HEH/tools.py at main · Cherishnoobs/HEH · 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
138
139
140
141
import numpy as np
import settings
import torch
import matplotlib.pyplot as plt
from settings import a, b,c
from scipy.stats import norm
def build_G_from_S(S, k):
# def build_G_from_S(S, k):
    # S: similarity matrix
    # k: number of nearest neighbors
    # G: graph
    # G = torch.ones(S.shape).cuda() * -1.5
    # # G = torch.zeros(S.shape).cuda()
    # G_ = S
    # # G_ = S_
        # S: similarity matrix
    # k: number of nearest neighbors
    # G: graph
    G = torch.ones(S.shape).cuda() * -1.5
    # G = torch.zeros(S.shape).cuda()
    G_ = torch.where(S > settings.threshold, S, -1.5).cuda()
    # G_ = S_
    # S_ = torch.where(S > 0 and S > settings.rthreshold, 1, S).cuda()
    # S_ = torch.where(S < 0 and S < settings.lthreshold, -1, S_).cuda()
    # G_= torch.where(S_ != 0, S, 0).cuda()
    for i in range(G_.shape[0]):
        idx = torch.argsort(-G_[i])[:k]
        G[i][idx] = G_[i][idx]
    del G_
    torch.cuda.empty_cache()
    return G
    # # S_ = torch.where(S > 0 and S > settings.rthreshold, 1, S).cuda()
    # # S_ = torch.where(S < 0 and S < settings.lthreshold, -1, S_).cuda()
    # # G_= torch.where(S_ != 0, S, 0).cuda()
    # for i in range(G_.shape[0]):
    #     idx = torch.argsort(-G_[i])[:k]
    #     G[i][idx] = G_[i][idx]
    # del G_
    # torch.cuda.empty_cache()
    # return G
    G = torch.ones(S.shape).cuda() * -1.5
    # G = torch.zeros(S.shape).cuda()

    # G_ = torch.where(S > settings.threshold, S, -1.5).cuda()
    G_ = S.cuda()

    # G_ = torch.where(S==1, S, -1.5).cuda()
    # G = G_

    # G_ = S_
    # S_ = torch.where(S > 0 and S > settings.rthreshold, 1, S).cuda()
    # S_ = torch.where(S < 0 and S < settings.lthreshold, -1, S_).cuda()
    # G_= torch.where(S_ != 0, S, 0).cuda()
    for i in range(G_.shape[0]):
        idx = torch.argsort(-G_[i])[:k]
        G[i][idx] = G_[i][idx]
    del G_
    torch.cuda.empty_cache()
    return G

def generate_robust_S(s, alpha=2, beta=2):
    # G: graph
    # S: similarity matrix
    # alpha: parameter for postive robustness
    # beta: parameter for negative robustness
    # S_robust: robust similarity matrix

    S = s

    # find maximum count of cosine distance

    max_count = 0
    max_cos = 0

    interval = 1/1000
    cur = -1.0
    for i in range(2000):
        cur_cnt = np.sum((S>cur) & (S<cur+interval))
        if max_count < cur_cnt:
            max_count = cur_cnt
            max_cos = cur
        cur += interval

    # split positive and negative similarity matrix

    flat_S = S.reshape((-1,1))
    left = flat_S[np.where(flat_S <= max_cos)[0]]
    right = flat_S[np.where(flat_S >= max_cos)[0]]

    # reconstruct
    left = np.concatenate([left, 2 * max_cos - left])
    right = np.concatenate([max_cos-np.maximum(right-max_cos,max_cos-right), right])

    # fit to gaussian distribution
    left_mean, left_std = norm.fit(left)
    right_mean, right_std = norm.fit(right)

    print('left mean: ', left_mean)
    print('left std: ', left_std)
    print('threshold:',left_mean-alpha*left_std)
    print('right mean: ', right_mean)
    print('right std: ', right_std)
    print('threshold:',right_mean+beta*right_std)

    S = np.where(S >= right_mean + beta * right_std, 1, S)
    # S = np.where(S <= left_mean - alpha * left_std, -1, S)
    S = np.where(S <= left_mean - alpha * left_std, 0, S)
    # 不显示 y轴
    plt.gca().axes.get_yaxis().set_visible(False)
        # draw the histogram
    plt.hist(left, bins=10000, density=True, alpha=0.6, color='g')
    plt.savefig('left.svg', format='svg')
    plt.close()
    # x_major_locator=MultipleLocator(0.5)
    # ax=plt.gca()
    # ax.xaxis.set_major_locator(x_major_locator)
    plt.hist(right, bins=10000, density=True, alpha=0.6, color='r')
    plt.savefig('right.svg', format='svg')
    plt.close()
    # x_major_locator=MultipleLocator(0.5)
    # ax=plt.gca()
    # ax.xaxis.set_major_locator(x_major_locator)
    plt.hist(flat_S, bins=10000, density=True, alpha=0.6, color='b')
    plt.savefig('flat_S.svg', format='svg')
    plt.close()
    return S

    # generate robust similarity matrix

    # show the fitting result
    # plt.hist(left, bins=100, density=True, alpha=0.6, color='g')
    # plt.hist(right, bins=100, density=True, alpha=0.6, color='r')
    # plt.show()
    # plt.close()


def cal_sim(S_I, S_T, S_F):
    return settings.a * S_I + settings.b * S_T + settings.c * (S_F + S_F.t()) / 2