Kaggle: House Prices: Advanced Regression Techniques

notebook来自https://www.kaggle.com/neviadomski/how-to-get-to-top-25-with-simple-model-sklearn

思路流程：

1.导入数据，查看数据结构和缺失值情况
重点在于查看缺失值情况的写法:
NAs = pd.concat([train.isnull().sum(), test.isnull().sum()], axis = 1, keys = ['train', 'test']) NAs[NAs.sum(axis=1) > 0]

2.数据预处理（删除无用特征，特征转化，缺失值填充，构造新特征，特征值标准化，转化为dummy）
Q:什么样的特征需要做转化？
A:如某些整型数据只表示类别，其数值本身没有意义，则应转化为dummy
重点学习手动将特征转化为dummy的方法（这里情况稍微还要复杂一点，因为存在同一特征对应两列的情况，如Condition1,Condition2)

3.随机打乱数据，分离训练集和测试集

4.构建多个单一模型

5.模型融合

问题：

1.如何判断一个特征是否是无用特征？

2.模型融合的方法？这里为什是np.exp(GB_model.predict(test_features)) + np.exp(ENS_model.predict(test_features_std))？

3.为什么label分布偏斜需要做转化？

In [33]: #Kaggle: House Prices: Advanced Regression Techniques import numpy as np import pandas as pd import matplotlib.pyplot as plt import seaborn as sns from sklearn import ensemble, linear_model, tree from sklearn.model_selection import train_test_split, cross_val_score from sklearn.metrics import mean_squared_error, r2_score from sklearn.utils import shuffle ​ %matplotlib inline import warnings warnings.filterwarnings('ignore') ​ train = pd.read_csv('downloads/train.csv') test = pd.read_csv('downloads/test.csv') In [8]: train.head() Out[8]:  IdMSSubClassMSZoningLotFrontageLotAreaStreetAlleyLotShapeLandContourUtilities...PoolAreaPoolQCFenceMiscFeatureMiscValMoSoldYrSoldSaleTypeSaleConditionSalePrice01234

1	60	RL	65.0	8450	Pave	NaN	Reg	Lvl	AllPub	...	0	NaN	NaN	NaN	0	2	2008	WD	Normal	208500
2	20	RL	80.0	9600	Pave	NaN	Reg	Lvl	AllPub	...	0	NaN	NaN	NaN	0	5	2007	WD	Normal	181500
3	60	RL	68.0	11250	Pave	NaN	IR1	Lvl	AllPub	...	0	NaN	NaN	NaN	0	9	2008	WD	Normal	223500
4	70	RL	60.0	9550	Pave	NaN	IR1	Lvl	AllPub	...	0	NaN	NaN	NaN	0	2	2006	WD	Abnorml	140000
5	60	RL	84.0	14260	Pave	NaN	IR1	Lvl	AllPub	...	0	NaN	NaN	NaN	0	12	2008	WD	Normal	250000

5 rows × 81 columns 

In [9]: #检查缺失值 NAs = pd.concat([train.isnull().sum(), test.isnull().sum()], axis = 1, keys = ['train', 'test']) #sum()默认的axis=0，即跨行 NAs[NAs.sum(axis=1) > 0] #只显示有缺失值的特征 Out[9]:  traintestAlleyBsmtCondBsmtExposureBsmtFinSF1BsmtFinSF2BsmtFinType1BsmtFinType2BsmtFullBathBsmtHalfBathBsmtQualBsmtUnfSFElectricalExterior1stExterior2ndFenceFireplaceQuFunctionalGarageAreaGarageCarsGarageCondGarageFinishGarageQualGarageTypeGarageYrBltKitchenQualLotFrontageMSZoningMasVnrAreaMasVnrTypeMiscFeaturePoolQCSaleTypeTotalBsmtSFUtilities

1369	1352.0
37	45.0
38	44.0
0	1.0
0	1.0
37	42.0
38	42.0
0	2.0
0	2.0
37	44.0
0	1.0
1	0.0
0	1.0
0	1.0
1179	1169.0
690	730.0
0	2.0
0	1.0
0	1.0
81	78.0
81	78.0
81	78.0
81	76.0
81	78.0
0	1.0
259	227.0
0	4.0
8	15.0
8	16.0
1406	1408.0
1453	1456.0
0	1.0
0	1.0
0	2.0

