深度学习笔记（5）---自学《动手学深度学习》----多项式函数拟合实验 内附代码片段及解释

多项式函数拟合实验

注：t.my_ones_packages是我自己根据《动手学深度学习》目前学习过程中出现的所有自定义函数进行集中得到的文档。
《动手学深度学习》也向读者提供了一个包含所有自定义函数的包“d2lzh”大家可以自行下载 侵删 链接如下 :link. 提取码: un5p 时间设置是永久的 如果失效可以联系我更新

import t.my_ones_packages as mop
from mxnet import autograd, gluon, nd
from mxnet.gluon import data as gdata, loss as gloss, nn

# 生成数据集
num_train, num_test, true_w, true_b = 100, 100, [1.2, -3.4, 5.6], 5  # 设置输入个数、输出个数真实的w和真实的b的值
features = nd.random.normal(shape=(num_train + num_test, 1))  # 设置特征为 随机的 200*1形状的矩阵
ploy_feature = nd.concat(features, nd.power(features, 2), nd.power(features, 3))  # 拼接三个矩阵


# 对应 w0*x0+w1*x1+w2*x2+b
labels = (true_w[0] * ploy_feature[:, 0] + true_w[1] * ploy_feature[:, 1] + true_w[2] * ploy_feature[:, 2] + true_b)
labels += nd.random.normal(scale=0.1, shape=labels.shape)


# 绘图函数
def semilogy(x_val, y_val, x_lab, y_lab, x2_val=None, y2_val=None, legend=None, figsize=(7.5, 5.5)):
    mop.set_figsize(figsize)
    mop.plt.xlabel(x_lab)  # x方向的数据来源
    mop.plt.ylabel(y_lab)  # y方向额数据来源
    mop.plt.semilogy(x_val, y_val)  # 递归调用自己

    if x2_val and y2_val:
        mop.plt.semilogy(x2_val, y2_val, linestyle=':')  # 递归调用自己
        # legend(loc,    frameon,       edgecolor,     facecolor,              title,     ['',''])
        # legend(图例位置，是否去掉图例边框，设置图例边框颜色，设置图例背景(若无边框，无效),设置图例标题，设置图例内的分类名字)
        mop.plt.legend(legend)

    mop.plt.show()  # 显示图像


num_epochs = 100
loss = gloss.L2Loss()  # 定义L2范数损失


def fit_and_plot(train_features, test_features, train_labels, test_labels):
    net = nn.Sequential()  # Sequential是容器
    net.add(nn.Dense(1))  # 添加输出层，输出层单元个数为1
    net.initialize()  # 对net初始化
    batch_size = min(10, train_labels.shape[0])
    train_iter = gdata.DataLoader(gdata.ArrayDataset(train_features, train_labels), batch_size, shuffle=True)  # shuffle对应是否打乱数据
    trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': 0.01})

    train_ls = []
    test_ls = []

    for _ in range(num_epochs):
        for x, y in train_iter:

            with autograd.record():
                l = loss(net(x), y)  # l是有关net(x)和y的小批量的损失
            l.backward()  # 使用反向传播自动计算梯度
            trainer.step(batch_size)  # 仅保存参数和超级参数 ，根据batch_size进行迭代，数据更新


        # append(x)函数是在列表最后加上x
        train_ls.append(loss(net(train_features), train_labels).mean().asscalar())
        test_ls.append(loss(net(test_features), test_labels).mean().asscalar())

    print('final epoch :train loss', train_ls[-1], 'test loss', test_ls[-1])

    semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
             range(1, num_epochs + 1), test_ls, ['train', 'test'])

    print('weight', net[0].weight.data().asnumpy(), '\nbias', net[0].bias.data().asnumpy())


# 三项多项式函数拟合

fit_and_plot(ploy_feature[:num_train, :], ploy_feature[num_train:, :], labels[:num_train], labels[num_train:])


# 线性函数拟合(欠拟合)
'''
fit_and_plot(features[:num_train, :], features[num_train:, :], labels[:num_train], labels[num_train:])
'''
# 训练样本不足(过拟合)
'''
fit_and_plot(ploy_feature[0:2, :], ploy_feature[num_train:, :], labels[0:2], labels[num_train:])
'''