feature selection(변수 선택법)¶
필요한 변수만 선택하여 모델의 예측력을 높이거나 과적합을 줄이는 것
알고리즘의 정확도가 유사하다면 변수의 개수가 적을수록 속도가 빠르다
방법
- 단변량(filter) : 특정 컬럼이 또 다른 컬럼과 상관관계가 있는지 확인하는 방법
- 전진/후진 선택법(wrapper) : 컬럼을 하나씩 추가하며 점수 확인 / 하나씩 빼며 점수 확인
- RFE : 사이킷런에서 제공하는 라이브러리를 이용하여 변수의 중요도를 확인하여 변수를 선택하는 방법
- embed(임베드) : 알고리즘의 성질 이용
- 알고리즘 내부에서 각 feature 구송 요소들이 얼마나 사용 되었는지를 알아보는 방식
- 알고리즘들을 이용해 변수의 중요도를 파악, 중요도 낮은 컬럼은 지울 수 있다
- 참고
- RFE와 embed 방식은 특정 알고리즘에 존재하며, 해당 방식을 사용할 수 없는 알고리즘들이 존재한다.(KNN, SVR(kernel=rbf)등 사용 못함)
- Tree계열 알고리즘들은 사용 가능하다(Random Forest, Decision Tree 등)
In [ ]:
# 단변량 : 비슷한 컬럼들을 확인 후 제거하는 것
# embed : Tree 계열 알고리즘에서만 사용 가능, 각 컬럼의 티어를 알려준다.
In [1]:
from sklearn.ensemble import RandomForestRegressor
import pandas as pd
df = pd.read_csv("../data_set/6.회귀/data_cleaning.csv")
df.head()
Out[1]:
| datetime | season | holiday | workingday | weather | temp | atemp | humidity | windspeed | casual | registered | count | year | month | day | hour | temp_int | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 0 | 2011-01-01 00:00:00 | 1 | 0 | 0 | 1 | 9.84 | 14.395 | 81 | 0.0 | 3 | 13 | 16 | 2011 | 1 | 1 | 0 | 9 |
| 1 | 2011-01-01 01:00:00 | 1 | 0 | 0 | 1 | 9.02 | 13.635 | 80 | 0.0 | 8 | 32 | 40 | 2011 | 1 | 1 | 1 | 9 |
| 2 | 2011-01-01 02:00:00 | 1 | 0 | 0 | 1 | 9.02 | 13.635 | 80 | 0.0 | 5 | 27 | 32 | 2011 | 1 | 1 | 2 | 9 |
| 3 | 2011-01-01 03:00:00 | 1 | 0 | 0 | 1 | 9.84 | 14.395 | 75 | 0.0 | 3 | 10 | 13 | 2011 | 1 | 1 | 3 | 9 |
| 4 | 2011-01-01 04:00:00 | 1 | 0 | 0 | 1 | 9.84 | 14.395 | 75 | 0.0 | 0 | 1 | 1 | 2011 | 1 | 1 | 4 | 9 |
In [2]:
df.columns
Out[2]:
Index(['datetime', 'season', 'holiday', 'workingday', 'weather', 'temp',
'atemp', 'humidity', 'windspeed', 'casual', 'registered', 'count',
'year', 'month', 'day', 'hour', 'temp_int'],
dtype='object')
In [3]:
f = ['season', 'holiday', 'workingday', 'weather', 'temp',
'atemp', 'humidity', 'windspeed', 'year', 'month', 'day', 'hour']
l = "count"
X, y = df[f], df[l]
In [4]:
from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
In [5]:
model = RandomForestRegressor()
model.fit(X_train, y_train)
print("train : ", model.score(X_train, y_train))
print("test : ", model.score(X_test, y_test))
train : 0.9916614979489478 test : 0.9410126086888452
In [14]:
# 서로의 상관관계가 높을 수록 그래프가 일정하게 상승 또는 하강한다
import seaborn as sns
sns.scatterplot(data=df, x="temp", y="atemp")
Out[14]:
<Axes: xlabel='temp', ylabel='atemp'>
In [11]:
# 각 컬럼 간의 상관관계를 보여준다
# temp 와 atemp 의 상관관계가 1에 가깝기 때문에
# 제거해도 될듯 하다
df.corr(numeric_only=True)
Out[11]:
| season | holiday | workingday | weather | temp | atemp | humidity | windspeed | casual | registered | count | year | month | day | hour | temp_int | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| season | 1.000000 | 0.029368 | -0.008126 | 0.008879 | 0.258689 | 0.264744 | 0.190610 | -0.147121 | 0.096758 | 0.164011 | 0.163439 | -0.004797 | 0.971524 | 0.001729 | -0.006546 | 0.257917 |
