l2的最优回归_L2范数惩罚项，高维线性回归

%matplotlib inlineimportmxnetfrom mxnet importnd,autogradfrom mxnet importgluon,initfrom mxnet.gluon importdata as gdata,loss as gloss,nnimportgluonbook as gb

n_train, n_test, num_inputs= 20,100,200true_w= nd.ones((num_inputs, 1)) * 0.01true_b= 0.05features= nd.random.normal(shape=(n_train+n_test, num_inputs))

labels= nd.dot(features,true_w) +true_b

labels+= nd.random.normal(scale=0.01, shape=labels.shape)

train_feature=features[:n_train,:]

test_feature=features[n_train:,:]

train_labels=labels[:n_train]

test_labels=labels[n_train:]#print(features,train_feature,test_feature)

#初始化模型参数

definit_params():

w= nd.random.normal(scale=1, shape=(num_inputs, 1))

b= nd.zeros(shape=(1,))

w.attach_grad()

b.attach_grad()return[w,b]#定义，训练，测试

batch_size= 1num_epochs= 100lr= 0.03train_iter= gdata.DataLoader(gdata.ArrayDataset(train_feature,train_labels),batch_size=batch_size,shuffle=True)#定义网络

deflinreg(X, w, b):return nd.dot(X,w) +b#损失函数

defsquared_loss(y_hat, y):"""Squared loss."""

return (y_hat - y.reshape(y_hat.shape)) ** 2 / 2

#L2 范数惩罚

defl2_penalty(w):return (w**2).sum() / 2

defsgd(params, lr, batch_size):for param inparams:

param[:]= param - lr * param.grad /batch_sizedeffit_and_plot(lambd):

w, b=init_params()

train_ls, test_ls=[], []for _ inrange(num_epochs):for X, y intrain_iter:

with autograd.record():#添加了 L2 范数惩罚项。

l = squared_loss(linreg(X, w, b), y) + lambd *l2_penalty(w)

l.backward()

sgd([w, b], lr, batch_size)

train_ls.append(squared_loss(linreg(train_feature, w, b),

train_labels).mean().asscalar())

test_ls.append(squared_loss(linreg(test_feature, w, b),

test_labels).mean().asscalar())

gb.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',

range(1, num_epochs + 1), test_ls, ['train', 'test'])print('L2 norm of w:', w.norm().asscalar())