def fit_and_plot(train_features, test_features, train_labels, test_labels):
# 写一个只有线性层的网络，打平连接到一个输出
# 这个过程自动初始化了参数所以不用额外处理了
net = torch.nn.Linear(train_features.shape[-1], 1)
# 设置批大小，记得限制最小值
batch_size = min(10, train_labels.shape[0])
# 读取和参数设置
dataset = torch.utils.data.TensorDataset(train_features, train_labels)
train_iter = torch.utils.data.DataLoader(dataset, batch_size, shuffle=True)
optimizer = torch.optim.SGD(net.parameters(), lr=0.01)
train_ls, test_ls = [], []
# 用下划线来标明这个循环的循环值是不需要用到的
for _ in range(num_epochs):
# 进行正常的训练和误差计算
for X, y in train_iter:
l = loss(net(X), y.view(-1, 1))
optimizer.zero_grad()
l.backward()
optimizer.step()
# 把数据的标签打平变为方便处理的标签组
train_labels = train_labels.view(-1, 1)
test_labels = test_labels.view(-1, 1)
# 计算出预测的误差并和上面的标签组合起来
train_ls.append(loss(net(train_features), train_labels).item())
test_ls.append(loss(net(test_features), test_labels).item())
# 打印结果
print('final epoch: train loss', train_ls[-1], 'test loss', test_ls[-1])
plot.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
range(1, num_epochs + 1), test_ls, ['train', 'test'])
print('weight:', net.weight.data, '\nbias:', net.bias.data)
def fit_and_plot_pytorch(wd):
net = nn.Linear(num_features, 1)
nn.init.normal_(net.weight, mean=0, std=1)
nn.init.normal_(net.bias, mean=0, std=1)
# weight_decay就是附加在损失函数上的惩罚系数,修改这个超参数会改变衰减比率
optim_w = torch.optim.SGD(params=[net.weight], lr=lr, weight_decay=wd)
# 不对bias进行惩罚
optim_b = torch.optim.SGD(params=[net.bias], lr=lr)
# 训练
train_ls, test_ls = [], []
for _ in range(num_epochs):
for X, y in train_iter:
# 具体的训练和之前的做法都一样
l = loss(net(X), y).mean()
optim_w.zero_grad()
optim_b.zero_grad()
l.backward()
# 反向传播后就让优化器迭代
optim_w.step()
optim_b.step()
train_ls.append(loss(net(train_features), train_labels).mean().item())
test_ls.append(loss(net(test_features), test_labels).mean().item())
plot.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
range(1, num_epochs + 1), test_ls, ['train', 'test'])
print('L2 norm of w:', net.weight.data.norm().item())
def train_and_pred(train_features, test_features, train_labels, test_data,
num_epochs, lr, weight_decay, batch_size):
# 适配网络
net = get_net(train_features.shape[1])
# 进行训练，这里是使用了前面猜测好的超参数进行的
train_ls, _ = train(net, train_features, train_labels, None, None,
num_epochs, lr, weight_decay, batch_size)
# 绘图和输出
plot.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'rmse')
print('train rmse %f' % train_ls[-1])
# detach用于得到当前计算图的变量
preds = net(test_features).detach().numpy()
# 保存到提交文档中
test_data['SalePrice'] = pd.Series(preds.reshape(1, -1)[0])
submission = pd.concat([test_data['Id'], test_data['SalePrice']], axis=1)
submission.to_csv('./Datasets/KaggleHouse/submission.csv', index=False)
def fit_and_plot(lambd):
w, b = init_params()
train_ls, test_ls = [], []
for _ in range(num_epochs):
for X, y in train_iter:
# 这里的损失项要加上前面写好的L2惩罚项,注意乘上lambd好控制衰减比率
l = loss(net(X, w, b), y) + lambd * l2_penalty(w)
l = l.sum()
if w.grad is not None:
w.grad.data.zero_()
b.grad.data.zero_()
l.backward()
# 进行梯度下降
linear_reg.sgd([w, b], lr, batch_size)
# 记录每次训练的loss
train_ls.append(
loss(net(train_features, w, b), train_labels).mean().item())
test_ls.append(
loss(net(test_features, w, b), test_labels).mean().item())
# 绘制和输出
print('L2 norm of w:', w.norm().item())
plot.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'loss',
range(1, num_epochs + 1), test_ls, ['train', 'test'])
def k_fold(k, X_train, y_train, num_epochs, learning_rate, weight_decay,
batch_size):
train_l_sum, valid_l_sum = 0, 0
# 对于每一种折法
for i in range(k):
# 得到k折的验证集和训练集
data = get_k_fold_data(k, i, X_train, y_train)
# 用X的特征数来生成一个对应大小的网络
net = get_net(X_train.shape[1])
# 将数据应用到上面的训练函数中，*data自动将四个返回值展开输入到参数中
train_ls, valid_ls = train(net, *data, num_epochs, learning_rate,
weight_decay, batch_size)
# -1指取出最后一格元素
train_l_sum += train_ls[-1]
valid_l_sum += valid_ls[-1]
# 绘图第一折的误差曲线，并输出每一折的训练结果
if i == 0:
plot.semilogy(range(1, num_epochs + 1), train_ls, 'epochs', 'rmse',
range(1, num_epochs + 1), valid_ls,
['train', 'valid'])
print('fold %d, train rmse %f, valid rmse %f' %
(i, train_ls[-1], valid_ls[-1]))
# 返回误差总和的平均值，valid部分可以较好地衡量泛化误差
return train_l_sum / k, valid_l_sum / k