使用Sklearn进行特征工程

不念不忘少年蓝@ 2022-09-11 10:19 252阅读 0赞

![在这里插入图片描述][watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16]

**sklearn中的数据预处理和特征工程：**

![54ce298b925644449850896ac947bf34.png][]  
![396eeaf78ffa4a218ca84a25c1202992.png][]

### 数据预处理 ###

#### 数据无量纲化 ####

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 1][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 2][]

from sklearn.preprocessing import MinMaxScaler
    
    data = [[-1, 2], [-0.5, 6], [0, 10], [1, 18]]
    
    #不太熟悉numpy的小伙伴，能够判断data的结构吗？
    #如果换成表是什么样子？
    
    import pandas as pd
    
    pd.DataFrame(data)
    # 0 1
    #0 -1.0 2
    #1 -0.5 6
    #2 0.0 10
    #3 1.0 18
    
    #实现归一化
    scaler = MinMaxScaler() #实例化
    scaler = scaler.fit(data) #fit，在这里本质是生成min(x)和max(x)
    result = scaler.transform(data) #通过接口导出结果
    
    result
    #array([[0. , 0. ],
    # [0.25, 0.25],
    # [0.5 , 0.5 ],
    # [1. , 1. ]])
    result_ = scaler.fit_transform(data) #训练和导出结果一步达成
    
    scaler.inverse_transform(result) #将归一化后的结果逆转
    
    #使用MinMaxScaler的参数feature_range实现将数据归一化到[0,1]以外的范围中
    data = [[-1, 2], [-0.5, 6], [0, 10], [1, 18]]
    scaler = MinMaxScaler(feature_range=[5,10]) #依然实例化

**标准化：**

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 3][]

from sklearn.preprocessing import StandardScaler
    
    data = [[-1, 2], [-0.5, 6], [0, 10], [1, 18]]
    
    scaler = StandardScaler() #实例化
    scaler.fit(data) #fit，本质是生成均值和方差
    
    scaler.mean_ #查看均值的属性mean_
    #array([-0.125, 9. ])
    scaler.var_ #查看方差的属性var_
    #array([ 0.546875, 35. ])
    x_std = scaler.transform(data) #通过接口导出结果
    
    x_std.mean() #导出的结果是一个数组，用mean()查看均值
    #0.0
    x_std.std() #用std()查看方差
    #1.0
    scaler.fit_transform(data) #使用fit_transform(data)一步达成结果
    scaler.inverse_transform(x_std) #使用inverse_transform逆转标准化

![0f345b57e1034046b3885d1af9971b18.png][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 4][]

#### 缺失值处理 ####

class sklearn.impute.SimpleImputer (missing_values=nan, strategy=’mean’, fill_value=None, verbose=0,
    copy=True)

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 5][]

import pandas as pd
    from sklearn.impute import SimpleImputer
    
    data = pd.read_csv("./Narrativedata.csv",index_col=0)
    
    age = data.iloc[:,0]
    Age = age.values.reshape(-1,1)#sklearn当中特征矩阵必须是二维
    
    imp_mean = SimpleImputer() #实例化，默认均值填补
    imp_median = SimpleImputer(strategy="median") #用中位数填补
    imp_0 = SimpleImputer(strategy="constant",fill_value=0) #用0填补
    
    imp_mean = imp_mean.fit_transform(Age) #fit_transform一步完成调取结果
    imp_median = imp_median.fit_transform(Age)
    imp_0 = imp_0.fit_transform(Age)
    
    #在这里我们使用中位数填补Age
    data.loc[:,"Age"] = imp_median
    
    data.info()
    #<class 'pandas.core.frame.DataFrame'>
    #Int64Index: 891 entries, 0 to 890
    #Data columns (total 4 columns):
     # Column Non-Null Count Dtype 
    #--- ------ -------------- ----- 
    # 0 Age 891 non-null float64
    # 1 Sex 891 non-null object 
    # 2 Embarked 889 non-null object 
    # 3 Survived 891 non-null object 
    #dtypes: float64(1), object(3)
    #memory usage: 74.8+ KB
    
    #使用众数填补Embarked
    Embarked = data.loc[:,"Embarked"].values.reshape(-1,1)
    imp_mode = SimpleImputer(strategy = "most_frequent")
    data.loc[:,"Embarked"] = imp_mode.fit_transform(Embarked)

