感知器算法的设计实现 python

1.数据生成及规范化处理
利用高斯模型，生成 N 类（N＞5）数据（2D or 3D），并对生成样本进行规范化处理
mu = np.array([[random.randint(5,95), random.randint(5,95)]])
Sigma = np.array([[8, 5], [3, 2]])
R = cholesky(Sigma)
s = np.dot(np.random.randn(6, 2), R) + mu

上图即为所有的分界面的图形，但由于无法去除无用的分类面，只能把所有n^2个分类面一一输出
3.生成测试数据列，并对测试数据进行分类判别。

# -*- coding: utf-8 -*-

import random
import copy
import numpy as np
import matplotlib.pyplot as plt
from numpy.linalg import cholesky
import sys

if __name__ == '__main__':
  
    base = 10.0
    data = []
    alldata = []
    numofN=6

    num = 0
    while num < numofN:
   
        mu = np.array([[random.randint(5,95), random.randint(5,95)]])#生成随机数据
        Sigma = np.array([[8, 5], [3, 2]])
        R = cholesky(Sigma)
        s = np.dot(np.random.randn(6, 2), R) + mu
   
        key = False
        for i in s.tolist():
            for j in alldata:
                if abs(i[0]-j[0]) + abs(i[1]-j[1]) <=10:
                    key = True
                    break
        if key:#随机数据碰撞则重来
            continue
        num += 1
   
        plt.plot(s[:,0],s[:,1],'*',markersize=10) 
   
        x = []
        for i in s :
            x.append(np.array(np.hstack((np.array(i),np.array([base])))))
            
        data.append(np.array(x))
   
        for i in s.tolist():
            alldata.append(i)

  
    wlist = []
    for i in range(0,numofN) :
        wlist.append(np.array([base,base,base]))#初始权向量
    
    ccc=0
    while True :
        time = 0
    
        for i in range(0,numofN):
            for yi in data[i]:
                for t in range(0,numofN):
                    if t == i:
                        continue
           
                    while np.dot(yi,wlist[i]) <= np.dot(yi,wlist[t]) + 0.5:#修正
                        wlist[i] += np.array(yi)
                        wlist[t] -= np.array(yi)
                        time += 1
                        
                        
            
    
    
        if time == 0:
            break

    for i in wlist:
        for j in wlist:
            if (i==j).all() :
                continue
            linex = np.array([0.0,100.0])
            plt.plot(linex,(((-1)*linex*(i[0]-j[0])-base*(i[2]-j[2]))/(i[1]-j[1])),alpha=0.5) #画分割线
            
   
    plt.xlim(0,100)
    plt.ylim(0,100)
    plt.show()
  
    plt.cla()
         
    

    picx = []
    picy = []
    for i in range(0,numofN):
        picx.append([])
        picy.append([])
        
    for i in range(0,100):
        for j in range(0,100):
            
            maxx = []
            for t in wlist:
                maxx.append(np.dot(np.array([i,j,base]),t))
            
            k = maxx.index(max(maxx))
            picx[k].append(i)
            picy[k].append(j)
    
    
   
        for i in data :
            plt.plot(i[:,0],i[:,1],'*',markersize=10) #
    for i in range(0,numofN):
        plt.plot(picx[i],picy[i],'.',alpha=0.3) #画出各个位置上的预测结果