sklearn standard scaler

데이터 표준화 적용시, sklearn의 standard scaler를 사용하는 경우, ddof=0의 편향 표준편차 를 사용한다.

https://scikit-learn.org/stable/modules/generated/sklearn.preprocessing.StandardScaler.html

StandardScaler

StandardScaler document에서 ddof=0 의 표준편차를 사용함을 말하고 있다.

 

아래와 같이 샘플을 만들어 Standard Scaler를 적용하면, ddof=0의 std를 사용한 계산과 결과가 같음을 알 수 있다.

 

 

 

In [2]:

import numpy as np
import pandas as pd
import sklearn

 

In [3]:

A=np.random.sample(10)*5+2
B=np.random.sample(10)*2+3
C=np.random.sample(10)*3-1
df=pd.DataFrame({'A':A, 'B':B, 'C':C})
df

 

Out[3]:

	A	B	C
0	3.609154	3.408440	1.151811
1	6.227276	3.419386	0.669558
2	6.363483	4.949257	0.655796
3	4.491161	3.223420	-0.369682
4	2.144723	4.559320	1.054213
5	2.170273	4.978961	1.713408
6	5.584039	3.042569	0.886624
7	5.224212	4.663551	-0.564353
8	4.030533	3.552307	1.247374
9	5.134493	3.350118	1.917439

 

 

In [4]:

from sklearn.preprocessing import StandardScaler
std_scaler=StandardScaler()
print('sklearn standard scaler\n',std_scaler.fit_transform(df))
print('\nddof=1\n',(df-df.mean())/df.std(ddof=1))
print('\nddof=0\n',(df-df.mean())/df.std(ddof=0))

 

 

sklearn standard scaler
 [[-0.62003685 -0.69093336  0.41679539]
 [ 1.20643463 -0.67599531 -0.22010499]
 [ 1.30145605  1.41180608 -0.2382801 ]
 [-0.00472555 -0.94342769 -1.59260461]
 [-1.64166337  0.87966226  0.28790019]
 [-1.62383922  1.45234197  1.15848269]
 [ 0.75769539 -1.19023447  0.06656856]
 [ 0.50667029  1.02190528 -1.84970211]
 [-0.32607169 -0.49459958  0.54300297]
 [ 0.44408032 -0.7705252   1.42794202]]

ddof=1
           A         B         C
0 -0.588219 -0.655477  0.395407
1  1.144524 -0.641305 -0.208810
2  1.234670  1.339357 -0.226052
3 -0.004483 -0.895014 -1.510877
4 -1.557419  0.834521  0.273126
5 -1.540509  1.377813  1.099033
6  0.718813 -1.129156  0.063152
7  0.480670  0.969464 -1.754781
8 -0.309339 -0.469218  0.515138
9  0.421292 -0.730984  1.354665

ddof=0
           A         B         C
0 -0.620037 -0.690933  0.416795
1  1.206435 -0.675995 -0.220105
2  1.301456  1.411806 -0.238280
3 -0.004726 -0.943428 -1.592605
4 -1.641663  0.879662  0.287900
5 -1.623839  1.452342  1.158483
6  0.757695 -1.190234  0.066569
7  0.506670  1.021905 -1.849702
8 -0.326072 -0.494600  0.543003
9  0.444080 -0.770525  1.427942