竞赛大佬方法：（它们都是使用集成学习的方法，来集成学习多个模型）
第二名和第七名：autogluon
第三名：h2o
第四名：随机森林
import numpy as np
import pandas as pd
import torch
from torch import nn
from d2l import torch as d2l
train_data = pd.read_csv('..\\data\\kaggle\\california-house-prices\\train.csv')
test_data = pd.read_csv('..\\data\\kaggle\\california-house-prices\\test.csv')
查看数据集的形状
print(train_data.shape, test_data.shape)
查看数据集train_data比test_data多一列：Sold Price(就是我们要预测的房价)
[i for i in train_data.columns if i not in test_data.columns]
看一下数据集的具体样子
train_data
查看训练集和测试集的特征名称
# 查看特征
# train_data.columns，test_data.columns
其中部分特征：
剔除特别不好的特征
由于去除了标签，可以将训练集和测试集合并：方便后续处理
all_features = pd.concat((train_data.iloc[:, 4:-1], test_data.iloc[:, 3:-1]))
查看所有样本的形状
all_features.shape # 样本数：47439+31626= 79065 特征数：41-5=36 40-4=36
查看特征中有哪些数据类型
# 查看特征中有哪些数据类型
all_features.dtypes.unique()
查看时间数据
all_features['Year built'], all_features['Listed On'], all_features['Last Sold On']
all_features['Listed On'] = pd.to_datetime(all_features['Listed On'], format="%Y-%m-%d")
all_features['Last Sold On'] = pd.to_datetime(all_features['Last Sold On'], format="%Y-%m-%d")
查看
all_features['Listed On'], all_features['Last Sold On']
numeric_features = all_features.dtypes[all_features.dtypes != 'object'].index
all_features[numeric_features] = all_features[numeric_features].apply(
lambda x: (x - x.mean()) / (x.std()))
# 在标准化数据后，所有均值消失，因此我们可以将缺失值设置为0
all_features[numeric_features] = all_features[numeric_features].fillna(0)
查看所有数值型特征的形状
all_features[numeric_features].shape
查看它们的特征名称是否与csv文件中的数值型特征一致
numeric_features
用一次独热编码替换它们
查看所有的离散型数据
print(all_features.dtypes[all_features.dtypes =='object'])
object_numeric_features = all_features.dtypes[all_features.dtypes == 'object'].index
for i in object_numeric_features:
print(len(all_features[i].unique()))
features = list(numeric_features)
features.append('Type') # 加上类别数相对较少的Type
print(features)
现在新的特征数：21 (20: 数值型特征， 1：文本型特征)
all_features = all_features[features] # 20 + 1 = 21
all_features.shape
all_features
查看全部特征的数据类型
# 查看全部新特征的数据类型
all_features.dtypes.unique()
print(all_features.dtypes[all_features.dtypes =='object'])
object_numeric_features = all_features.dtypes[all_features.dtypes == 'object'].index
object_numeric_features
n_train = train_data.shape[0] # train_data.shape:[47439, 41] train_data.shape[0]:47439
train_features = torch.tensor(all_features[:n_train].values, dtype=torch.float32) # 训练集：前47439个样本
test_features = torch.tensor(all_features[n_train:].values, dtype=torch.float32) # 验证集：后31626（79065-47439）个样本
train_labels = torch.tensor(train_data['Sold Price'].values.reshape(-1, 1), dtype=torch.float32) # 训练集的标签
# 是否使用GPU训练
if not torch.cuda.is_available():
print('CUDA is not available. Training on CPU ...')
else:
print('CUDA is available. Training on GPU ...')
