AAAI2022-HCM/utils.py at main · DAMO-DI-ML/AAAI2022-HCM · 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
import numpy as np
from sklearn.preprocessing import OneHotEncoder
from astropy import stats


def normalize_biweight(x, eps=1e-10):
    median = np.median(x)
    scale = stats.biweight.biweight_scale(x)
    if np.std(x) < 1e+2 or np.isnan(scale) or scale < 1e-4:
        norm =  (x-np.mean(x))/np.std(x)
    else:
        norm = (x - median) / (scale + eps)
    return norm


def normalize(x):
    norm = lambda x: (x-np.mean(x))/np.std(x)
    return np.apply_along_axis(norm, 0, x)


def data_preprocess(df):

    def _encoding(i):
        if df.iloc[:,i].dtype == 'O' or df.iloc[:, i].dtype.name == 'category':
            tempX = df.iloc[:, i].values.reshape(-1, 1)
            enc = OneHotEncoder(handle_unknown='ignore')
            enc.fit(tempX)
            out = enc.transform(tempX).toarray()
        else:
            out = df.iloc[:, i].values.reshape(-1, 1)
        return out
    p = df.shape[1]
    X_encode = [_encoding(i) for i in np.arange(p)]
    return X_encode


def evaluate_binary(trueG, estG):
    TP, TN, FP, FN, FD, MD, FPMD = 0, 0, 0, 0, 0, 0, 0
    n_node = trueG.shape[0]
    for i in range(1, n_node):
        for j in range(i):
            if trueG[i, j] == 1 and trueG[j, i] == 0 and estG[i, j] == 1 and \
                    estG[j, i] == 0:
                TP += 1
            if trueG[i, j] == 0 and trueG[j, i] == 1 and estG[i, j] == 0 and \
                    estG[j, i] == 1:
                TP += 1
            if trueG[i, j] == 0 and trueG[j, i] == 0 and estG[i, j] == 0 and \
                    estG[j, i] == 0:
                TN += 1
            if trueG[i, j] == 0 and trueG[j, i] == 0 and estG[i, j] == 1 and \
                    estG[j, i] == 0:
                FP += 1
            if trueG[i, j] == 0 and trueG[j, i] == 0 and estG[i, j] == 0 and \
                    estG[j, i] == 1:
                FP += 1
            if trueG[i, j] == 1 and trueG[j, i] == 0 and estG[i, j] == 0 and \
                    estG[j, i] == 0:
                FN += 1
            if trueG[i, j] == 0 and trueG[j, i] == 1 and estG[i, j] == 0 and \
                    estG[j, i] == 0:
                FN += 1
            if trueG[i, j] == 1 and trueG[j, i] == 0 and estG[i, j] == 0 and \
                    estG[j, i] == 1:
                FD += 1
            if trueG[i, j] == 0 and trueG[j, i] == 1 and estG[i, j] == 1 and \
                    estG[j, i] == 0:
                FD += 1
            if trueG[i, j] == 0 and trueG[j, i] == 1 and estG[i, j] == 1 and \
                    estG[j, i] == 1:
                MD += 1
            if trueG[i, j] == 1 and trueG[j, i] == 0 and estG[i, j] == 1 and \
                    estG[j, i] == 1:
                MD += 1
            if trueG[i, j] == 0 and trueG[j, i] == 0 and estG[i, j] == 1 and \
                    estG[j, i] == 1:
                FPMD += 1
    if (TP + FP + FD)>0:
        Precision = TP / (TP + FP + FD)
    else:
        Precision = 0.0
    Recall = TP / sum(sum(trueG))

    if (TP + FN + FD) > 0:
        Recall1 = TP / (TP + FN + FD)
    else:
        Recall1 = 0.0
    SHD = sum(sum((trueG != estG) | np.transpose((trueG != estG)))) / 2
    return {'TP': TP, 'TN': TN, 'FP': FP, 'FN': FN, 'FD': FD, 'MD': MD,
            'FPMD': FPMD,'Precision': Precision, 'Recall': Recall,
            'Recall_NOMD': Recall1, 'SHD': SHD}


def skeleton_metrics(trueG, estG):
    TP,TN,FP,FN = 0,0,0,0
    n = trueG.shape[0]
    for i in range(n):
        for j in range(i):
            if trueG[i, j] == 1 or trueG[j, i] == 1:
                if estG[i, j] != 0 or estG[j, i] != 0:
                    TP += 1
                else:
                    FN += 1
            else:
                if estG[i, j] == 0 and estG[j, i] == 0:
                    TN += 1
                else:
                    FP += 1
    return {'TP': TP, 'TN': TN, 'FP': FP, 'FN': FN}


def check_connect_skel(skeleton, i, j):
    depth = set.union(set(np.where(skeleton[:,i]==1)[0]),
                      set(np.where(skeleton[i,:]==1)[0]))
    checked = depth
    while depth:
        if j in depth:
            return True
        next = {}
        for k in depth:
            next = set.union(next, set.union(set(np.where(skeleton[:,k]==1)[0]),
                                             set(np.where(skeleton[k,:]==1)[0])))
        depth = set.difference(next, checked)
        checked = set.union(checked, depth)
    return False


def reachable(dag, fr, to):
    depth = set(np.where(dag[fr,:]==1)[0])
    checked = depth
    while depth:
        if to in depth:
            return True
        next = set()
        for k in depth:
            next = set.union(next, set(np.where(dag[k,:]==1)[0]))
        depth = set.difference(next, checked)
        checked = set.union(checked, depth)
    return False