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]:
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]:
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)
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)
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]:
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)
In [18]:
features.head()
Out[18]:
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)
In [20]:
features.dtypes[features.dtypes == 'object'].index
Out[20]:
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)
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)
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)
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))
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)
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)
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)