get_dummies (Pandas) 和 OneHotEncoder (Scikit-learn) 的优缺点是什么? [英] What are the pros and cons between get_dummies (Pandas) and OneHotEncoder (Scikit-learn)?

问题描述

我正在学习将分类变量转换为机器学习分类器的数值的不同方法.我遇到了 pd.get_dummies 方法和 sklearn.preprocessing.OneHotEncoder()，我想看看它们在性能和使用方面有何不同.

I'm learning different methods to convert categorical variables to numeric for machine-learning classifiers. I came across the pd.get_dummies method and sklearn.preprocessing.OneHotEncoder() and I wanted to see how they differed in terms of performance and usage.

我在 https://xgdgsc.wordpress.com/2015/03/20/note-on-using-onehotencoder-in-scikit-learn-to-work-on-categorical-features/ 因为 sklearn 文档对此功能没有太大帮助.我有一种感觉我没有做对......但是

I found a tutorial on how to use OneHotEncoder() on https://xgdgsc.wordpress.com/2015/03/20/note-on-using-onehotencoder-in-scikit-learn-to-work-on-categorical-features/ since the sklearn documentation wasn't too helpful on this feature. I have a feeling I'm not doing it correctly...but

有人可以解释一下使用 pd.dummies 而不是 sklearn.preprocessing.OneHotEncoder() 的利弊吗?反之亦然?我知道OneHotEncoder() 为您提供了一个稀疏矩阵，但除此之外，我不确定它是如何使用的以及与 pandas 方法相比有什么好处.我的使用效率低吗?

Can some explain the pros and cons of using pd.dummies over sklearn.preprocessing.OneHotEncoder() and vice versa? I know that OneHotEncoder() gives you a sparse matrix but other than that I'm not sure how it is used and what the benefits are over the pandas method. Am I using it inefficiently?

import pandas as pd
import numpy as np
from sklearn.datasets import load_iris
sns.set()

%matplotlib inline

#Iris Plot
iris = load_iris()
n_samples, m_features = iris.data.shape

#Load Data
X, y = iris.data, iris.target
D_target_dummy = dict(zip(np.arange(iris.target_names.shape[0]), iris.target_names))

DF_data = pd.DataFrame(X,columns=iris.feature_names)
DF_data["target"] = pd.Series(y).map(D_target_dummy)
#sepal length (cm)  sepal width (cm)  petal length (cm)  petal width (cm)  
#0                  5.1               3.5                1.4               0.2   
#1                  4.9               3.0                1.4               0.2   
#2                  4.7               3.2                1.3               0.2   
#3                  4.6               3.1                1.5               0.2   
#4                  5.0               3.6                1.4               0.2   
#5                  5.4               3.9                1.7               0.4   

DF_dummies = pd.get_dummies(DF_data["target"])
#setosa  versicolor  virginica
#0         1           0          0
#1         1           0          0
#2         1           0          0
#3         1           0          0
#4         1           0          0
#5         1           0          0

from sklearn.preprocessing import OneHotEncoder, LabelEncoder
def f1(DF_data):
    Enc_ohe, Enc_label = OneHotEncoder(), LabelEncoder()
    DF_data["Dummies"] = Enc_label.fit_transform(DF_data["target"])
    DF_dummies2 = pd.DataFrame(Enc_ohe.fit_transform(DF_data[["Dummies"]]).todense(), columns = Enc_label.classes_)
    return(DF_dummies2)

%timeit pd.get_dummies(DF_data["target"])
#1000 loops, best of 3: 777 µs per loop

%timeit f1(DF_data)
#100 loops, best of 3: 2.91 ms per loop

推荐答案

OneHotEncoder 无法直接处理字符串值.如果你的名义特征是字符串，那么你需要先将它们映射成整数.

OneHotEncoder cannot process string values directly. If your nominal features are strings, then you need to first map them into integers.

pandas.get_dummies 正好相反.默认情况下，它只会将字符串列转换为 one-hot 表示，除非指定了列.

pandas.get_dummies is kind of the opposite. By default, it only converts string columns into one-hot representation, unless columns are specified.

这篇关于get_dummies (Pandas) 和 OneHotEncoder (Scikit-learn) 的优缺点是什么?的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！