| holiday | 0.029368 | 1.000000 | -0.250491 | -0.007074 | 0.000295 | -0.005215 | 0.001929 | 0.008409 | 0.043799 | -0.020956 | -0.005393 | 0.012021 | 0.001731 | -0.015877 | -0.000354 | 0.000192 |
| workingday | -0.008126 | -0.250491 | 1.000000 | 0.033772 | 0.029966 | 0.024660 | -0.010880 | 0.013373 | -0.319111 | 0.119460 | 0.011594 | -0.002482 | -0.003394 | 0.009829 | 0.002780 | 0.029603 |
| weather | 0.008879 | -0.007074 | 0.033772 | 1.000000 | -0.055035 | -0.055376 | 0.406244 | 0.007261 | -0.135918 | -0.109340 | -0.128655 | -0.012548 | 0.012144 | -0.007890 | -0.022740 | -0.054556 |
| temp | 0.258689 | 0.000295 | 0.029966 | -0.055035 | 1.000000 | 0.984948 | -0.064949 | -0.017852 | 0.467097 | 0.318571 | 0.394454 | 0.061226 | 0.257589 | 0.015551 | 0.145430 | 0.999313 |
| atemp | 0.264744 | -0.005215 | 0.024660 | -0.055376 | 0.984948 | 1.000000 | -0.043536 | -0.057473 | 0.462067 | 0.314635 | 0.389784 | 0.058540 | 0.264173 | 0.011866 | 0.140343 | 0.984431 |
| humidity | 0.190610 | 0.001929 | -0.010880 | 0.406244 | -0.064949 | -0.043536 | 1.000000 | -0.318607 | -0.348187 | -0.265458 | -0.317371 | -0.078606 | 0.204537 | -0.011335 | -0.278011 | -0.064205 |
| windspeed | -0.147121 | 0.008409 | 0.013373 | 0.007261 | -0.017852 | -0.057473 | -0.318607 | 1.000000 | 0.092276 | 0.091052 | 0.101369 | -0.015221 | -0.150192 | 0.036157 | 0.146631 | -0.017660 |
| casual | 0.096758 | 0.043799 | -0.319111 | -0.135918 | 0.467097 | 0.462067 | -0.348187 | 0.092276 | 1.000000 | 0.497250 | 0.690414 | 0.145241 | 0.092722 | 0.014109 | 0.302045 | 0.467047 |
| registered | 0.164011 | -0.020956 | 0.119460 | -0.109340 | 0.318571 | 0.314635 | -0.265458 | 0.091052 | 0.497250 | 1.000000 | 0.970948 | 0.264265 | 0.169451 | 0.019111 | 0.380540 | 0.318048 |
| count | 0.163439 | -0.005393 | 0.011594 | -0.128655 | 0.394454 | 0.389784 | -0.317371 | 0.101369 | 0.690414 | 0.970948 | 1.000000 | 0.260403 | 0.166862 | 0.019826 | 0.400601 | 0.394003 |
| year | -0.004797 | 0.012021 | -0.002482 | -0.012548 | 0.061226 | 0.058540 | -0.078606 | -0.015221 | 0.145241 | 0.264265 | 0.260403 | 1.000000 | -0.004932 | 0.001800 | -0.004234 | 0.060692 |
| month | 0.971524 | 0.001731 | -0.003394 | 0.012144 | 0.257589 | 0.264173 | 0.204537 | -0.150192 | 0.092722 | 0.169451 | 0.166862 | -0.004932 | 1.000000 | 0.001974 | -0.006818 | 0.256862 |
| day | 0.001729 | -0.015877 | 0.009829 | -0.007890 | 0.015551 | 0.011866 | -0.011335 | 0.036157 | 0.014109 | 0.019111 | 0.019826 | 0.001800 | 0.001974 | 1.000000 | 0.001132 | 0.016202 |
| hour | -0.006546 | -0.000354 | 0.002780 | -0.022740 | 0.145430 | 0.140343 | -0.278011 | 0.146631 | 0.302045 | 0.380540 | 0.400601 | -0.004234 | -0.006818 | 0.001132 | 1.000000 | 0.145353 |
