小白学习之pytorch框架(7)之实战Kaggle比赛：房价预测(K折交叉验证、*args、**kwargs)

本篇博客代码来自于《动手学深度学习》pytorch版，也是代码较多，解释较少的一篇。不过好多方法在我以前的博客都有提，所以这次没提。还有一个原因是，这篇博客的代码，只要好好看看肯定能看懂(前提是python语法大概了解)，这是我不加很多解释的重要原因。

K折交叉验证实现

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32

def get_k_fold_data(k, i, X, y):     # 返回第i折交叉验证时所需要的训练和验证数据，分开放，X_train为训练数据，X_valid为验证数据     assert k > 1     fold_size = X.shape[0] // k  # 双斜杠表示除完后再向下取整     X_train, y_train = None, None     for j in range(k):         idx = slice(j * fold_size, (j + 1) * fold_size)  #slice(start,end,step)切片函数         X_part, y_part = X[idx, :], y[idx]         if j == i:             X_valid, y_valid = X_part, y_part         elif X_train is None:             X_train, y_train = X_part, y_part         else:             X_train = torch.cat((X_train, X_part), dim=0) #dim=0增加行数，竖着连接             y_train = torch.cat((y_train, y_part), dim=0)     return X_train, y_train, X_valid, y_valid   def k_fold(k, X_train, y_train, num_epochs,learning_rate, weight_decay, batch_size):     train_l_sum, valid_l_sum = 0, 0     for i in range(k):         data = get_k_fold_data(k, i, X_train, y_train) # 获取k折交叉验证的训练和验证数据         net = get_net(X_train.shape[1])  #get_net在这是一个基本的线性回归模型，方法实现见附录1         train_ls, valid_ls = train(net, *data, num_epochs, learning_rate,                                    weight_decay, batch_size)  #train方法见后面附录2         train_l_sum += train_ls[-1]         valid_l_sum += valid_ls[-1]         if i == 0:             d2l.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'rmse',                          range(1, num_epochs + 1), valid_ls,                          ['train', 'valid'])   #画图，且是对y求对数了，x未变。方法实现见附录3         print('fold %d, train rmse %f, valid rmse %f' % (i, train_ls[-1], valid_ls[-1]))     return train_l_sum / k, valid_l_sum / k

 *args：表示接受任意长度的参数，然后存放入一个元组中；如def fun(*args) print(args)，‘fruit'，'animal'，'human'作为参数传进去，输出(‘fruit'，'animal'，'human')

**kwargs：表示接受任意长的参数，然后存放入一个字典中；如

?

1 2 3

def fun(**kwargs):       for key, value in kwargs.items():         print("%s:%s" % (key,value)

fun(a=1,b=2,c=3)会输出 a=1 b=2 c=3

附录1

?

1 2 3 4 5 6 7

loss = torch.nn.MSELoss()   def get_net(feature_num):     net = nn.Linear(feature_num, 1)     for param in net.parameters():         nn.init.normal_(param, mean=0, std=0.01)     return net

附录2

?

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18

def train(net, train_features, train_labels, test_features, test_labels, num_epochs, learning_rate,weight_decay, batch_size):     train_ls, test_ls = [], []     dataset = torch.utils.data.TensorDataset(train_features, train_labels)     train_iter = torch.utils.data.DataLoader(dataset, batch_size, shuffle=True) #TensorDataset和DataLoader的使用请查看我以前的博客           #这里使用了Adam优化算法     optimizer = torch.optim.Adam(params=net.parameters(), lr= learning_rate, weight_decay=weight_decay)     net = net.float()     for epoch in range(num_epochs):         for X, y in train_iter:             l = loss(net(X.float()), y.float())             optimizer.zero_grad()             l.backward()             optimizer.step()         train_ls.append(log_rmse(net, train_features, train_labels))         if test_labels is not None:             test_ls.append(log_rmse(net, test_features, test_labels))     return train_ls, test_ls

 附录3

?

1 2 3 4 5 6 7 8

def semilogy(x_vals, y_vals, x_label, y_label, x2_vals=None, y2_vals=None, legend=None, figsize=(3.5, 2.5)):     set_figsize(figsize)     plt.xlabel(x_label)     plt.ylabel(y_label)     plt.semilogy(x_vals, y_vals)     if x2_vals and y2_vals:         plt.semilogy(x2_vals, y2_vals, linestyle=':')         plt.legend(legend)

 注：由于最近有其他任务，所以此博客写的匆忙，等我有时间后会丰富，也可能加详细解释。