手寫一個機器學習的入門算法-感知器算法

用4x+5y=2000做爲分界線製造了100個點；

初始分界線爲0,0；

通過1000輪糾正後，結果是：

22 x+31 y = 11876

對比結果4 x + 5 y = 2000

仍是比較接近的。

剛開始更新w的那行代碼搞錯了，覺得是用predict去糾正，其實應該用sample的真實值去糾正。

import random;

def find_split(points):
    w=(0,0,0)
    for _ in range(1,2000):
        print 'w='+str(w);
        for pt in points:
            (x1,x2,z) = pt;
            (w1,w2,w3)=w;
            predict = int((w1+w2*x1+w3*x2)>0)*2-1
            if predict!=z:
                print 'wrong: '+str(pt)
                w=(w1+z,w2+z*x1,w3+z*x2);
                # break;
            else:
                print 'right: '+str(pt)
    return w;

def test_split(points,w):
    points_2 = filter(lambda pt:((int(w[0]+w[1]*pt[0]+w[2]*pt[1])>=0)*2-1)==pt[2],points)
    return points_2;

def init_points(max_x,max_y,num_of_pts):
    points=[];
    for i in range(1,num_of_pts,1):
        x = int(random.random()*max_x);
        y = int(random.random()*max_y);
        z = int((4*x+5*y)>=2000)*2-1
        points.append((x,y,z));
    return points;

if __name__ == '__main__':    points = init_points(400,500,100);    print points;    line = find_split(points);    print(line);    pts = test_split(points,line);    print points;    print len(pts);