| temp_int | 0.257917 | 0.000192 | 0.029603 | -0.054556 | 0.999313 | 0.984431 | -0.064205 | -0.017660 | 0.467047 | 0.318048 | 0.394003 | 0.060692 | 0.256862 | 0.016202 | 0.145353 | 1.000000 |
In [13]:
import matplotlib.pyplot as plt
plt.subplots(figsize=(12,12))
sns.heatmap(df.corr(numeric_only=True), annot=True)
Out[13]:
<Axes: >
In [16]:
# 상관관계가 비슷한 것들은 / 이 형태나 \ 이 형태로 값이
# 일정하게 줄어들거나 늘어난다
sns.pairplot(df, height=1)
Out[16]:
<seaborn.axisgrid.PairGrid at 0x16a895c6710>
In [18]:
# atemp 삭제
f = ['season', 'holiday', 'workingday', 'weather', 'temp',
'humidity', 'windspeed', 'year', 'month', 'day', 'hour']
# atemp 컬럼을 삭제해도 예측 결과 값에는 영향을 크게 끼치지 않았음
# atemp 와 temp 가 거의 동일한 컬럼이었기 때문...!
X_train, X_test, y_train, y_test = train_test_split(df[f], df[l], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
print("train : ", model.score(X_train, y_train))
print("test : ", model.score(X_test, y_test))
train : 0.9917278017758199 test : 0.9399504472888517
In [ ]:
In [ ]:
wrapper(전진/후진 선택)¶
- feature를 추가 / 제거하면서 모델의 성능을 확인한다
- combinations : 모든 값들의 경우의 수를 뽑아준다
In [22]:
from itertools import combinations
sample_bag = [1,2,3,4]
for c in combinations(sample_bag, 2) :
# sample_bag 를 2개씩 뽑으라는 뜻
print(c, type(c))
(1, 2) <class 'tuple'> (1, 3) <class 'tuple'> (1, 4) <class 'tuple'> (2, 3) <class 'tuple'> (2, 4) <class 'tuple'> (3, 4) <class 'tuple'>
In [25]:
all_result = []
# combinations() 가 컬럼들을 2개씩 묶어준다
for c in combinations(f, 2):
print(c)
('season', 'holiday')
('season', 'workingday')
('season', 'weather')
('season', 'temp')
('season', 'humidity')
('season', 'windspeed')
('season', 'year')
('season', 'month')
('season', 'day')
('season', 'hour')
('holiday', 'workingday')
('holiday', 'weather')
('holiday', 'temp')
('holiday', 'humidity')
('holiday', 'windspeed')
('holiday', 'year')
('holiday', 'month')
('holiday', 'day')
('holiday', 'hour')
('workingday', 'weather')
('workingday', 'temp')
('workingday', 'humidity')
('workingday', 'windspeed')
('workingday', 'year')
('workingday', 'month')
('workingday', 'day')
('workingday', 'hour')
('weather', 'temp')
('weather', 'humidity')
('weather', 'windspeed')
('weather', 'year')
('weather', 'month')
('weather', 'day')
('weather', 'hour')
('temp', 'humidity')
('temp', 'windspeed')
('temp', 'year')
('temp', 'month')
('temp', 'day')
('temp', 'hour')
('humidity', 'windspeed')
('humidity', 'year')
('humidity', 'month')
('humidity', 'day')
('humidity', 'hour')
('windspeed', 'year')
('windspeed', 'month')
('windspeed', 'day')
('windspeed', 'hour')
('year', 'month')
('year', 'day')
('year', 'hour')
('month', 'day')
('month', 'hour')
('day', 'hour')
In [27]:
for c in combinations(f, 2):
# 컬럼을 list 자료형으로 형변환해서 사용해야 한다
print( df[list(c)] )
season holiday
0 1 0
1 1 0
2 1 0
3 1 0
4 1 0
... ... ...
