python|特征工程之分类变量的处理

分类变量是表示类别或标记的。与数值型变量不同,分类变量的值是不能被排序的,故而又称为无序变量。
one-hot编码 【python|特征工程之分类变量的处理】独热编码(one-hot encoding)通常用于处理类别间不具有大小关系的特征。独热编码使用一组比特位表示不同的类别,每个比特位表示一个特征。因此,一个可能有k个类别的分类变脸就可以编码成为一个长度为k的特征向量。若变量不能同时属于多个类别,那这组值就只有一个比特位是‘开’的。
独热编码的优缺点:

  1. 独热编码解决了分类器不好处理属性数据的问题,在一定程度上也起到了扩充特征的作用。它的值只有0和1,不同的类型存储在垂直的空间。
  2. 当类别的数量很多时,特征空间会变得非常大。在这种情况下,一般可以用PCA来减少维度。而且one hot encoding+PCA这种组合在实际中也非常有用。使用稀疏向量节省空间配合特征选择降低维度
import pandas as pd from sklearn import linear_model

df = pd.DataFrame({'city':['SF','SF','SF','NYC','NYC','NYC','Seattle','Seattle','Seattle'], 'Rent':[3999, 4000, 4001, 3499, 3500, 3501, 2499, 2500, 2501]})

df['Rent'].mean()

3333.3333333333335

#将分类变量转换为one-hot编码并拟合一个线性回归模型 one_hot_df = pd.get_dummies(df, prefix=['city']) one_hot_df

Rent city_NYC city_SF city_Seattle
0 3999 0 1 0
1 4000 0 1 0
2 4001 0 1 0
3 3499 1 0 0
4 3500 1 0 0
5 3501 1 0 0
6 2499 0 0 1
7 2500 0 0 1
8 2501 0 0 1
model = linear_model.LinearRegression() model.fit(one_hot_df[['city_NYC', 'city_SF', 'city_Seattle']], one_hot_df['Rent']) model.coef_#获取线性回归模型的系数

array([ 166.66666667,666.66666667, -833.33333333])

model.intercept_#获取线性回归模型的截距

3333.3333333333335

model.score(one_hot_df[['city_NYC', 'city_SF', 'city_Seattle']], one_hot_df['Rent'])#获取模型的拟合优度R2

0.9999982857172245

使用one-hot编码时,截距表示目标变量rent的整体均值,每个线性系数表示相应城市的Rent均值与整体Rent均值有多大
虚拟编码 虚拟编码在进行表示时只使用k-1个特征,除去了额外的自由度。没有被使用的那个特征通过一个全零向量来表示,它称为参照类。虚拟编码和one-hot都可以通过pandas.get_dummies实现
#用虚拟编码训练一个线性回归模型,指定drop_first标志来生成虚拟编码

dummy_df = pd.get_dummies(df, prefix=['city'], drop_first=True) dummy_df

Rent city_SF city_Seattle
0 3999 1 0
1 4000 1 0
2 4001 1 0
3 3499 0 0
4 3500 0 0
5 3501 0 0
6 2499 0 1
7 2500 0 1
8 2501 0 1
model.fit(dummy_df[['city_SF', 'city_Seattle']], dummy_df['Rent']) model.coef_

array([500., -1000.])

model.intercept_

3500.0

model.score(dummy_df[['city_SF', 'city_Seattle']], dummy_df['Rent'])

0.9999982857172245

使用虚拟编码时,偏差系数表示相应变量y对于参照类的均值,该例中参照类是city_NYC。第i个特征的系数等于第i个类别的均值与参照类均值的差。
效果编码 效果编码与虚拟编码非常相似,区别在于参照类的用全部由-1组成的向量表示的
effect_df = dummy_df.copy() effect_df.loc[3:5, ['city_SF','city_Seattle']]= -1.0 effect_df

Rent city_SF city_Seattle
0 3999 1.0 0.0
1 4000 1.0 0.0
2 4001 1.0 0.0
3 3499 -1.0 -1.0
4 3500 -1.0 -1.0
5 3501 -1.0 -1.0
6 2499 0.0 1.0
7 2500 0.0 1.0
8 2501 0.0 1.0
model.fit(effect_df[['city_SF', 'city_Seattle']], effect_df['Rent'])

LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None, normalize=False)

model.coef_

array([ 666.66666667, -833.33333333])

model.intercept_

3333.3333333333335

model.score(effect_df[['city_SF', 'city_Seattle']], effect_df['Rent'])

