-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathhesslr.py
211 lines (172 loc) · 6.3 KB
/
hesslr.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
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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
from phe import paillier
from sklearn.model_selection import train_test_split
import numpy as np
import pandas as pd
import random
import time
import sklearn.metrics as metrics
import matplotlib.pyplot as plt
import ass
import ssm
def normalization(data):
mu = data.mean(axis=0)
std = data.std(axis=0)
return (data - mu) / std
def guiyihua(data):
_range = np.max(data) - np.min(data)
return (data - np.min(data)) / _range
def load_data(file_name):
df = pd.read_csv(file_name)
# diabetes 8*features
fg = df.iloc[:, :4].to_numpy()
fh = df.iloc[:, 4:-1].to_numpy()
# breast
# fg = df.iloc[:, :10].to_numpy()
# fh = df.iloc[:, 10:-1].to_numpy()
fg = normalization(fg)
fh = normalization(fh)
# print("fg:",fg)
ones = np.ones(shape=fg.shape[0])
fg = np.c_[fg, ones]
# print("features:", features[0])
# print('fixed features shape: ', features_g.shape)
fg_train,fg_test,fh_train,fh_test=train_test_split(fg,fh,test_size=0.3,random_state=1)
labels = np.squeeze(df.iloc[:, -1].to_numpy().reshape(1, -1))
# labels = np.squeeze(df.iloc[:, -1:].to_numpy().reshape(1, -1))
# labels = normalization(labels)
# labels = labels*2-1
labels_train,labels_test = train_test_split(labels,test_size=0.3,random_state=1)
# print('labels shape: ', labels.shape)
return fg_test,fg_train, fh_test,fh_train, labels_test,labels_train
def data_iter(batch_size, x1, x2, y):
num_examples = len(y)
indices = list(range(num_examples))
np.random.shuffle(indices)
for i in range(0, num_examples,batch_size):
# batch_indices = indices[i:i+batch_size]
batch_indices = indices[i:i+batch_size]
yield x1[batch_indices], x2[batch_indices], y[batch_indices]
def sigmoid(x):
return 1.0 / (1.0 + np.exp(-x))
def compute_loss(X1,X2, y, w11,w12,w21,w22):
# w1 = w11+w12
# w2 = w21+w22
z = np.dot(X1,w11)+np.dot(X1,w12)+np.dot(X2,w21)+np.dot(X2,w22)
# y_hat = 0.5+0.125*z
y_hat = sigmoid(z)
# print("y_hat:",y_hat)
loss = -np.sum(y * np.log(y_hat) + (1 - y) * np.log(1 - y_hat+1e-5))
# loss = np.sum(np.log(2)-1/2*y*wx+1/8*wx*wx)
loss /= len(X1)
print("loss:",loss)
return loss
# x.shape=(batch,n)
def compute_gradient(pk,sk,X1,X2, y, w11,w12,w21,w22):
z11 = np.dot(X1,w11)
# print("w11",w11)
# print("w12",w12)
# print("w21",w21)
# print("w22",w22)
z121,z122 = ssm.ssm(pk,sk,X1,w12)
z22 = np.dot(X2,w22)
z211,z212 = ssm.ssm(pk,sk,X2,w21)
z1 = z11+z121+z211
z1_e = np.asarray([pk.encrypt(m) for m in z1])
# z1_e2 = np.asarray([pk.encrypt(m) for m in z1*z1])
# z1_e3 = np.asarray([pk.encrypt(m) for m in z1*z1*z1])
z2 = z22+z122+z212
z = z1_e+z2
# z_3e = z1_e3+3*z1_e2*z2+3*z1_e*z2*z2+z2*z2*z2
# y_hat = 0.5+0.197*z-0.004*z_3e
y_hat = 0.5+0.125*z
