python备忘录

数值型数据处理_preprocessing

缩放特征数组

import numpy as np
from sklearn import preprocessing

feature=np.array([[-100],[100]])
minmax_scale=preprocessing.MinMaxScaler(feature_range=(0,1))
minmax_scale=prepeocessing.MinMaxScaler(feature_range=(0,1))

scale_feature=minmax_scale.fit_transform(feature)  

#根据中位数/四分位数进行缩放
robust_scaler=preprocessing.RobustScaler()

robust_scaler.fit_transform(x)

特征标准化

x=np.array([[-100],[100]])
scaler=preprocessing.StandardScaler()
standardized=scaler.fit_transform(x)

归一化

from sklearn.prerocessing import Normalizer

normalizer=Normalizer(norm='l2')
normalizer.transform(features)

features_l2_norm=Normalizer(norm='l2').transform(features)
#norm='l1'时，元素总和为1

创建多项式特征
当特征和目标值之间非线性关系时，创建多项式特征

from sklearn.preprocessing import PolynomialFeatures

features=np.array([[-100],[100]])

polynomial_interaction=PolynomialFeatures(degree=2,incluse_bias=Fale)
#degree表示最高项数
interaction=PolynomialFeatures(degree=2,interaction_only=True,include_bias=False)
#仅包含交互特征
polynomial_interaction.fit_transform(features)
interaction.fit_transforms(features)

特征转换

from sklearn.preprocessing import FunctionTransformer
features=np.array([[-100],[100]])

def add_ten(x):
    return x+10

ten_transformer=FunctionTransformer(add_ten)

ten_transformer.transform(features)

特征离散化

from sklearn.preprocessing import Binarizer

#二值化
age=np.array([[6],[12],[18],[36],[64]])
binarizer=Binarizer(18)
binarizer.fit_transform(age)

#离散化
np.digitize(age,bins=[20,30,60])
#bins中数值代表每个区间左边界（左闭右开）
#需要时加入参数right=True，左闭右闭

缺失值填充

#KNN预测缺失值
from fancyimpute import KNN
from sklearn.preprocessing import StandardScaler
from sklearn.datasets import make_blobs

features,_=make_blobs(n_sample=1000,n_feature=2,random_state=1)

scaler=StandardScaler()
standardized_features=scaler.fit_transform(features)

#设定一个缺失值
true_value=standardized_features[0,0]
standardized_features[0,0]=np.nan

features_knn_imputed=KNN(k=5,verbose=0).complete(standardized_features)


#imputer模块，平均数/中位数/众数填充
from sklearn.preprocessing import Imputer

mean_imputer=Imputer(strategy='mean',axis=0)
features_mean_imputed=mean_imputer.fit_transform(features)