机器学习算法导论代码---lasso_cd

# Lasso回归问题的坐标下降算法
import numpy as np
class Lasso:
    def __init__(self,Lambda=1):
        self.Lambda=Lambda

    def soft_threshold(self,t,x):
        if x>t:
            return x-t
        elif x>=-t:
            return 0
        else:
            return x+t
    def fit(self,X,y,N=1000):
        m,n=X.shape
        alpha=2*np.sum(X**2,axis=0)/m
        w=np.zeros(n)
        for t in range(N):
            j=t%n
            w[j]=0
            e_j=X.dot(w.reshape(-1,1))-y
            beta_j=2*X[:,j].dot(e_j)/m
            w[j]=self.soft_threshold(self.Lambda/alpha[j],-beta_j/alpha[j])
        self.w=w
    def predict(self,X):
        return X.dot(self.w.reshape(-1,1))