-
- 0.前言
- 1.导入数据
- 2.查看房价分布
- 3.填充缺失数据
- 4.建模
- 5.提交结果
0.前言
本文对Kaggle房价的训练集和测试集进行分析,采用正则线性回归,对房价进行了预测.本人将思路记录下来,以供参考.如有不足之处,欢迎指正.
1.导入数据
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
# 忽略警告
import warnings
warnings.filterwarnings('ignore')
# 读取训练集和测试集
train = pd.read_csv('train.csv')
train_len = len(train)
test = pd.read_csv('test.csv')
# 查看训练集
train.head()
| Id | MSSubClass | MSZoning | LotFrontage | LotArea | Street | Alley | LotShape | LandContour | Utilities | … | PoolArea | PoolQC | Fence | MiscFeature | MiscVal | MoSold | YrSold | SaleType | SaleCondition | SalePrice |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | 60 | RL | 65.0 | 8450 | Pave | NaN | Reg | Lvl | AllPub | … | NaN | NaN | NaN | 2 | 2008 | WD | Normal | 208500 | ||
| 1 | 2 | 20 | RL | 80.0 | 9600 | Pave | NaN | Reg | Lvl | AllPub | … | NaN | NaN | NaN | 5 | 2007 | WD | Normal | 181500 | |
| 2 | 3 | 60 | RL | 68.0 | 11250 | Pave | NaN | IR1 | Lvl | AllPub | … | NaN | NaN | NaN | 9 | 2008 | WD | Normal | 223500 | |
| 3 | 4 | 70 | RL | 60.0 | 9550 | Pave | NaN | IR1 | Lvl | AllPub | … | NaN | NaN | NaN | 2 | 2006 | WD | Abnorml | 140000 | |
| 4 | 5 | 60 | RL | 84.0 | 14260 | Pave | NaN | IR1 | Lvl | AllPub | … | NaN | NaN | NaN | 12 | 2008 | WD | Normal | 250000 |
5 rows × 81 columns
# 查看测试集, 缺少最后一列SalePrice
test.head()
| Id | MSSubClass | MSZoning | LotFrontage | LotArea | Street | Alley | LotShape | LandContour | Utilities | … | ScreenPorch | PoolArea | PoolQC | Fence | MiscFeature | MiscVal | MoSold | YrSold | SaleType | SaleCondition |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1461 | 20 | RH | 80.0 | 11622 | Pave | NaN | Reg | Lvl | AllPub | … | 120 | NaN | MnPrv | NaN | 6 | 2010 | WD | Normal | ||
| 1 | 1462 | 20 | RL | 81.0 | 14267 | Pave | NaN | IR1 | Lvl | AllPub | … | NaN | NaN | Gar2 | 12500 | 6 | 2010 | WD | Normal | |
| 2 | 1463 | 60 | RL | 74.0 | 13830 | Pave | NaN | IR1 | Lvl | AllPub | … | NaN | MnPrv | NaN | 3 | 2010 | WD | Normal | ||
| 3 | 1464 | 60 | RL | 78.0 | 9978 | Pave | NaN | IR1 | Lvl | AllPub | … | NaN | NaN | NaN | 6 | 2010 | WD | Normal | ||
| 4 | 1465 | 120 | RL | 43.0 | 5005 | Pave | NaN | IR1 | HLS | AllPub | … | 144 | NaN | NaN | NaN | 1 | 2010 | WD | Normal |
5 rows × 80 columns
# 合并训练集和测试集,去掉房价一列
all_data = pd.concat([train, test], axis = , ignore_index= True)
all_data.drop(labels = ["SalePrice"],axis = , inplace = True)
2.查看房价分布
由于特征太多，我们在此不查看各特征与房价的关系，只看房价的分布。
# 查看训练集的房价分布，左图是原始房价分布，右图是将房价对数化之后的分布
fig = plt.figure(figsize=(,))
ax1 = fig.add_subplot()
ax2 = fig.add_subplot()
g1 = sns.distplot(train['SalePrice'],hist = True,label='skewness:{:.2f}'.format(train['SalePrice'].skew()),ax = ax1)
g1.legend()
g1.set(xlabel = 'Price')
g2 = sns.distplot(np.log1p(train['SalePrice']),hist = True,label='skewness:{:.2f}'.format(np.log1p(train['SalePrice']).skew()),ax=ax2)
g2.legend()
g2.set(xlabel = 'log(Price+1)')
plt.show()
# 由于房价是有偏度的,将房价对数化
train['SalePrice'] = np.log1p(train['SalePrice'])
# 将有偏的数值特征对数化
num_features_list = list(all_data.dtypes[all_data.dtypes != "object"].index)
for i in num_features_list:
if all_data[i].dropna().skew() > :
all_data[i] = np.log1p(all_data[i])
# 将类别数值转化为虚拟变量
all_data = pd.get_dummies(all_data)
3.填充缺失数据
由于缺失值很多，我们在此不逐一预测，仅用均值来填充。
# 查看缺失值
all_data.isnull().sum()
1stFlrSF 0
2ndFlrSF 0
3SsnPorch 0
BedroomAbvGr 0
BsmtFinSF1 1
BsmtFinSF2 1
BsmtFullBath 2
BsmtHalfBath 2
BsmtUnfSF 1
EnclosedPorch 0
Fireplaces 0
FullBath 0
GarageArea 1
GarageCars 1
GarageYrBlt 159
GrLivArea 0
HalfBath 0
Id 0
KitchenAbvGr 0
LotArea 0
LotFrontage 486
LowQualFinSF 0
MSSubClass 0
MasVnrArea 23
MiscVal 0
MoSold 0
OpenPorchSF 0
OverallCond 0
OverallQual 0
PoolArea 0
...