In [10]: #打印R2和RMSE得分 def print_score (prediction, labels):print('R2: {}'.format(r2_score(prediction, labels)))print('RMSE: {}'.format(np.sqrt(mean_squared_error(prediction, labels)))) ​ #对给定的模型进行评估，分别打印训练集上的得分和测试集上的得分 def train_test_score(estimator, x_train, x_test, y_train, y_test):train_predictions = estimator.predict(x_train)print('------------train-----------')print_score(train_predictions, y_train)print('------------test------------')test_predictions = estimator.predict(x_test)print_score(test_predictions, y_test) In [11]: #将标签从训练集中分离出来 train_label = train.pop('SalePrice') ​ #将训练集特征和测试集特征拼在一起，便于一起删除无用的特征 features = pd.concat([train, test], keys = ['train', 'test']) ​ #删除无用特征（为什么说它们是无用特征并没有解释） features.drop(['Utilities', 'RoofMatl', 'MasVnrArea', 'BsmtFinSF1', 'BsmtFinSF2', 'BsmtUnfSF', 'Heating', 'LowQualFinSF','BsmtFullBath', 'BsmtHalfBath', 'Functional', 'GarageYrBlt', 'GarageArea', 'GarageCond', 'WoodDeckSF','OpenPorchSF', 'EnclosedPorch', '3SsnPorch', 'ScreenPorch', 'PoolArea', 'PoolQC', 'Fence', 'MiscFeature', 'MiscVal'],axis=1, inplace=True) print(features.shape) (2919, 56) In [12]: #将series数据转化为str #问题：什么样的数据需要转化为str #答：将原来的某些整型数据转化为str，这些整型数据数字大小本身并没有含义，而只是代表一个类，所以转化为str后，后续再转化为dummy features['MSSubClass'] = features['MSSubClass'].astype(str) #pandas调用特征的两种方法：.feature和['feature']，两者效果相同，下面就是.feature方法 features.OverallCond = features.OverallCond.astype(str) features['KitchenAbvGr'] = features['KitchenAbvGr'].astype(str) features['YrSold'] = features['YrSold'].astype(str) features['MoSold'] = features['MoSold'].astype(str) ​ #用众数填充缺失值 features['MSZoning'] = features['MSZoning'].fillna(features['MSZoning'].mode()[0]) features['MasVnrType'] = features['MasVnrType'].fillna(features['MasVnrType'].mode()[0]) features['Electrical'] = features['Electrical'].fillna(features['Electrical'].mode()[0]) features['KitchenQual'] = features['KitchenQual'].fillna(features['KitchenQual'].mode()[0]) features['SaleType'] = features['SaleType'].fillna(features['SaleType'].mode()[0]) ​ #用某个特定值填充缺失值 features['LotFrontage'] = features['LotFrontage'].fillna(features['LotFrontage'].mean()) features['Alley'] = features['Alley'].fillna('NOACCESS') for col in ('BsmtQual', 'BsmtCond', 'BsmtExposure', 'BsmtFinType1', 'BsmtFinType2'):features[col] = features[col].fillna('NoBSMT') features['TotalBsmtSF'] = features['TotalBsmtSF'].fillna(0) features['FireplaceQu'] = features['FireplaceQu'].fillna('NoFP') for col in ('GarageType', 'GarageFinish', 'GarageQual'):features[col] = features[col].fillna('NoGRG') features['GarageCars'] = features['GarageCars'].fillna(0.0) ​ #构造新特征 features['TotalSF'] = features['TotalBsmtSF'] + features['1stFlrSF'] + features['2ndFlrSF'] features.drop(['TotalBsmtSF', '1stFlrSF', '2ndFlrSF'], axis=1, inplace=True) print(features.shape) (2919, 54) In [13]: #查看房价分布情况 ax = sns.distplot(train_label) In [14]: #发现图像整体向左倾斜，所以做log转变 train_label = np.log(train_label) ax = sns.distplot(train_label) In [15]: #对数字特征做标准化处理 num_features = features.loc[:,['LotFrontage', 'LotArea', 'GrLivArea', 'TotalSF']] num_features_standarized = (num_features - num_features.mean()) / num_features.std() num_features_standarized.head() Out[15]:   LotFrontageLotAreaGrLivAreaTotalSFtrain01234