#用pandas和numpy进行填补
    import pandas as pd
    
    data = pd.read_csv("./Narrativedata.csv",index_col=0)
    
    data.loc[:,"Age"] = data.loc[:,"Age"].fillna(data.loc[:,"Age"].median())
    #.fillna 在DataFrame里面直接进行填补'
    
    data.dropna(axis=0,inplace=True)
    #.dropna(axis=0)删除所有有缺失值的行，.dropna(axis=1)删除所有有缺失值的列
    #参数inplace，为True表示在原数据集上进行修改，为False表示生成一个复制对象，不修改原数据，默认False

#### 离散数据处理：将文字型数据转换为数值型 ####

**1.preprocessing.LabelEncoder：标签专用，能够将分类转换为分类数值**

from sklearn.preprocessing import LabelEncoder
    
    y = data.iloc[:,-1] #要输入的是标签，不是特征矩阵，所以允许一维
    
    le = LabelEncoder() #实例化
    le = le.fit(y) #导入数据
    label = le.transform(y)   #transform接口调取结果
    
    le.classes_ #属性.classes_查看标签中究竟有多少类别
    #array(['No', 'Unknown', 'Yes'], dtype=object)
    
    data.iloc[:,-1] = label #让标签等于我们运行出来的结果
    
    data.head()
    # Age Sex Embarked Survived
    #0 22.0 male S 0
    #1 38.0 female C 2
    #2 26.0 female S 2
    #3 35.0 female S 2
    #4 35.0 male S 0

**2.preprocessing.OrdinalEncoder：特征专用，能够将分类特征转换为分类数值**

from sklearn.preprocessing import OrdinalEncoder
    
    #接口categories_对应LabelEncoder的接口classes_，一模一样的功能
    data_ = data.copy()
    
    OrdinalEncoder().fit(data_.iloc[:,1:-1]).categories_
    #[array(['female', 'male'], dtype=object), array(['C', 'Q', 'S'], dtype=object)]
    
    data_.iloc[:,1:-1] = OrdinalEncoder().fit_transform(data_.iloc[:,1:-1])
    
    data_.head()
    # Age Sex Embarked Survived
    #0 22.0 1.0 2.0 0
    #1 38.0 0.0 0.0 2
    #2 26.0 0.0 2.0 2
    #3 35.0 0.0 2.0 2
    #4 35.0 1.0 2.0 0

**3.preprocessing.OneHotEncoder：独热编码，创建哑变量**

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 6][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 7][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 8][]

from sklearn.preprocessing import OneHotEncoder
    
    X = data.iloc[:,1:-1]
    
    enc = OneHotEncoder(categories='auto').fit(X)
    result = enc.transform(X).toarray()
    #array([[0., 1., 0., 0., 1.],
    # [1., 0., 1., 0., 0.],
    # [1., 0., 0., 0., 1.],
    
    #依然可以直接一步到位，但为了给大家展示模型属性，所以还是写成了三步
    #OneHotEncoder(categories='auto').fit_transform(X).toarray()
    
    enc.get_feature_names()
    #array(['x0_female', 'x0_male', 'x1_C', 'x1_Q', 'x1_S'], dtype=object)
    
    #axis=1,表示跨行进行合并，也就是将量表左右相连，如果是axis=0，就是将量表上下相连
    newdata = pd.concat([data,pd.DataFrame(result)],axis=1)
    
    # Age Sex Embarked Survived 0 1 2 3 4
    #0 22.0 male S 0 0.0 1.0 0.0 0.0 1.0
    #1 38.0 female C 2 1.0 0.0 1.0 0.0 0.0
    #2 26.0 female S 2 1.0 0.0 0.0 0.0 1.0
    #3 35.0 female S 2 1.0 0.0 0.0 0.0 1.0
    #4 35.0 male S 0 0.0 1.0 0.0 0.0 1.0
    
    newdata.drop(["Sex","Embarked"],axis=1,inplace=True)
    
    newdata.columns = ["Age","Survived","Female","Male","Embarked_C","Embarked_Q","Embarked_S"]
    
    # Age Survived Female Male Embarked_C Embarked_Q Embarked_S
    #0 22.0 0 0.0 1.0 0.0 0.0 1.0
    #1 38.0 2 1.0 0.0 1.0 0.0 0.0
    #2 26.0 2 1.0 0.0 0.0 0.0 1.0
    #3 35.0 2 1.0 0.0 0.0 0.0 1.0
    #4 35.0 0 0.0 1.0 0.0 0.0 1.0

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 9][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 10][]

#### **连续数据处理：二值化或分段** ####

sklearn.preprocessing.Binarizer

![e7924c9ccdef4223920bdf447a4b92ca.png][]

from sklearn.preprocessing import Binarizer
    
    data_2 = data.copy()
    
