replace_target_iter=100, memory_size=2000,
e_greedy_increment=0.001,)
total_steps = 0
ep_rhistory = []
for i_episode in range(500):
observation = env.reset()
ep_r = 0
while True:
# env.render()
action = RL.choose_action(observation.flatten())
observation_, reward, done, info = env.step(action)
# the smaller theta and closer to center the better
# print(observation_)
# x, x_dot, theta, theta_dot = observation_
# r1 = (env.x_threshold - abs(x))/env.x_threshold - 0.8
# r2 = (env.theta_threshold_radians - abs(theta))/env.theta_threshold_radians - 0.5
# reward = r1 + r2
RL.store_transition(observation.flatten(), action, reward, observation_.flatten())
ep_r += reward
if total_steps > 1000:
RL.learn()
#Initializing
cross = crossroads_map(x, y)
visual = Visual()
obs = []
for xx in x:
for yy in y:
lab = str(xx) + str(yy)
obs = np.concatenate(
(obs, cross[lab].car_nums, cross[lab].light_state), axis=None)
#Training steps
for steps in range(200000):
visual.visual_before(cross, x, y, times, b, bias, bias_t)
action = RL.choose_action(obs)
action_set = [[0 for i in range(grid_y + 1)]
for j in range(grid_x + 1)]
peri_cars = [[([0] * 4) for i in range(grid_y + 1)]
for j in range(grid_x + 1)]
in_cars = [[([0] * 4) for i in range(grid_y + 1)]
for j in range(grid_x + 1)]
#light state changes, cars numbers change, interactions between crossroads and peripherals
for xx in x:
for yy in y:
lab = str(xx) + str(yy)
#10->binary coding for action(1 value), like if action=128, 9 bits binary coding
#of it is 010000000, indicating a 3*3 grid with each crossroad having action of
#'0''1''0''0''0''0''0''0''0'(storing in action set), each action is either '0' or '1',
#for 'change state' or 'keep on'. The binary number is set to be (grid_x*grid_y) bits,