ML实战：手动实现异常检测算法

• 这里使用了改良版本，即使用多元高斯分布函数，提高了对数据相关性的适应

代码实现

Detaction类

import numpy as np
from math import exp,pow,pi

np.set_printoptions(suppress=True)

class Detection:
    def __init__(self,x):
        '''
        :param x:训练集
        :param n:特征数
        :param u:每个特征的均值
        :param cov:协方差矩阵
        '''
        self.x=x
        self.n=len(x[0])

    def fit(self):
        #参数拟合
        m=len(self.x)
        self.u=np.sum(self.x,axis=0)/m
        temp=self.x-self.u
        self.cov=np.matmul(temp.T,temp)/m

    def predict(self,x):
        #预测函数
        cov=np.linalg.inv(self.cov)
        p=exp(((x-self.u)@cov@(x.T-self.u.T))*(-0.5))/pow(np.linalg.det(2*pi*self.cov),0.5)#多元高斯分布函数
        if p<10e-4:
            return 1
        else:
            return 0

主函数

from sklearn.datasets import make_blobs
import matplotlib.pyplot as plt
import numpy as np
import sys
from Detection_class import Detection
from sklearn.model_selection import train_test_split

np.set_printoptions(suppress=True)

color=['blue','red']
label=['Normal','Abnormal']

#生成数据，1000个正常样本，以及三类异常样本各10个
x, y = make_blobs(n_samples=[1000,10,10,10], n_features=2, random_state=np.random.randint(0,30))
x_postive=x[y==0,:]
x_error=x[y!=0,:]
x_train,x_test=train_test_split(x_postive,test_size=0.3,random_state=np.random.randint(0,30))
y_test=[0]*len(x_test)+[1]*len(x_error)
y_test=np.array(y_test)
x_test=np.r_[x_test,x_error]

#调用fit算法，实现参数拟合
detection=Detection(x_train)
detection.fit()

#处理测试集的预测结果
y_predict=[]
for i in x_test:
    y_predict.append(detection.predict(i))
y_predict=np.array(y_predict)
print(len(y_predict[y_predict==0]))

#实现结果可视化，与真实值对比
plt.figure(figsize=(12, 5), dpi=80)
for i in range(2):
    plt.scatter(x_test[y_test==i, 0], x_test[y_test==i,1],marker='o',s=8,c=color[i],label=label[i])
plt.title('Real Data')
plt.legend(loc=[1, 0])
plt.savefig('E:\python\ml\ml by myself\Abnormal_Detection\Abnormal_Detection_real.png')
for i in range(2):
    plt.scatter(x_test[y_predict==i, 0], x_test[y_predict==i,1],marker='o',s=8,c=color[i],label=label[i])
plt.title('Predict Result')
plt.savefig('E:\python\ml\ml by myself\Abnormal_Detection\Abnormal_Detection_predict.png')

sys.exit(0)

结果