    X = data_2.iloc[:,0].values.reshape(-1,1) #类为特征专用，所以不能使用一维数组
    transformer = Binarizer(threshold=30).fit_transform(X) #将年龄中大于30的分类为1，小于等于30的分类为0
    #array([0., 1., 0., 1., 1., 0., 1., 0., 0., 0., 0., 1., 0., 1., 0., 1., 0.,
    # 0., 1., 0., 1., 1., 0., 0., 0., 1., 0., 0., 0., 0., 1., 0., 0., 1.,
    # 0., 1., 0., 0., 0., 0., 1., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0. ...]

preprocessing.KBinsDiscretizer

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 11][]

from sklearn.preprocessing import KBinsDiscretizer
    
    X = data.iloc[:,0].values.reshape(-1,1) 
    est = KBinsDiscretizer(n_bins=3, encode='ordinal', strategy='uniform')
    est.fit_transform(X)
    
    #查看转换后分的箱：变成了一列中的三箱
    set(est.fit_transform(X).ravel())
    #{0.0, 1.0, 2.0}
    
    #uniform:表示等宽分箱
    est = KBinsDiscretizer(n_bins=3, encode='onehot', strategy='uniform')
    #查看转换后分的箱：变成了哑变量
    est.fit_transform(X).toarray()
    #array([[1., 0., 0.],
    # [0., 1., 0.],
    # [1., 0., 0.],

### 特征选择 ###

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 12][]

#### Filter过滤法 ####

![af22e799eb14461ebe2eeaeed3fbd43d.png][]

**方差过滤**：

![eb8f410744d248eca228e83b06a949f6.png][]

import pandas as pd
    from sklearn.feature_selection import VarianceThreshold
    
    data = pd.read_csv("./digit recognizor.csv")
    
    x = data.iloc[:,1:]
    y = data.iloc[:,0]
    
    #取除特征中方差为0的特征，即所有值都相等的特征
    selector = VarianceThreshold() #实例化，不填参数默认方差为0
    X_var0 = selector.fit_transform(x) #获取删除不合格特征之后的新特征矩阵
    
    X_var0.shape
    #(42000, 784)-》 (42000, 708)

设置阈值来削减特征：

import numpy as np
    
    X.var().values
    #pixel0 0.000000
    #pixel1 0.000000
    #pixel2 0.000000 ...
    
    np.median(X.var().values)
    #1352.2867031797243
    
    #以方差的中位数为阈值，可以削减一半的特征
    X_fsvar = VarianceThreshold(np.median(X.var().values)).fit_transform(X)
    
    X_fsvar.shape
    #(42000, 392)

![e4a1e42ac33d4fcdb134d698c2cbefe9.png][]

#若特征是伯努利随机变量，假设p=0.8，即二分类特征中某种分类占到80%以上的时候删除特征
    X_bvar = VarianceThreshold(.8 * (1 - .8)).fit_transform(x)
    X_bvar.shape
    #(42000, 685)

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 13][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 14][]

#### 相关性过滤 ####

![f37c27fe56234439b99c88cc6702ac39.png][]

**卡方过滤**

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 15][]

from sklearn.ensemble import RandomForestClassifier as RFC
    from sklearn.model_selection import cross_val_score
    from sklearn.feature_selection import SelectKBest
    from sklearn.feature_selection import chi2
    
    #假设在这里我一直我需要300个特征
    X_fschi = SelectKBest(chi2, k=300).fit_transform(X_fsvar, y)
    X_fschi.shape
    #(42000, 300)
    
    #验证模型效果
    cross_val_score(RFC(n_estimators=10,random_state=0),X_fschi,y,cv=5).mean()
    #0.9344761904761905

**选取超参数K**

#绘制学习曲线
    %matplotlib inline
    import matplotlib.pyplot as plt
    
    score = []
    #range(350,200,-10) [350, 340, 330, 320, 310, 300, 290, 280, 270, 260, 250, 240, 230, 220, 210]
    for i in range(350,200,-10):
        X_fschi = SelectKBest(chi2, k=i).fit_transform(X_fsvar, y)
        once = cross_val_score(RFC(n_estimators=10,random_state=0),X_fschi,y,cv=5).mean()
        score.append(once)
        
    plt.plot(range(350,200,-10),score)
    plt.show()

![在这里插入图片描述][watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 16]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 17][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 18][]

chivalue, pvalues_chi = chi2(X_fsvar,y)
    chivalue #卡方值
    #array([ 945664.84392643, 1244766.05139164, 1554872.30384525,
    # 1834161.78305343, 1903618.94085294, 1845226.62427198,
    # 1602117.23307537, 708535.17489837, 974050.20513718 ...]
    
    pvalues_chi #P值
    #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., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0., 0....]
    
