ProjectZoe/optimal_solver.py at main · sqmscm/ProjectZoe · GitHub

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
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
"""
turn: 0: white, 1: black
piece: -1: black, 0: space, 1: white, 2: obstacle
"""
import math
import config
import numpy as np
from collections import defaultdict


def comb(n, k):
    a = 1
    for i in range(n - k + 1, n + 1):
        a *= i
    return a // math.factorial(k)


allowed_rows_cols = {x: defaultdict(dict) for x in config.supported_model}

side_combinations = {}
position_combinations = {}
obstacles_combinations = {}

piece_position_table = {}
inv_piece_position_table = {}
side_position_table = {}
inv_side_position_table = {}


def make_position_table(t, comb_num, limit):
    k = list(range(t))
    position_table = []
    inv_position_table = {}
    for i in range(comb_num):
        position_table.append(k.copy())
        inv_position_table.update({frozenset(k.copy()): i})
        for j in range(t - 1, -1, -1):
            if k[j] < limit - (t - j):
                k[j] += 1
                for h in range(j + 1, t):
                    k[h] = k[h - 1] + 1
                break
            else:
                if j > 0 and k[j - 1] + 1 < k[j] - (t - j):
                    k[j - 1] = k[j - 1] + 1
                    for h in range(j, t):
                        k[h] = k[h - 1] + 1
                    break
    # print(position_table, len(position_table))
    return position_table, inv_position_table


# return 1d state with respect to state number
def num_to_state(n, rows, cols):
    state = [0] * rows * cols

    turn = n % 2
    n //= 2
    position_num = n % position_combinations[(rows, cols)]
    n //= position_combinations[(rows, cols)]
    side_num = n % side_combinations[(rows, cols)]
    n //= side_combinations[(rows, cols)]
    obstacles_num = n

    positions = piece_position_table[(rows, cols)][position_num]
    whites = side_position_table[(rows, cols)][side_num]
    blacks = [i for i in range(4) if i not in whites]

    state[positions[whites[0]]] = state[positions[whites[1]]] = 1
    state[positions[blacks[0]]] = state[positions[blacks[1]]] = -1

    for i in range(rows * cols):
        if state[i] in [-1, 1]:
            continue
        is_obstacle = obstacles_num % 2
        if is_obstacle == 1:
            state[i] = 2
        obstacles_num //= 2

    return state, turn


def state_to_num(state, turn, rows, cols, state_dim=2):
    if state_dim == 2:
        state = [i for j in state for i in j]  # flatten

    num = turn

    position = []
    for i in range(rows * cols):
        if state[i] in [-1, 1]:
            position.append(i)
    position_num = inv_piece_position_table[(rows, cols)][frozenset(position)]
    num += position_num * 2

    side = []
    tmp = 0
    for i in range(rows * cols):
        if state[i] == 1:  # from the perspective of white
            side.append(tmp)
        if state[i] in [-1, 1]:
            tmp += 1
    side_num = inv_side_position_table[(rows, cols)][frozenset(side)]
    num += side_num * 2 * position_combinations[(rows, cols)]

    obstacle_num = 0
    tmp = 1
    for i in range(rows * cols):
        if state[i] == 2:
            obstacle_num += tmp
        if state[i] not in [-1, 1]:
            tmp *= 2
    num += obstacle_num * 2 * position_combinations[(rows, cols)] * side_combinations[(rows, cols)]

    return num


def load_from_disk(rows, cols, file_name='NxNoptimal_bothsides_'):
    distance_to_win = np.load(f"model/{file_name}dtw_r{rows}_c{cols}.npy").tolist()
    parent = np.load(f"model/{file_name}par_r{rows}_c{cols}.npy").tolist()
    return distance_to_win, parent


for r, c in allowed_rows_cols:
    dtw, par = load_from_disk(r, c)
    print(f'loaded model: {r}x{c} {par[:5]} {dtw[:5]}')
    allowed_rows_cols[(r, c)]['dtw'] = dtw
    allowed_rows_cols[(r, c)]['par'] = par

    side_combinations[(r, c)] = comb(4, 2)
    position_combinations[(r, c)] = comb(r * c, 4)
    obstacles_combinations[(r, c)] = 2 ** (r * c - 4)

    piece_position_table[(r, c)], inv_piece_position_table[(r, c)] = \
        make_position_table(4, position_combinations[(r, c)], r * c)
    side_position_table[(r, c)], inv_side_position_table[(r, c)] = \
        make_position_table(2, side_combinations[(r, c)], 4)