def main(_):
# initialize parameters
DATA_DIR = ""
NUM_ACTIONS = 3 # number of valid actions (left, stay, right)
GAMMA = 0.99 # decay rate of past observations
INITIAL_EPSILON = 0.1 # starting value of epsilon
FINAL_EPSILON = 0.0001 # final value of epsilon
MEMORY_SIZE = 50000 # number of previous transitions to remember
NUM_EPOCHS_OBSERVE = 1
NUM_EPOCHS_TRAIN = 1000
BATCH_SIZE = 1
NUM_EPOCHS = NUM_EPOCHS_OBSERVE + NUM_EPOCHS_TRAIN
model_file = "pong_net"
if os.path.exists(model_file):
# load the model
pass
else:
# build the model
pass
pong_game = PongGame()
experience = collections.deque(maxlen=MEMORY_SIZE)
num_games, num_wins = 0, 0
epsilon = INITIAL_EPSILON
for e in range(NUM_EPOCHS):
loss = 0.0
pong_game.reset_game()
# get first state
a_0 = 1 # (0 = left, 1 = stay, 2 = right)
x_t, r_0, game_over = pong_game.step(a_0)
s_t = preprocess_frames(x_t)
while not game_over:
pass
import numpy as np from pong import PongGame if __name__ == '__main__': env = PongGame(competitive = False) obs = env.reset() while 1: env.render(mode='human') obs, reward, dones, win = env.step([np.random.uniform(-1,1),np.random.uniform(-1,1)])