    #k取多少？我们想要消除所有p值大于设定值，比如0.05或0.01的特征：
    k = chivalue.shape[0] - (pvalues_chi > 0.05).sum()
    #392
    
    #X_fschi = SelectKBest(chi2, k=填写具体的k).fit_transform(X_fsvar, y)
    #cross_val_score(RFC(n_estimators=10,random_state=0),X_fschi,y,cv=5).mean()

![de1f4f740c9d475c97c2700344bebcd8.png][]

**F检验**

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 19][]

from sklearn.feature_selection import f_classif
    
    F, pvalues_f = f_classif(X_fsvar,y)
    
    F
    #array([1.12236836, 1.69713477, 0.19542821, ..., 3.03522216, 3.73716286, 0.56568448])
    
    pvalues_f
    #array([0.28966539, 0.19296363, 0.65853243, ..., 0.08178345, 0.05349733,0.45215625])
    
    k = F.shape[0] - (pvalues_f > 0.05).sum()
    #392

**互信息法**

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 20][]

from sklearn.feature_selection import mutual_info_classif as MIC
    
    result = MIC(X_fsvar,y)
    
    k = result.shape[0] - sum(result <= 0)
    #392
    #X_fsmic = SelectKBest(MIC, k=填写具体的k).fit_transform(X_fsvar, y)
    #cross_val_score(RFC(n_estimators=10,random_state=0),X_fsmic,y,cv=5).mean()

![1d4d89fad51f4b7ab8a04133076f8741.png][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 21][]

#### Embedded嵌入法 ####

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 22][]

class sklearn.feature_selection.SelectFromModel (estimator, threshold=None, prefit=False, norm_order=1,
    max_features=None)

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 23][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 24][]

from sklearn.feature_selection import SelectFromModel
    from sklearn.ensemble import RandomForestClassifier as RFC
    
    RFC_ = RFC(n_estimators =10,random_state=0)
    X_embedded = SelectFromModel(RFC_,threshold=0.005).fit_transform(X,y)
    #在这里我只想取出来有限的特征。0.005这个阈值对于有780个特征的数据来说，是非常高的阈值，因为平均每个特征只能够分到大约0.001的feature_importances_
    
    X_embedded.shape
    #模型的维度明显被降低了
    #(42000, 47)
    
    #同样的，我们也可以画学习曲线来找最佳阈值
    #======【TIME WARNING：10 mins】======#
    import numpy as np
    import matplotlib.pyplot as plt
    
    RFC_.fit(X,y).feature_importances_
    #array([0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,
    # 0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00,
    # 0.00000000e+00, 0.00000000e+00, 0.00000000e+00, 0.00000000e+00...]
    
    threshold = np.linspace(0,(RFC_.fit(X,y).feature_importances_).max(),20)
    score = []
    for i in threshold:
        X_embedded = SelectFromModel(RFC_,threshold=i).fit_transform(X,y)
        once = cross_val_score(RFC_,X_embedded,y,cv=5).mean()
        score.append(once)
    plt.plot(threshold,score)
    plt.show()

![在这里插入图片描述][watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 25]

#### Wrapper包装法 ####

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 26][]

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 27][]

class sklearn.feature_selection.RFE (estimator, n_features_to_select=None, step=1, verbose=0)

![51aa28fbc11342dc8adb4cc8b22a551e.png][]

from sklearn.feature_selection import RFE
    
    RFC_ = RFC(n_estimators =10,random_state=0)
    selector = RFE(RFC_, n_features_to_select=340, step=50).fit(X, y)
    
    selector.support_.sum()
    #340
    
    selector.ranking_
    #array([10, 9, 8, 7, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6,
    # 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 7, 6, 6,
    # 5, 6, 5, 6, 6, 6, 6, 6, 6, 6, 6, 6, 6, 7, 6, 7, 7,
    
    X_wrapper = selector.transform(X)
    cross_val_score(RFC_,X_wrapper,y,cv=5).mean()
    #0.9379761904761905

score = []
    for i in range(1,751,50):
        X_wrapper = RFE(RFC_,n_features_to_select=i, step=50).fit_transform(x,y)
        once = cross_val_score(RFC_,X_wrapper,y,cv=5).mean()
        score.append(once)
    plt.figure(figsize=[20,5])
    plt.plot(range(1,751,50),score)
    plt.xticks(range(1,751,50))
    plt.show()

![watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 28][]

![07a2a9d113244cfa9446ddd2cfdcae40.png][]

[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16]: /images/20220828/a879c776282742de8688e65b3550595d.png
[54ce298b925644449850896ac947bf34.png]: /images/20220828/5342e3a7a1364536a7d55190e262c4cb.png
[396eeaf78ffa4a218ca84a25c1202992.png]: /images/20220828/f1f0e5796b2d4b13aeab4f8e1e2a7f9a.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 1]: /images/20220828/e1774c91d80a4f35a57ca0ebdd5d596f.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 2]: /images/20220828/7a469df400724923a67c9a0499635f6e.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 3]: /images/20220828/1fd9a6631ece442aba59f825b54f6674.png
[0f345b57e1034046b3885d1af9971b18.png]: /images/20220828/5efb8e3287a440328648a548547404a3.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 4]: /images/20220828/7f505415467545d0a3115bd8b8aa3f4a.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 5]: /images/20220828/a50b4e4b57cf4497b1b9dc39c18b7279.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 6]: /images/20220828/6fae96b8434f4970897f55469274579c.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 7]: /images/20220828/28c7493e07844eb0aaef46266d793c76.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 8]: /images/20220828/4ba8cfbcda4446249e486d189af2fa30.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 9]: /images/20220828/b991bf7aa1cd4bbeb2375dd3832a255d.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 10]: /images/20220828/d5bd2fcc7e194b8e92e10b2393866353.png
[e7924c9ccdef4223920bdf447a4b92ca.png]: /images/20220828/086fd14471b84507b9da612233f4e513.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 11]: /images/20220828/a558e09b9c2c49558f0b6d1a37cab0fd.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 12]: /images/20220828/dd5b2d79c3934dbf8199af5abeb5d9cc.png
[af22e799eb14461ebe2eeaeed3fbd43d.png]: /images/20220828/801c989d835340fc9ac6d9a8d358a98e.png
[eb8f410744d248eca228e83b06a949f6.png]: /images/20220828/0b5937495a3c486cb7300e68b102dbaa.png
[e4a1e42ac33d4fcdb134d698c2cbefe9.png]: /images/20220828/54afc0ec7927467688119d1f90b3995a.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 13]: /images/20220828/acbdb49afce941048dea8bf072770f30.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 14]: /images/20220828/31f6b6ba5ac3469980d1f12cd1d07888.png
[f37c27fe56234439b99c88cc6702ac39.png]: /images/20220828/4f4f4630aa984e6ca362ce0744c7651c.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 15]: /images/20220828/d2445198fb9741eabf8d0312869571f9.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 16]: /images/20220828/14edb07c9c1246c2938777983e4f922e.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 17]: /images/20220828/0fb007364d404df6b224b1669195653a.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 18]: /images/20220828/850c14ff72a441b189dfe0389da6957a.png
[de1f4f740c9d475c97c2700344bebcd8.png]: /images/20220828/6e249fe1e939461690e5e0a6df77de48.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 19]: /images/20220828/76df6fb094de44af96ea5a6ec4625bf6.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 20]: /images/20220828/2b52b95558df4cc096292e42f23b7642.png
[1d4d89fad51f4b7ab8a04133076f8741.png]: /images/20220828/9e93c8d91411425b9dea3e1b735020da.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 21]: /images/20220828/f6271ede6f254160b65d72e40f3ddb0b.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 22]: /images/20220828/620b92291a6b4877ad278550d767546e.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 23]: /images/20220828/20c561c24ae94025beab9754ad788ccb.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 24]: /images/20220828/76bd0c014b374e4fbbd976a2a67750e7.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 25]: /images/20220828/3992f701bd6942358d1d20e60b587b54.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 26]: /images/20220828/d45e74f44992410c8557b70bd8121f5d.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 27]: /images/20220828/ab818b83ffce44b88d2079d07ea98610.png
[51aa28fbc11342dc8adb4cc8b22a551e.png]: /images/20220828/dda4c2406baf4ea7ac1dfe5f4db5538c.png
[watermark_type_ZHJvaWRzYW5zZmFsbGJhY2s_shadow_50_text_Q1NETiBA6aG-5Y2B5pa5_size_20_color_FFFFFF_t_70_g_se_x_16 28]: /images/20220828/115b47919ff0426798b330725b2c9f84.png
[07a2a9d113244cfa9446ddd2cfdcae40.png]: /images/20220828/7ea69ad5db4d4ce1b07614bba4b014b8.png