RoofMatl_Metal 0
RoofMatl_Roll 0
RoofMatl_Tar&Grv 0
RoofMatl_WdShake 0
RoofMatl_WdShngl 0
RoofStyle_Flat 0
RoofStyle_Gable 0
RoofStyle_Gambrel 0
RoofStyle_Hip 0
RoofStyle_Mansard 0
RoofStyle_Shed 0
SaleCondition_Abnorml 0
SaleCondition_AdjLand 0
SaleCondition_Alloca 0
SaleCondition_Family 0
SaleCondition_Normal 0
SaleCondition_Partial 0
SaleType_COD 0
SaleType_CWD 0
SaleType_Con 0
SaleType_ConLD 0
SaleType_ConLI 0
SaleType_ConLw 0
SaleType_New 0
SaleType_Oth 0
SaleType_WD 0
Street_Grvl 0
Street_Pave 0
Utilities_AllPub 0
Utilities_NoSeWa 0
Length: 289, dtype: int64
# 将缺失值用该列的均值填充
all_data = all_data.fillna(all_data.mean())
# 将测试集和训练集分开
X_train = all_data[:train_len]
X_test = all_data[train_len:]
Y_train = train['SalePrice']
4.建模
from sklearn.linear_model import Ridge, LassoCV
from sklearn.model_selection import cross_val_score
# 定义交叉验证,用均方根误差来评价模型的拟合程度
def rmse_cv(model):
rmse = np.sqrt(-cross_val_score(model, X_train, Y_train, scoring = 'neg_mean_squared_error', cv=))
return rmse
# Ridge模型
model_ridge = Ridge()
alphas = [, , , , , , , , , , ]
cv_ridge = [rmse_cv(Ridge(alpha = a)).mean() for a in alphas]
cv_ridge = pd.Series(cv_ridge, index = alphas)
cv_ridge
# 交叉验证可视化
fig = plt.figure(figsize=(,))
cv_ridge.plot(title = 'Cross Validation Score with Model Ridge')
plt.xlabel("alpha")
plt.ylabel("rmse")
plt.show()
# 当alpha为10时,均方根误差最小
cv_ridge.min()
0.12699476769354789
# lasso模型,均方根误差的均值更小,因此最终选择lasso模型
model_lasso = LassoCV(alphas = [, , , ]).fit(X_train, Y_train)
rmse_cv(model_lasso).mean()
0.12296228157910054
# 查看模型系数, lasso模型能选择特征,将不重要的特征系数设置为0
coef = pd.Series(model_lasso.coef_, index = X_train.columns)
print("Lasso picked {} variables and eliminated the other {} variables".format(sum(coef != ), sum(coef==)))
Lasso picked 110 variables and eliminated the other 179 variables
# 查看重要的特征, GrLivArea地上面积是最重要的正相关特征
imp_coef = pd.concat([coef.sort_values().head(),coef.sort_values().tail()])
fig = plt.figure(figsize=(,))
imp_coef.plot(kind = "barh")
plt.title("Coefficients in the Lasso Model")
plt.show()
# 查看残差
est = pd.DataFrame({"est":model_lasso.predict(X_train), "true":Y_train})
plt.rcParams["figure.figsize"] = [,]
est["resi"] = est["true"] - est["est"]
est.plot(x = "est", y = "resi",kind = "scatter")
plt.show()
# xgboost模型
import xgboost as xgb
dtrain = xgb.DMatrix(X_train, label = Y_train)
dtest = xgb.DMatrix(X_test)
# 交叉验证
params = {"max_depth":, "eta":}
cv_xgb = xgb.cv(params, dtrain, num_boost_round=, early_stopping_rounds=)
cv_xgb.loc[:,["test-rmse-mean", "train-rmse-mean"]].plot()
plt.show()
# 训练模型
model_xgb = xgb.XGBRegressor(n_estimators=, max_depth=, learning_rate=)
model_xgb.fit(X_train, Y_train)
XGBRegressor(base_score=0.5, booster='gbtree', colsample_bylevel=1,
colsample_bytree=1, gamma=0, learning_rate=0.1, max_delta_step=0,
max_depth=2, min_child_weight=1, missing=None, n_estimators=360,
n_jobs=1, nthread=None, objective='reg:linear', random_state=0,
reg_alpha=0, reg_lambda=1, scale_pos_weight=1, seed=None,
silent=True, subsample=1)
# 查看两种模型的预测结果, 将结果指数化
lasso_preds = np.expm1(model_lasso.predict(X_test))
xgb_preds = np.expm1(model_xgb.predict(X_test))
predictions = pd.DataFrame({"xgb":xgb_preds, "lasso":lasso_preds})
predictions.plot(x = "xgb", y = "lasso", kind = "scatter")
plt.show()
5.提交结果
# 最终结果采用两种模型预测的加权平均值,提交结果
preds = *lasso_preds + *xgb_preds
result = pd.DataFrame({"id":test.Id, "SalePrice":preds})
result.to_csv('result.csv', index = False)
结果排在前19%, 还有改进的空间, 要继续努力呀.