def updateQ(c_state, action, alpha, gamma): num_s = add_state(c_state) num_a = lists.action_list.index(action) reward = 0 #get_reward() piece = c_state[1] n_board_state, line_reward = pt.answer(c_state[0], piece[1], piece[2], action[0], action[1]) for i in range(len(n_board_state)): for j in range(len(n_board_state[i])): if n_board_state[i][j] > 0 and n_board_state[i][j] < 8: n_board_state[i][j] = 1 n_state = [n_board_state, pt.getNextPiece()] num_sn = add_state(n_state) if num_sn: pass else:num_sn = add_state(n_state) # else: # lists.state_list.append(n_state) # num_sn = lists.state_list.index(n_state) Q_s_a = get_value(num_s, num_a) r_s_a, height = get_reward(c_state, action) max_Q = get_max_Qvalue(num_sn) value = Q_s_a + alpha * (r_s_a + gamma * max_Q - Q_s_a) if len(lists.Qtable)<=num_s or len(lists.Qtable[num_s])<=num_a: for i in range(len(lists.Qtable), num_s+1): lists.Qtable.append([]) for i in range(len(lists.Qtable[num_s]), num_a+1): lists.Qtable[num_s].append(0) lists.Qtable[num_s][num_a] = value return
def get_reward(c_state, action):
#- c_board_state: 現在の盤面の状態
#- scores = [40, 100, 300, 1200]
#- index: action_listの中でQを最大化するactionを示す番号
c_board_state,piece = c_state
n_board_state, line_reward = pt.answer(c_board_state, piece[1], piece[2], action[0], action[1])
erace_line = line(line_reward) #- 消したライン数
block_reward = reward = 0
## for i in range(0,len(n_board_state)):
## for j in range(0,len(n_board_state[i])):
## if n_board_state[i][j] != 0 and n_board_state[i][j] != 8: n_board_state[i][j] = 1
#- 一番下の空きスペースを埋めた分だけrewardを与える．隙間を作ったら減点
## trans_c = zip(*c_board_state)
## trans_n = zip(*n_board_state)
max_h = 7#21
## for width in range(1, 5):#11
## r = 0
# if trans_c[width][:22 - erace_line] != trans_n[width][erace_line:22]:
## if trans_c[width][:8 - erace_line] != trans_n[width][erace_line:8]:
## r = 3
# l = [num for num in list(trans_n[width]) if num > 0]
# max_1 = l[0]
# start = list(trans_n[width]).index(max_1)
# if l.count(0) + 2 <= len(l): max_2 = l[1]
# else: max_2 = 8
# if max_2 == max_1:
# max_2 = [num for num in list(trans_n[width]) if num != max_1 and num > 0][0]
# print "max_1:",max_1,"max_2:",max_2
# start = list(trans_n[width]).index(max_1)
# final = list(trans_n[width]).index(max_2)
## if [num for num in list(trans_n[width]) if num == 1] != []:
## start = list(trans_n[width]).index(1)
## r -= list(trans_n[width])[erace_line + start:].count(0)
## else: start = 8#22
## r -= 6 - start#20
## if max_h > start:
## max_h = start
## block_reward += r
block_reward = 100/(numpy.max(c_board_state) - numpy.min(c_board_state))
reward = block_reward + line_reward
return reward, max_h
[8,0,0,0,0,0,0,0,0,0,0,8],
[8,0,0,0,0,0,0,0,0,0,0,8],
[8,0,0,0,0,0,0,0,0,0,0,8],
[8,0,0,0,0,0,0,0,0,0,0,8],
[8,0,0,0,0,0,0,0,0,0,0,8],
[8,0,0,0,0,0,0,0,0,0,0,8],
[8,8,8,8,8,8,8,8,8,8,8,8],
[8,8,8,8,8,8,8,8,8,8,8,8],
[8,8,8,8,8,8,8,8,8,8,8,8],
[8,8,8,8,8,8,8,8,8,8,8,8],
[8,8,8,8,8,8,8,8,8,8,8,8],
[8,8,8,8,8,8,8,8,8,8,8,8]]
b = pt.getBoard()
ans = pt.answer(pt.getBoard(),pt.getPiece()[1], pt.getPiece()[2], 0, 2)
#ans = pt.answer([2,3,4,5,4,3],pt.getPiece()[1], pt.getPiece()[2], 0, 2)
pprint(ans)
pt.rotate('ccw')
pt.rotate('ccw')
# メインループ
while(pt.alive()):
#########################
# ループごとの解析、操作をここに書く
pt.drop()
# 次のループへ
pt.loop()
step += 1