0.9999982857172245

处理大型分类变量 特征散列化 散列函数是一种确定性函数,它可以将一个可能无界的整数映射到一个有限的整数范围【1,m】中。
import pandas as pd import json js = [] with open('yelp_academic_dataset_review.json') as f: for i in range(10000): js.append(json.loads(f.readline())) f.close()review_df = pd.DataFrame(js)# 定义m为唯一的business_id的数量 m = len(review_df.business_id.unique())

m

4174

from sklearn.feature_extraction import FeatureHasher

h = FeatureHasher(n_features = m , input_type='string') f = h.transform(review_df['business_id'])

review_df['business_id'].unique().tolist()[0:5]

['9yKzy9PApeiPPOUJEtnvkg', 'ZRJwVLyzEJq1VAihDhYiow', '6oRAC4uyJCsJl1X0WZpVSA', '_1QQZuf4zZOyFCvXc0o6Vg', '6ozycU1RpktNG2-1BroVtw']

f.toarray()

array([[0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.], ..., [0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.], [0., 0., 0., ..., 0., 0., 0.]])

from sys import getsizeof

print('Our pandas Series, in bytes: ', getsizeof(review_df['business_id'])) print('Our hashed numpy array, in bytes: ', getsizeof(f))

Our pandas Series, in bytes:790152 Our hashed numpy array, in bytes:56

分箱计数
import pandas as pd

df = pd.read_csv('train_subset.csv')

len(df['device_id'].unique()) #查看训练集中有多少个唯一的特征

1075

df.head()

id click hour C1 banner_pos site_id site_domain site_category app_id app_domain ... device_type device_conn_type C14 C15 C16 C17 C18 C19 C20 C21
0 1000009418151094273 0 14102100 1005 0 1fbe01fe f3845767 28905ebd ecad2386 7801e8d9 ... 1 2 15706 320 50 1722 0 35 -1 79
1 10000169349117863715 0 14102100 1005 0 1fbe01fe f3845767 28905ebd ecad2386 7801e8d9 ... 1 0 15704 320 50 1722 0 35 100084 79
2 10000371904215119486 0 14102100 1005 0 1fbe01fe f3845767 28905ebd ecad2386 7801e8d9 ... 1 0 15704 320 50 1722 0 35 100084 79
3 10000640724480838376 0 14102100 1005 0 1fbe01fe f3845767 28905ebd ecad2386 7801e8d9 ... 1 0 15706 320 50 1722 0 35 100084 79
4 10000679056417042096 0 14102100 1005 1 fe8cc448 9166c161 0569f928 ecad2386 7801e8d9 ... 1 0 18993 320 50 2161 0 35 -1 157
5 rows × 24 columns
def click_counting(x, bin_column): clicks = pd.Series( x[x['click'] > 0][bin_column].value_counts(), name='clicks') no_clicks = pd.Series( x[x['click'] < 1][bin_column].value_counts(), name='no_clicks')counts = pd.DataFrame([clicks, no_clicks]).T.fillna('0') counts['total'] = counts['clicks'].astype( 'int64') + counts['no_clicks'].astype('int64')return counts

def bin_counting(counts): counts['N+'] = counts['clicks'].astype('int64').divide( counts['total'].astype('int64')) counts['N-'] = counts['no_clicks'].astype('int64').divide( counts['total'].astype('int64')) counts['log_N+'] = counts['N+'].divide(counts['N-'])#If we wanted to only return bin-counting properties, we would filter here bin_counts = counts.filter(items=['N+', 'N-', 'log_N+']) return counts, bin_counts

bin_column = 'device_id' device_clicks = click_counting(df.filter(items = [bin_column, 'click']), bin_column) device_all, device_bin_counts = bin_counting(device_clicks)

len(device_bin_counts)

1075

device_all.sort_values(by = 'total', ascending = False).head(4)

clicks no_clicks total N+ N- log_N+
a99f214a 1561 7163 8724 0.178932 0.821068 0.217925
c357dbff 2 15 17 0.117647 0.882353 0.133333
a167aa83 0 9 9 0.000000 1.000000 0.000000
3c0208dc 0 9 9 0.000000 1.000000 0.000000
from sys import getsizeof

print('Our pandas Series, in bytes: ', getsizeof(df.filter(items=['device_id', 'click']))) print('Our bin-counting feature, in bytes: ', getsizeof(device_bin_counts))

Our pandas Series, in bytes:730152 Our bin-counting feature, in bytes:95699

参考:
爱丽丝·郑、阿曼达·卡萨丽,精通特征工程

    推荐阅读