# plot
if (arg.show_pre_image):
# img = s_[-1]
if (arg.show_orig_img):
img = pre_process_screen(img, ret_type=arg.img_ret_type)
add_rects(img=img, point=position_, env=env)
# img = cv2.resize(img, (800, 600), interpolation=cv2.INTER_AREA)
else:
img = s_[-1]
if (arg.show_resize_img):
img = cv2.resize(img, (800, 600),
interpolation=cv2.INTER_AREA)
if (cv_img(img)):
break_flag = 1
break
if (model.memory_counter > MEMORY_CAPACITY
and steps % arg.LEARN_TMP_STEP == 0):
model.learn()
# print('----------------------------- learn ------------------', model.learn_step_counter)
# if done:
# print('Ep: ', i_episode,
# '| Ep_r: ', round(ep_r, 2))
# ---------- importrant data ----------------------
# ------------- r_d and r_p
# print('r: {:5.1f}, steps: {:3}, r_d: {:5.2f}, r_p: {:5.2f}, d/p: {:-8.3f}'.format( r, steps, r_d, r_p, d_p))
for i in range(actions.__len__()): # batch times
if (tt.stop_alt()):
break_flag = 1
break
steps += 1
a = actions[i]
s_, r, done, info = env.step(a)
if (not done):
position_, press_shift, pos_passed = info
s_ = preprocess_state(s_, position_, env)
# plot
if (arg.show_pre_image and cv_img(s_[-1])):
break_flag = 1
break
else:
r = arg.reward_done
position_ = (0, 0)
s_ = preprocess_state(s_, position_, env)
if (arg.show_pre_image and cv_img(s_[-1])):
break_flag = 1
break
tt.sleep(0.5)
break
# memory.push(s, r, done, info, s_)
memory.push(s, a, r, info, s_)
point=env.end,
ptype=arg.ptype_dest,
pcolor=arg.pcolor_dest,
ww=20,
hh=20)
# if (cv_img(img)): break_flag = 1; break
# plt_img(img)
# img = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
img.shape
img.dtype
# img = cv2.resize(img, (600, 400), interpolation=cv2.INTER_AREA)
# img = cv2.resize(img, (800, 600), interpolation=cv2.INTER_AREA)
# cv2.rectangle(img, xy0, xy1, (0, 0, 0), -1)
if (arg.show_pre_image and cv_img(img)):
break_flag = 1
break
position = position_
rewards.append(r)
s = s_
if (arg.debug_reward_d_p):
print('{:5.2}, {:5.3f}, {:5},'.format(
rewards[-1], env.rewards_d[-1], env.rewards_p[-1]))
# print('{:5.2f}, {:5.3f}, {:5.3f},'.format( sum(rewards), sum(env.rewards_d), sum(env.rewards_p) ) )
if (done or step >= MAX_STEP - 1):
r = arg.reward_done
death_n = 0
results = ((0, 0), (100, 100))
print('--- death! ---')
# tt.sleep(1)
for j in range(10): #plot death
tt.sleep(0.01)
screen0 = wind.grab_screen()
# screen = screen0
#screen = pre_process_screen(screen0)
screen = cv2.cvtColor(screen0,
cv2.COLOR_BGR2RGB) # RGB
if (cv_img(screen)): break
wind.key_dp(vk.r, tmp)
continue
xy0, xy1 = results[0], results[1]
# xy0, xy1 = (xy0[0] - 10, xy0[0] - 10), (xy1[0] + 10, xy1[0] + 10)
position = np.array((np.array(xy0) + np.array(xy1)) / 2,
dtype=int)
# print(results)
print('第{:4}次, position: {}'.format(i, position))
cv2.rectangle(screen, xy0, xy1, (255, 220, 0), -1)
#screen = cv2.cvtColor(screen0, cv2.COLOR_BGR2GRAY)