-0.202033	-0.217841	0.413476	0.022999
0.501785	-0.072032	-0.471810	-0.029167
-0.061269	0.137173	0.563659	0.196886
-0.436639	-0.078371	0.427309	-0.092511
0.689469	0.518814	1.377806	0.988072

In [16]: ax = sns.pairplot(num_features_standarized) In [17]: #重点 #convert categorical data to dummies #将所有condition不重复的记录在一个set中 conditions = set([x for x in features['Condition1']] + [x for x in features['Condition2']]) #自定义dummy变量，行数为阳历数，列数为原condition数据转化为dummy后的维数 dummies = pd.DataFrame(data = np.zeros((len(features.index), len(conditions))), index = features.index, columns = conditions) #遍历所有样例，将原来的condition信息转化为对应的dummy信息 for i, cond in enumerate(zip(features['Condition1'], features['Condition2'])): #用ix找到位置，注意cond可能包含Condition1和Condition2两个位置的信息，对应dummies数组的两个点，所以需要用ix而不能简单的直接用dummies[i,cond]dummies.ix[i, cond] = 1 #将dummy后的特征数据拼接到原features后面，并给dummy特征的index增加前缀 features = pd.concat([features, dummies.add_prefix('Cond_')], axis = 1) #最后就可以删除原来的Condition特征 features.drop(['Condition1', 'Condition2'], axis = 1, inplace =True) print(features.shape) (2919, 61) In [18]: features.head() Out[18]:   IdMSSubClassMSZoningLotFrontageLotAreaStreetAlleyLotShapeLandContourLotConfig...TotalSFCond_PosACond_ArteryCond_PosNCond_RRAnCond_RRAeCond_FeedrCond_NormCond_RRNnCond_RRNetrain01234

1	60	RL	65.0	8450	Pave	NOACCESS	Reg	Lvl	Inside	...	2566.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0
2	20	RL	80.0	9600	Pave	NOACCESS	Reg	Lvl	FR2	...	2524.0	0.0	0.0	0.0	0.0	0.0	1.0	1.0	0.0	0.0
3	60	RL	68.0	11250	Pave	NOACCESS	IR1	Lvl	Inside	...	2706.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0
4	70	RL	60.0	9550	Pave	NOACCESS	IR1	Lvl	Corner	...	2473.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0
5	60	RL	84.0	14260	Pave	NOACCESS	IR1	Lvl	FR2	...	3343.0	0.0	0.0	0.0	0.0	0.0	0.0	1.0	0.0	0.0

