感知机学习算法 python实现

参考李航《统计学习方法》 一开始的感知机章节，看着不太复杂就实现一下。。。

"""
感知机学习算法的原始形式
例2.1
"""
import numpy as np

class Perceptron:
    def __init__(self,w,b,alpha):
        self.w = w
        self.b = b
        self.alpha = alpha

    def loss(self,x,y):
        return np.sum( y*(np.dot(x, self.w) + self.b) )
    
    def sgd(self,x,y):  # 随机梯度下降函数
        self.w += self.alpha * y * x
        self.b += self.alpha * y

    def train(self,X,Y):
        while(True):
            M = len(X)   # 错误分类数
            for i in range(len(X)):
                if self.loss(X[i],Y[i])<=0:
                    self.sgd(X[i],Y[i])
                    print "w:",self.w," b:",self.b
                else:
                    M -= 1
            if not M:
                print "final optimal:","w:",self.w," b:",self.b
                break

class Perceptron_dual:
    def __init__(self,alpha,b,ita):
        self.alpha = alpha
        self.b = b
        self.ita = ita

    def gram(self,X):
        return np.dot(X,X.T)

    def train(self,X,Y):
        g = self.gram(X)
        
        M = len(X)   # 错误分类数
        while(True):
            M = len(X)   # 错误分类数
            for j in range(len(X)):
                if Y[j] * (np.sum(self.alpha * Y * g[j]) + self.b) <= 0:
                    self.alpha[j] += self.ita
                    self.b += self.ita * Y[j]
                    print "a:",self.alpha," b:",self.b
                else:
                    M -= 1
            if M == 0:
                print "final optimal:","a:",self.alpha," b:",self.b
                break

if __name__ == "__main__":

    X = np.array([[3,3],[4,3],[1,1]])
    
    Y = np.array([1,1,-1])
    perc_d = Perceptron_dual(np.zeros(Y.shape),0,1)
    perc_d.train(X, Y)