e = y_hat-y
sk_y_hat = np.asarray([sk.decrypt(m) for m in y_hat])
y1,y2 = ass.asslist(sk_y_hat,len(sk_y_hat))
e1 = y1
e2 = y2-y
g11 = np.dot(X1.T,e1)
tmp = np.dot(X1.T,e2)
g121,g122 = ass.asslist(tmp,len(tmp))
grad2 = np.dot(X2.T,e)
sk_g2 = np.asarray([sk.decrypt(m) for m in grad2])
g21,g22 = ass.asslist(sk_g2,len(sk_g2))
n = len(X1)
return g11/n,g121/n,g122/n,g21/n,g22/n
def fit(X1,X2, y,fg_test,fh_test,labels_test):
print('fit start')
np.random.seed(1)
losslist=[]
acclist=[]
auclist=[]
w1 = np.ones(X1.shape[1])
w2 = np.ones(X2.shape[1])
w11,w12 = ass.asslist(w1,len(w1))
w21,w22 = ass.asslist(w2,len(w2))
pk,sk = paillier.generate_paillier_keypair(n_length=1024)
batch_size = 64
learning_rate = 0.1
iter_max = 30
oldloss = 0
for n_iter in range(1, iter_max+1):
# compute loss
loss = compute_loss(X1,X2, y, w11,w12,w21,w22)
losslist.append(loss)
# print(f'current loss: {loss}')
# if abs(loss-oldloss) <= 1e-5:
# print(f'loss <= 1e-5, fit finish')
# break
oldloss = loss
for (batch_X1,batch_X2, batch_y) in data_iter(batch_size, X1,X2, y):
g11,g121,g122,g21,g22 = compute_gradient(pk,sk,batch_X1,batch_X2, batch_y, w11,w12,w21,w22)
# print("length of batch_X:",n_iter,"----",batch_X.shape[0])
w11 -= learning_rate * (g11+g121)
w12 -= learning_rate * g122
w21 -= learning_rate * g21
w22 -= learning_rate * g22
# print("w11",w11)
# print("w12",w12)
# print("w21",w21)
# print("w22",w22)
print("current iter:",n_iter)
acc,predlist = predict(fg_test, fh_test, labels_test, w11,w12,w21,w22)
acclist.append(acc)
fpr, tpr, thresholds = metrics.roc_curve(labels_test,predlist)
auc = metrics.auc(fpr, tpr)
# print(auc)
auclist.append(auc)
# print("w1:",w1)
# print("w2:",w2)
return w11,w12,w21,w22,losslist,acclist,auclist
def predict(X1,X2, y, w11,w12,w21,w22):
count = 0
w1 = w11+w12
w2 = w21+w22
pred = sigmoid(np.dot(X1, w1)+np.dot(X2, w2))
count = sum((pred > 0.5)*1 == y)
# count = sum((pred > 0.5)*1 == (y+1)/2)
print("count", count)
return 100 * count / len(y),pred
if __name__ == '__main__':
fg_test,fg_train, fh_test,fh_train, labels_test,labels_train = load_data('diabetes.csv')
t1 = time.time()
w11,w12,w21,w22,losslist,acclist,auclist = fit(fg_train, fh_train,labels_train,fg_test,fh_test,labels_test)
print("w1:",w11+w12)
print("w2:",w21+w22)
print(f'cost：{time.time()-t1:.3f}s')
# predict_result = predict(fg_test, fh_test,labels_test, w1,w2)
# print(f'predict_result: {predict_result}%')
print("losslist:", losslist)
print("acclsit:", acclist)
print("auclist:", auclist)
plt.plot(np.linspace(0, len(losslist), len(losslist)), losslist)
plt.ylabel('loss')
plt.xlabel('iter')
plt.show()
plt.plot(np.linspace(0, len(acclist), len(acclist)), acclist)
plt.ylabel('acc')
plt.xlabel('iter')
plt.show()
plt.plot(np.linspace(0, len(auclist), len(auclist)), auclist)
plt.ylabel('auc')
plt.xlabel('iter')
plt.show()