Explainable_Federated_Learning/utilities_trustfed.py at main · Krishna-Rani-t/Explainable_Federated_Learning · 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
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164

import math
import torch
import numpy as np
from sklearn.metrics import roc_auc_score
from sklearn.metrics import classification_report, confusion_matrix
from sklearn.metrics import log_loss, brier_score_loss


def find_statistical_parity_score(protected_attr,labels,predictions):
        protected_pos = 0.
        protected_neg = 0.
        non_protected_pos = 0.
        non_protected_neg = 0.

        saValue = 0
        for i in range(len(protected_attr)):
            # protected population
            if protected_attr[i] == saValue:
                if predictions[i] == 1:
                    protected_pos += 1.
                else:
                    protected_neg += 1.
                # correctly classified


            else:
                if predictions[i] == 1:
                    non_protected_pos += 1.
                else:
                    non_protected_neg += 1.


        if((protected_pos + protected_neg) == 0):
            C_prot = 0
        else:
            C_prot = (protected_pos) / (protected_pos + protected_neg)
        if((non_protected_pos + non_protected_neg) == 0):
            C_non_prot =0
        else:
            C_non_prot = (non_protected_pos) / (non_protected_pos + non_protected_neg)

        stat_par = C_non_prot - C_prot
        return stat_par

def find_eqop_score(protected_attr,labels,predictions):
        protected_pos = 0.
        protected_neg = 0.
        non_protected_pos = 0.
        non_protected_neg = 0.

        tp_protected = 0.
        tn_protected = 0.
        fp_protected = 0.
        fn_protected = 0.

        tp_non_protected = 0.
        tn_non_protected = 0.
        fp_non_protected = 0.
        fn_non_protected = 0.

        saValue = 0
        for idx in range(len(protected_attr)):
            # protrcted population
            if protected_attr[idx] == saValue:
                if predictions[idx] == 1:
                    protected_pos += 1.
                else:
                    protected_neg += 1.

                # correctly classified
                if labels[idx] == predictions[idx]:
                    if labels[idx] == 1:
                        tp_protected += 1.
                    else:
                        tn_protected += 1.
                # misclassified
                else:
                    if labels[idx] == 1:
                        fn_protected += 1.
                    else:
                        fp_protected += 1.

            else:
                if predictions[idx] == 1:
                    non_protected_pos += 1.
                else:
                    non_protected_neg += 1.

                # correctly classified
                if labels[idx] == predictions[idx]:
                    if labels[idx] == 1:
                        tp_non_protected += 1.
                    else:
                        tn_non_protected += 1.
                # misclassified
                else:
                    if labels[idx] == 1:
                        fn_non_protected += 1.
                    else:
                        fp_non_protected += 1.

        if((tp_protected + fn_protected)==0):
            tpr_protected = 0
        else:
            tpr_protected = tp_protected / (tp_protected + fn_protected)
        #tnr_protected = tn_protected / (tn_protected + fp_protected)

        if((tp_non_protected + fn_non_protected) == 0):
            tpr_non_protected=0
        else:
            tpr_non_protected = tp_non_protected / (tp_non_protected + fn_non_protected)
        #tnr_non_protected = tn_non_protected / (tn_non_protected + fp_non_protected)


        eqop = tpr_non_protected - tpr_protected
        return eqop


# Evaluation of the global model on the test data
def all_metrics(y_true,y_pre):
        #print(y_true)
        #p#rint(y_pre)
        conf = (confusion_matrix(y_true,y_pre.round()))
        TN = conf[0][0]
        FP = conf[0][1]
        FN = conf[1][0]
        TP = conf[1][1]
        #print(TN)
        #print(FP)
        #print(FN)
        #print(TP)
        sensitivity = TP/(TP+FN)
        specificity = TN/(FP+TN)
        BalanceACC = (sensitivity+specificity)/2
        G_mean = math.sqrt(sensitivity*specificity)
        FN_rate= FN/(FN+TP)
        FP_rate = FP/(FP+TN)
        Precision = TP/(TP+FP)
        f1_sc = 2*(sensitivity * Precision) / (sensitivity + Precision)
        acc = (TP+TN)/(TP+TN+FN+FP)
        auc = roc_auc_score(y_true, y_pre)
        return sensitivity,specificity,BalanceACC,G_mean,FN_rate,FP_rate,Precision,f1_sc,acc, auc

def find_class_weights(labels,majority_label=0,minority_label=1):
        unique, counts = np.unique(labels, return_counts=True)
        count_ap_dict = dict(zip(unique, counts))

        majority_class_weight = 1
        minority_class_weight = count_ap_dict.get(majority_label,0)/count_ap_dict.get(minority_label,1)
        #class_weights={majority_label:1,minority_label:minority_class_weight}
        #class_weights = [1,minority_class_weight]
        class_weights = []
        for i in range(len(labels)):
            if labels[i]==minority_label:
                class_weights.append(minority_class_weight)
            else:
                class_weights.append(1)
        t = torch.tensor(class_weights)
        return t