Python Gower

# Gower 모듈 설치

# 범주형, 연속형 데이터를 가진 행들간의 비유사도 계산

!pip install gower

예제

customers = {

    'age':[22,25,30,34,45,34,50,47,59,62],

    'gender':['m','m','f','m','f','m','f','f','m','f'],

    'marriage':['y','n','n','y','y','n','y','y','n','y'],

    'salary':[3400,3500,4300,3900,4020,4800, 5030,2900,4500,3400],

    'children':[True, False,False,True,True,False,True,False,True,False],

    'purchase_type':['low','low','low','heavy','heavy','low','heavy','low','heavy','low']

}

import pandas as pd

df = pd.DataFrame(customers)

df.index = ['user01','user02','user03','user04','user05','user06','user07','user08','user09','user10']

df

	age	gender	marriage	salary	children	purchase_type
user01	22	m	y	3400	True	low
user02	25	m	n	3500	False	low
user03	30	f	n	4300	False	low
user04	34	m	y	3900	True	heavy
user05	45	f	y	4020	True	heavy
user06	34	m	n	4800	False	low
user07	50	f	y	5030	True	heavy
user08	47	f	y	2900	False	low
user09	59	m	n	4500	True	heavy
user10	62	f	y	3400	False	low

import gower

# 비유사도

gower_matrix = gower.gower_matrix(df) # 각 행간의 비유사도 계산

df_gower = pd.DataFrame(gower_matrix,index=df.index,columns=df.index)

df_gower

	user01	user02	user03	user04	user05	user06	user07	user08	user09	user10
user01	0.000000	0.353658	0.603756	0.255790	0.477680	0.492880	0.577543	0.476624	0.573572	0.500000
user02	0.353658	0.000000	0.250098	0.568799	0.790689	0.139221	0.890552	0.471948	0.553247	0.495325
user03	0.603756	0.250098	0.000000	0.714632	0.584409	0.222457	0.640454	0.347046	0.636483	0.370423
user04	0.255790	0.568799	0.714632	0.000000	0.221890	0.570423	0.321753	0.632414	0.317782	0.655790
user05	0.477680	0.790689	0.584409	0.221890	0.000000	0.773533	0.099863	0.429304	0.429225	0.452680
user06	0.492880	0.139221	0.222457	0.570423	0.773533	0.000000	0.751330	0.536170	0.460974	0.559546
user07	0.577543	0.890552	0.640454	0.321753	0.099863	0.751330	0.000000	0.512500	0.412304	0.510876
user08	0.476624	0.471948	0.347046	0.632414	0.429304	0.536170	0.512500	0.000000	0.841862	0.101624
user09	0.573572	0.553247	0.636483	0.317782	0.429225	0.460974	0.412304	0.841862	0.000000	0.765239
user10	0.500000	0.495325	0.370423	0.655790	0.452680	0.559546	0.510876	0.101624	0.765239	0.000000

from sklearn.cluster import DBSCAN

dbscan = DBSCAN(eps=0.3, min_samples=2, metric='precomputed') # metric='precomputed' : 미리 계산함

dbscan.fit(df_gower)

dbscan.labels_

array([ 0,  1,  1,  0,  0,  1,  0,  2, -1,  2], dtype=int64)

df['cluster'] = dbscan.labels_

df

	age	gender	marriage	salary	children	purchase_type	cluster
user01	22	m	y	3400	True	low	0
user02	25	m	n	3500	False	low	1
user03	30	f	n	4300	False	low	1
user04	34	m	y	3900	True	heavy	0
user05	45	f	y	4020	True	heavy	0
user06	34	m	n	4800	False	low	1
user07	50	f	y	5030	True	heavy	0
user08	47	f	y	2900	False	low	2
user09	59	m	n	4500	True	heavy	-1
user10	62	f	y	3400	False	low	2