纯Python实现机器学习算法13：Lasso回归

import numpy as npimport pandas as pd
data = np.genfromtxt('mystery.dat', delimiter = ',')# 选择特征与标签x = data[:,0:100] y = data[:,100].reshape(-1,1)# 加一列X = np.column_stack((np.ones((x.shape[0],1)),x))
# 划分训练集与测试集X_train, y_train = X[:70], y[:70]X_test, y_test = X[70:], y[70:]print(X_train.shape, y_train.shape, X_test.shape, y_test.shape)

# 定义参数初始化函数def initialize(dims):    w = np.zeros((dims, 1))    b = 0    return w, b

# 定义符号函数def sign(x):    if x > 0:        return 1    elif x < 0:        return -1    else:        return 0 # 利用numpy对符号函数进行向量化vec_sign = np.vectorize(sign)vec_sign(np.zeros((3,1))) 

# 定义lasso损失函数def l1_loss(X, y, w, b, alpha):    num_train = X.shape[0]    num_feature = X.shape[1]    y_hat = np.dot(X, w) + b    loss = np.sum((y_hat-y)**2)/num_train + np.sum(alpha*abs(w))    dw = np.dot(X.T, (y_hat-y)) /num_train + alpha * vec_sign(w)    db = np.sum((y_hat-y)) /num_train    return y_hat, loss, dw, db

# 定义训练过程def lasso_train(X, y, learning_rate=0.01, epochs=300):    loss_list = []    w, b = initialize(X.shape[1])    for i in range(1, epochs):        y_hat, loss, dw, db = l1_loss(X, y, w, b, 0.1)        w += -learning_rate * dw        b += -learning_rate * db        loss_list.append(loss)                if i % 50 == 0:            print('epoch %d loss %f' % (i, loss))        params = {            'w': w,            'b': b        }        grads = {            'dw': dw,            'db': db        }    return loss, loss_list, params, grads 

# 执行训练示例loss, loss_list, params, grads = lasso_train(X_train, y_train, 0.01, 500)

import numpy as npfrom sklearn.metrics import r2_score

class Lasso():    def __init__(self):        pass        def prepare_data(self):        data = np.genfromtxt('./example.dat', delimiter = ',')        x = data[:, 0:100]        y = data[:, 100].reshape(-1, 1)        X = np.column_stack((np.ones((x.shape[0], 1)), x))        X_train, y_train = X[:70], y[:70]        X_test, y_test = X[70:], y[70:]        return X_train, y_train, X_test, y_test        def initialize_params(self, dims):        w = np.zeros((dims, 1))        b = 0        return w, b
    def sign(self, x):        if x > 0:            return 1        elif x < 0:            return -1        else:            return 0            def l1_loss(self, X, y, w, b, alpha):        num_train = X.shape[0]        num_feature = X.shape[1]                y_hat = np.dot(X, w) + b        loss = np.sum((y_hat - y) ** 2) / num_train + np.sum(alpha*abs(w))        dw = np.dot(X.T, (y_hat - y)) / num_train + alpha*np.vectorize(self.sign)(w)        db = np.sum((y_hat - y)) / num_train        return y_hat, loss, dw, db        def lasso_train(self, X, y, learning_rate, epochs):        loss_list = []        w, b = self.initialize_params(X.shape[1])        for i in range(1, epochs):            y_hat, loss, dw, db = self.l1_loss(X, y, w, b, 0.1)            w += -learning_rate * dw            b += -learning_rate * db            loss_list.append(loss)                        if i % 300 == 0:                print('epoch %d loss %f' % (i, loss))                        params = {                'w': w,                'b': b            }            grads = {                'dw': dw,                'db': db            }        return loss, loss_list, params, grads            def predict(self, X, params):        w = params['w']        b = params['b']        y_pred = np.dot(X, w) + b        return y_pred    if __name__ == '__main__':    lasso = Lasso()    X_train, y_train, X_test, y_test = lasso.prepare_data()    loss, loss_list, params, grads = lasso.lasso_train(X_train, y_train, 0.01, 3000)    print(params)    y_pred = lasso.predict(X_test, params)    print(r2_score(y_test, y_pred))

# 导入线性模型模块from sklearn import linear_model# 创建lasso模型实例sk_lasso = linear_model.Lasso(alpha=0.1)# 对训练集进行拟合sk_lasso.fit(X_train, y_train)# 打印模型相关系数print("sklearn Lasso intercept :", sk_lasso.intercept_)print("\nsklearn Lasso coefficients :\n", sk_lasso.coef_)print("\nsklearn Lasso number of iterations :", sk_lasso.n_iter_)