if xx + 1 <= grid_x and in_cars[xx + 1][yy][3] > 0:
cross_ = cross[lab]
cross_.car_nums[3] += in_cars[xx + 1][yy][3]
cross[lab] = cross_
#in the same diagram as the above 'visual_before', showing the incoming cars
visual.visual_peri(peri_cars, x, y, times, b, bias, bias_, bias_t,
grid_x, grid_y)
reward = 0
for xx in x:
for yy in y:
lab = str(xx) + str(yy)
for i in range(4):
reward = reward - cross[lab].car_nums[i]**2
#show the result of state tranformation in another diagram 'visual_after', i.e. the result of 'visual_before'
visual.visual_after(cross, x, y, times, b, bias, bias_t)
time.sleep(10)
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)
RL.store_transition(obs, action, reward, obs_)
if steps > 200:
RL.learn()
if steps % 50 == 0:
peri_cars1, in_cars1 = cross1.state_change(action1)
peri_cars2, in_cars2 = cross2.state_change(action2)
print(peri_cars1, in_cars1, peri_cars2, in_cars2)
visual.visual_peri(peri_cars1,peri_cars2)
reward=0
for i in range (4):
if cross2.q_states[i]==1:
cross2.car_nums[i]+=in_cars1
if cross1.q_states[i]==1:
cross1.car_nums[i]+=in_cars2
reward = reward - cross1.car_nums[i]**2 - cross2.car_nums[i]**2
visual.visual_after(cross1, cross2)
obs_=np.concatenate((cross1.car_nums, cross1.light_state, cross2.car_nums, cross2.light_state),axis=None)
RL.store_transition(obs,action,reward,obs_)
if steps>200:
RL.learn()
if steps%50==0:
print(steps,reward)
reward_set.append(reward)
step_set.append(steps)
#plt.scatter(steps, reward)
obs=obs_
# window.mainloop()
plt.plot(step_set,reward_set)
