-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathneural_player.py
113 lines (94 loc) · 5 KB
/
neural_player.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
# coding: utf-8
import random
import copy
import numpy as np
import torch
import game
import player
import recently_hands
DEBUG = False
a1_recently_hands = recently_hands.Recently_hands()
a2_recently_hands = recently_hands.Recently_hands()
class DQNPlayer(player.Player):
def __init__(self, model):
super().__init__() #親クラスのコンストラクタ
self.model = model #評価値生成に用いるモデルを設定
if torch.cuda.is_available(): #GPUが使える場合は、モデルをGPUに転送
self.model.cuda()
def select(self, field, player):
value = field.value
own_state = field.own_state
opponent_state = field.opponent_state
pos = field.conv_agent_field([field.conv_turn_pos(player)['x'], field.conv_turn_pos(player)['y']])
#大きさをフィールドの最大値に固定
value = np.pad(value, [(0, game.MAX_BOARD_SIZE - field.value.shape[0]),(0, game.MAX_BOARD_SIZE - field.value.shape[1])], 'constant')
own_state = np.pad(own_state, [(0, game.MAX_BOARD_SIZE - field.value.shape[0]),(0, game.MAX_BOARD_SIZE - field.value.shape[1])], 'constant')
opponent_state = np.pad(opponent_state, [(0, game.MAX_BOARD_SIZE - field.value.shape[0]),(0, game.MAX_BOARD_SIZE - field.value.shape[1])], 'constant')
pos = np.pad(pos, [(0, game.MAX_BOARD_SIZE - field.value.shape[0]),(0, game.MAX_BOARD_SIZE - field.value.shape[1])], 'constant')
inputs = np.array([value, own_state, opponent_state, pos]).reshape(1, 4, game.MAX_BOARD_SIZE, game.MAX_BOARD_SIZE) #入力データを結合させる サイズは[バッチ, チャンネル, height, width]
self.model.eval() #推論モード
inputs = torch.from_numpy(inputs).float()
if torch.cuda.is_available(): #GPUを使える時
inputs = torch.autograd.Variable(inputs.cuda())
else:
inputs = torch.autograd.Variable(inputs)
out = self.model(inputs)
if DEBUG is True:
print(out)
max_sorted = torch.sort(out, descending=True) #価値が高い順に並べる
sorted_directions = max_sorted[1][0].tolist() #移動方向のみのリストを生成
# print("sorted_direction", sorted_directions)
for move_direction in sorted_directions:
hand = field.conv_direction_hand(move_direction, own_state, opponent_state, [field.conv_turn_pos(player)['x'], field.conv_turn_pos(player)['y']])
# 近づかないように
if hand is not None:
if player == game.OWN_2:
a1_x = field.own_a1['x']
a1_y = field.own_a1['y']
hand_x = hand[0]['x']
hand_y = hand[0]['y']
dis_x = hand_x - a1_x
dis_y = hand_y - a1_y
if dis_x < 2 or dis_y < 2:
continue
# 停留をナシにする
if move_direction == 0:
continue
if DEBUG is True:
print("dict : ", move_direction, "hand : ", hand)
if player == game.OWN_1:
if a1_recently_hands.check(hand, times = 3) is False:
# print("continue hand", hand)
continue
elif player == game.OWN_2:
if a2_recently_hands.check(hand, times = 3) is False:
# print("continue hand", hand)
continue
if hand is None: continue #不可能な手だったら、次点の手について処理する
if player == game.OWN_1:
a1_recently_hands.put(hand)
if DEBUG is True:
print("a1_recently_hands.hands", a1_recently_hands.hands)
elif player == game.OWN_2:
a2_recently_hands.put(hand)
if DEBUG is True:
print("a2_recently_hands.hands", a2_recently_hands.hands)
if DEBUG is True:
print("hand", hand)
return hand
if DEBUG is True:
print("player:", player, " put random hand")
#continueしすぎて可能な手を全部スキップした（なんで？）
hands = field.hands(field.own_state, field.opponent_state, player) #可能な手
if len(hands) == 0: #手がない
hand = field.conv_direction_hand(0 , own_state, opponent_state, [field.conv_turn_pos(player)['x'], field.conv_turn_pos(player)['y']]) #停留
else:
choice = random.randrange(len(hands)) #ランダムに一つ選択
hand = hands[choice]
return hand
# class MinimaxDQNPlayer(player.Player):
# def __init__(self, model):
# super().__init__() #親クラスのコンストラクタ
# self.model = model #評価値生成に用いるモデルを設定
# def select(self, field, player):
# pass