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
in_features = train_features.shape[1]
def get_net():
net = nn.Sequential(
nn.Linear(in_features, 1024), nn.ReLU(), # in_features = 47439
nn.Linear(1024, 512), nn.ReLU(),
nn.Linear(512, 256), nn.ReLU(),
nn.Linear(256, 128), nn.ReLU(), # 5--折验证：平均训练log rmse：0.241757,平均验证log rmse: 0.277515
# nn.Linear(128, 64), nn.ReLU(), # 5--折验证：平均训练log rmse：0.268310,平均验证log rmse: 0.304618
nn.Linear(128, 1))
return net
loss = nn.MSELoss()
def log_rmse(net, features, labels):
# 为了在取对数时进一步稳定该值， 将小于1的值设置为1
clipped_preds = torch.clamp(net(features), 1, float('inf'))
# torch.clamp()函数的功能将输入input张量每个元素的值压缩到区间 [1,+∞]，并返回结果到一个新张量。
rmse = torch.sqrt(loss(torch.log(clipped_preds), torch.log(labels))) # rmse = log(y_hat)与log(y)的均方损误差在平方
# torch.sqrt()：逐元素计算张量的平方根
return rmse.item()
def train(net, train_features, train_labels, test_features, test_labels,
num_epochs, learning_rate, weight_decay, batch_size):
net = net.to(device) # ---------------------模型传入GPU----------------
train_ls, test_ls = [], []
train_iter = d2l.load_array((train_features, train_labels), batch_size)
optimizer = torch.optim.Adam(net.parameters(), lr=learning_rate, weight_decay=weight_decay)
for epoch in range(num_epochs):
for X, y in train_iter:
X=X.to(device) # ---------------------输入传入GPU----------------
y=y.to(device) # ---------------------标签传入GPU----------------
optimizer.zero_grad()
l = loss(net(X), y).to(device) # ---------------------损失传入GPU----------------
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
def get_k_fold_data(k, i, X, y):
"""把数据分为k块，将第i块拿出来作为验证集，其余k-1块重新合并作为训练集"""
assert k > 1
fold_size = X.shape[0] // k
X_train, y_train = None, None
for j in range(k):
idx = slice(j * fold_size, (j + 1) * fold_size) # 一折有多少个样本
X_part, y_part = X[idx, :], y[idx]
if j == i: # i表示第几折：
X_valid, y_valid = X_part, y_part # j == i：将第i折作为验证集
elif X_train is None: # 表示第一次出现
X_train, y_train = X_part, y_part # 就把这一折存起来，作为训练集
else:
# torch.cat():将两个张量（tensor）按指定维度拼接在一起
X_train = torch.cat([X_train, X_part], 0) # 否则，就把X_train与其余折数据合并作为验证集
y_train = torch.cat([y_train, y_part], 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):
"""返回训练和验证误差的平均值"""
train_l_sum, valid_l_sum = 0, 0
for i in range(k):
data = get_k_fold_data(k, i, X_train, y_train)
net = get_net().to(device) # ---------------------模型传入GPU----------------
train_ls, valid_ls = train(net, *data, # *是解码，变成前面返回的四个数值
num_epochs, learning_rate, weight_decay, batch_size)
# [-1]：
# train函数返回的是整个训练中所有epoch的训练和验证损失数组，数组中每个元素是每一轮计算的损失
# 用-1表示（只取最后一轮的损失）最后一个epoch来代表这一折训练的效果
train_l_sum += train_ls[-1]
valid_l_sum += valid_ls[-1]
if i == 0: # 从0开始画图
d2l.plot(list(range(1, num_epochs + 1)), [train_ls, valid_ls],
xlabel='epoch', ylabel='rmse', xlim=[1, num_epochs],
legend=['train', 'valid'], yscale='log')
print(f'折{i + 1}, 训练log rmse {float(train_ls[-1]):f},'
f'验证log rms{float(valid_ls[-1]):f}')
return train_l_sum / k, valid_l_sum / k
%%time
train_features = train_features.to(device) # ---------------------输入传入GPU----------------
train_labels = train_labels.to(device) # ---------------------标签传入GPU----------------
k, num_epochs, lr, weight_decay, batch_size = 5, 100, 0.01, 0.01, 256
train_l, valid_l = k_fold(k, train_features, train_labels,
num_epochs, lr, weight_decay, batch_size)
print(f'{k}--折验证：平均训练log rmse：{float(train_l):f},'
f'平均验证log rmse: {float(valid_l):f}')
def train_and_pred(train_features, test_featrue, train_labels, test_data,
num_epochs, lr, weight_decay, batch_size):
net = get_net().to(device) # ---------------------模型传入GPU----------------
train_ls, _ = train(net, train_features, train_labels, None, None,
num_epochs, lr, weight_decay, batch_size)
d2l.plot(np.arange(1, num_epochs + 1), [train_ls], xlabel='epoch',
ylabel='log rmse', xlim=[1, num_epochs], yscale='log')
print(f'train log mse {float(train_ls[-1]):f}')
# 将网络应用于测试集。
preds = net(test_featrue).detach().cpu().numpy()
# 将其重新格式化以导出到Kaggle
test_data['SalePrice'] = pd.Series(preds.reshape(1, -1)[0])
submission = pd.concat([test_data['Id'], test_data['SalePrice']], axis=1)
submission.to_csv('submission.csv', index=False)
%%time
test_features = test_features.to(device) # ---------------------测试输入传入GPU----------------
train_features = train_features.to(device) # ---------------------训练输入传入GPU----------------
train_labels = train_labels.to(device) # ---------------------训练标签传入GPU----------------
train_and_pred(train_features, test_features, train_labels, test_data,
num_epochs, lr, weight_decay, batch_size)