5 rows × 61 columns

In [19]: #convert Exterior to dummies Exterior = set([x for x in features['Exterior1st']] + [x for x in features['Exterior2nd']]) dummies = pd.DataFrame(data = np.zeros([len(features.index), len(Exterior)]), index = features.index, columns = Exterior) for i, ext in enumerate(zip(features['Exterior1st'], features['Exterior2nd'])):dummies.ix[i, ext] = 1 features = pd.concat([features, dummies.add_prefix('Ext_')], axis = 1) features.drop(['Exterior1st', 'Exterior2nd', 'Ext_nan'], axis = 1, inplace = True) print(features.shape) (2919, 78) In [20]: features.dtypes[features.dtypes == 'object'].index Out[20]: Index(['MSSubClass', 'MSZoning', 'Street', 'Alley', 'LotShape', 'LandContour','LotConfig', 'LandSlope', 'Neighborhood', 'BldgType', 'HouseStyle','OverallCond', 'RoofStyle', 'MasVnrType', 'ExterQual', 'ExterCond','Foundation', 'BsmtQual', 'BsmtCond', 'BsmtExposure', 'BsmtFinType1','BsmtFinType2', 'HeatingQC', 'CentralAir', 'Electrical', 'KitchenAbvGr','KitchenQual', 'FireplaceQu', 'GarageType', 'GarageFinish','GarageQual', 'PavedDrive', 'MoSold', 'YrSold', 'SaleType','SaleCondition'],dtype='object') In [21]: #遍历特定类型数据的方法：for col in features.dtypes[features.dtypes == 'object'].index #convert all other categorical vars to dummies for col in features.dtypes[features.dtypes == 'object'].index:for_dummy = features.pop(col)features = pd.concat([features, pd.get_dummies(for_dummy, prefix = col)], axis = 1) print(features.shape) (2919, 263) In [22]: #用之前几个标准化的数据更新features ​ features_standardized = features.copy() features_standardized.update(num_features_standarized) In [23]: #重新分离训练集和测试集 ​ #首先分离没有标准化的features train_features = features.loc['train'].drop(['Id'], axis=1).select_dtypes(include=[np.number]).values test_features = features.loc['test'].drop(['Id'], axis=1).select_dtypes(include=[np.number]).values ​ #再分离标准化的数据 train_features_std = features_standardized.loc['train'].drop(['Id'], axis=1).select_dtypes(include=[np.number]).values test_features_std = features_standardized.loc['test'].drop(['Id'], axis=1).select_dtypes(include=[np.number]).values print(train_features.shape) print(train_features_std.shape) (1460, 262) (1460, 262) In [24]: #shuffle train dataset train_features_std, train_features, train_label = shuffle(train_features_std, train_features, train_label, random_state = 5) In [25]: #split train and test data x_train, x_test, y_train, y_test = train_test_split(train_features, train_label, test_size = 0.1, random_state = 200) x_train_std, x_test_std, y_train_std, y_test_std = train_test_split(train_features_std, train_label, test_size = 0.1, random_state = 200) In [26]: #构建第一个模型：ElasticNet ENSTest = linear_model.ElasticNetCV(alphas=[0.0001, 0.0005, 0.001, 0.01, 0.1, 1, 10], l1_ratio=[.01, .1, .5, .9, .99], max_iter=5000).fit(x_train_std, y_train_std) train_test_score(ENSTest, x_train_std, x_test_std, y_train_std, y_test_std) ------------train----------- R2: 0.9009283127352861 RMSE: 0.11921419084690392 ------------test------------ R2: 0.8967299522701895 RMSE: 0.11097042840114624 In [27]: #测试模型的交叉验证得分 score = cross_val_score(ENSTest, train_features_std, train_label, cv = 5) print('Accurary: %0.2f +/- %0.2f' % (score.mean(), score.std()*2)) Accurary: 0.88 +/- 0.10 In [28]: #构建第二个模型：GradientBoosting GB = ensemble.GradientBoostingRegressor(n_estimators=3000, learning_rate = 0.05, max_depth = 3, max_features = 'sqrt', min_samples_leaf = 15,min_samples_split = 10, loss = 'huber').fit(x_train_std, y_train_std) train_test_score(GB, x_train_std, x_test_std, y_train_std, y_test_std) ------------train----------- R2: 0.9607778449577035 RMSE: 0.07698826081848897 ------------test------------ R2: 0.9002871760789876 RMSE: 0.10793269100940146 In [29]: #构建第二个模型：GradientBoosting GB = ensemble.GradientBoostingRegressor(n_estimators=3000, learning_rate = 0.05, max_depth = 3, max_features = 'sqrt', min_samples_leaf = 15,min_samples_split = 10, loss = 'huber').fit(x_train_std, y_train_std) train_test_score(GB, x_train_std, x_test_std, y_train_std, y_test_std) Accurary: 0.90 +/- 0.04 In [30]: #模型融合 GB_model = GB.fit(train_features, train_label) ENS_model = ENSTest.fit(train_features_std, train_label) In [31]: #为什么模型融合公式是这样的？ Final_score = (np.exp(GB_model.predict(test_features)) + np.exp(ENS_model.predict(test_features_std))) / 2 In [32]: #写入csv文件 pd.DataFrame({'Id':test.Id, 'SalePrice':Final_score}).to_csv('submit.csv', index=False)

转载于:https://www.cnblogs.com/RB26DETT/p/11566650.html

总结

以上是生活随笔为你收集整理的Kaggle: House Prices: Advanced Regression Techniques的全部内容，希望文章能够帮你解决所遇到的问题。

如果觉得生活随笔网站内容还不错，欢迎将生活随笔推荐给好友。