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
|
import pandas as pd
import numpy as np
# 数据加载和预处理
data = pd.read_csv("iris.txt", header=None)
name_list = ["setosa", "versicolor", "virginica"]
# 标签编码
def safe_index(x, name_list):
try:
return name_list.index(x)
except ValueError:
return -1
def data_process(X,y):
index=np.arange(0,len(y))
np.random.shuffle(index)
return X[index,:],y[index]
def split_dataset(X, y, train_ratio=0.8, random_state=None):
if random_state is not None:
np.random.seed(random_state)
n_samples = X.shape[0]
indices = np.arange(n_samples)
np.random.shuffle(indices)
X_shuffled = X[indices]
y_shuffled = y[indices]
train_size = int(train_ratio * n_samples)
X_train, X_predict = X_shuffled[:train_size], X_shuffled[train_size:]
y_train, y_predict = y_shuffled[:train_size], y_shuffled[train_size:]
return X_train/10, y_train, X_predict/10, y_predict
data[4] = data[4].apply(lambda x: safe_index(x, name_list))
data = data[data[4] != -1] # 过滤掉无效数据
print(data)
# 提取特征和标签
X = data.iloc[:, :-1].values
y = data.iloc[:, -1].values
y = np.where(y == 0, -1, 1) # 将标签转换为 -1 和 1
x_train, y_train, x_test, y_test = split_dataset(X, y)
class SVM:
def __init__(self,x_train,y_train,x_test,y_test,C=1,max_iter=100,tol=0.001):
self.X=x_train
self.y=y_train
self.C=C
self.max_iter=max_iter
self.a=np.zeros_like(self.y)
self.b=0
self.K_matrix=self.compute_k_matrix()
self.tol=tol
self.X_test=x_test
self.y_test=y_test
def compute_k_matrix(self):
n_samples=self.X.shape[0]
K=np.zeros((n_samples,n_samples))
for i in range(n_samples):
K[i,:]=np.exp(-np.linalg.norm(self.X[i,:]-self.X,axis=1))
return K
def f(self,index):
K=self.K_matrix[index,:]
return np.sum(self.a*self.y*K)+self.b
def choose_index(self):
indices=np.where((self.a>0) & (self.a<self.C))[0]
if len(indices)==0:
indices=np.arange(len(self.y))
index1=np.random.choice(indices)
E=np.array([self.compute_E(i) for i in range(len(self.y))])
delta_E=np.abs(E-self.compute_E(index1))
index2=np.argmax(delta_E)
return index1,index2
def compute_E(self, index):
return self.f(index) - self.y[index]
def smo(self):
iter_num=0
while iter_num<self.max_iter:
max_inner_iter = 100
inner_iter = 0
while inner_iter<max_inner_iter:
index1,index2=self.choose_index()
a1_old,a2_old=self.a[index1],self.a[index2]
y1,y2=self.y[index1],self.y[index2]
if y1!=y2:
L=max(0,a2_old-a1_old)
H=min(self.C,self.C+a2_old-a1_old)
else:
L = max(0, a1_old + a2_old - self.C)
H = min(self.C, a1_old + a2_old)
eta=self.K_matrix[index1,index1]+self.K_matrix[index2,index2]-2*self.K_matrix[index1,index2]
if eta <= 1e-10:
continue
a2_new=self.a[index2]+self.y[index2]*(self.compute_E(index1)-self.compute_E(index2))/eta
if L == H:
continue
a2_new=np.clip(a2_new,L,H)
a1_new=self.a[index1]+self.y[index1]*self.y[index2]*(self.a[index2]-a2_new)
self.a[index1],self.a[index2]=a1_new,a2_new
b1 = self.b - self.compute_E(index1) \
- self.y[index1] * (a1_new - a1_old) * self.K_matrix[index1, index1] \
- self.y[index2] * (a2_new - a2_old) * self.K_matrix[index1, index2]
b2 = self.b - self.compute_E(index2) \
- self.y[index1] * (a1_new - a1_old) * self.K_matrix[index1, index2] \
- self.y[index2] * (a2_new - a2_old) * self.K_matrix[index2, index2]
if 0 < a1_new < self.C:
self.b = b1
elif 0 < a2_new < self.C:
self.b = b2
else:
self.b = (b1 + b2) / 2
inner_iter+=1
if np.abs(a1_new - a1_old)<self.tol and np.abs(a2_new - a2_old)<self.tol:
break
if iter_num%10==0:
print(f"第{iter_num}轮,预测准确率为{round(self.calculate_accuracy(), 3)}")
print(f"L:{L},H:{H},eta:{eta}")
print(f"a[:20]:{self.a[:20]},b:{self.b}")
iter_num+=1
def predict(self,X):
K=np.exp(-np.linalg.norm(self.X-X,axis=1))
if np.sum(self.a*self.y*K)+self.b>=0:
return 1
else:
return -1
def calculate_accuracy(self):
correct_predictions = 0
n_samples = len(self.y_test)
for i in range(n_samples):
y_pred = self.predict(self.X_test[i]) # 对单个样本预测
if y_pred == y_test[i]:
correct_predictions += 1
accuracy = correct_predictions / n_samples
return accuracy
if __name__=="__main__":
print(x_train)
svm=SVM(x_train,y_train,x_test,y_test)
svm.smo()
y_pred=np.zeros_like(y_test)
for i in range(len(svm.X_test)):
y_pred[i]=svm.predict(svm.X_test[i,:])
print(f"预测值{y_pred}")
print(f"真实值{y_test}")
|
