps=MAXTURNS,
episodes=EPISODES,
info=False,
info_step=50)
print(
"epoch = {:2} | won {:3}/{:3} matches | mean rewards = {:7.2f} | mean losses = {:7.2f} | epoch time = {:6.2f} sec"
@app.route('/background_process_test')
def background_process_test():
global AI_TURN
if(AI_TURN):
AI_TURN = False
flat_board = get_flat_board()
action, _ = model.action_value(np.asarray(flat_board)[None, :])
coord = env.flatMoveToCoord(int(action))
if (flat_board[int(action)] == 0):
stone = driver.find_element_by_css_selector("#board5 > div.row > div > div:nth-child({}) > div:nth-child({})".format(coord[0]+1,coord[1]+1))
stone.click()
print ("put stone on ({},{})".format(coord[0],coord[1]))
else:
print("invalid move ({},{}) passing instead".format(coord[0],coord[1]))
pass_move = driver.find_element_by_css_selector("#board5 > div.goPlayers > div.stats.white > div.pass")
pass_move.click()
return "Nothing"
else:
AI_TURN = True
return "Nothing"
if __name__ == "__main__":
env = GoGame(5, maxTurns=50)
model = Model(num_actions=env.action_space.n)
model.action_value(env.reset()[None, :])
model.load_weights(MODEL_NAME)
driver = webdriver.Chrome("C:\\Users\\flori\\Desktop\\q_learning\\go_game\\chromedriver.exe")
driver.get("http://127.0.0.1:5000")
app.run()