10881 4 0
10882 4 0
10883 4 0
10884 4 0
10885 4 0
[10886 rows x 2 columns]
season workingday
0 1 0
1 1 0
2 1 0
3 1 0
4 1 0
... ... ...
10881 4 1
10882 4 1
10883 4 1
10884 4 1
10885 4 1
[10886 rows x 2 columns]
season weather
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1
... ... ...
10881 4 1
10882 4 1
10883 4 1
10884 4 1
10885 4 1
[10886 rows x 2 columns]
season temp
0 1 9.84
1 1 9.02
2 1 9.02
3 1 9.84
4 1 9.84
... ... ...
10881 4 15.58
10882 4 14.76
10883 4 13.94
10884 4 13.94
10885 4 13.12
[10886 rows x 2 columns]
season humidity
0 1 81
1 1 80
2 1 80
3 1 75
4 1 75
... ... ...
10881 4 50
10882 4 57
10883 4 61
10884 4 61
10885 4 66
[10886 rows x 2 columns]
season windspeed
0 1 0.0000
1 1 0.0000
2 1 0.0000
3 1 0.0000
4 1 0.0000
... ... ...
10881 4 26.0027
10882 4 15.0013
10883 4 15.0013
10884 4 6.0032
10885 4 8.9981
[10886 rows x 2 columns]
season year
0 1 2011
1 1 2011
2 1 2011
3 1 2011
4 1 2011
... ... ...
10881 4 2012
10882 4 2012
10883 4 2012
10884 4 2012
10885 4 2012
[10886 rows x 2 columns]
season month
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1
... ... ...
10881 4 12
10882 4 12
10883 4 12
10884 4 12
10885 4 12
[10886 rows x 2 columns]
season day
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1
... ... ...
10881 4 19
10882 4 19
10883 4 19
10884 4 19
10885 4 19
[10886 rows x 2 columns]
season hour
0 1 0
1 1 1
2 1 2
3 1 3
4 1 4
... ... ...
10881 4 19
10882 4 20
10883 4 21
10884 4 22
10885 4 23
[10886 rows x 2 columns]
holiday workingday
0 0 0
1 0 0
2 0 0
3 0 0
4 0 0
... ... ...
10881 0 1
10882 0 1
10883 0 1
10884 0 1
10885 0 1
[10886 rows x 2 columns]
holiday weather
0 0 1
1 0 1
2 0 1
3 0 1
4 0 1
... ... ...
10881 0 1
10882 0 1
10883 0 1
10884 0 1
10885 0 1
[10886 rows x 2 columns]
holiday temp
0 0 9.84
1 0 9.02
2 0 9.02
3 0 9.84
4 0 9.84
... ... ...
10881 0 15.58
10882 0 14.76
10883 0 13.94
10884 0 13.94
10885 0 13.12
[10886 rows x 2 columns]
holiday humidity
0 0 81
1 0 80
2 0 80
3 0 75
4 0 75
... ... ...
10881 0 50
10882 0 57
10883 0 61
10884 0 61
10885 0 66
[10886 rows x 2 columns]
holiday windspeed
0 0 0.0000
1 0 0.0000
2 0 0.0000
3 0 0.0000
4 0 0.0000
... ... ...
10881 0 26.0027
10882 0 15.0013
10883 0 15.0013
10884 0 6.0032
10885 0 8.9981
[10886 rows x 2 columns]
holiday year
0 0 2011
1 0 2011
2 0 2011
3 0 2011
4 0 2011
... ... ...
10881 0 2012
10882 0 2012
10883 0 2012
10884 0 2012
10885 0 2012
[10886 rows x 2 columns]
holiday month
0 0 1
1 0 1
2 0 1
3 0 1
4 0 1
... ... ...
10881 0 12
10882 0 12
10883 0 12
10884 0 12
10885 0 12
[10886 rows x 2 columns]
holiday day
0 0 1
1 0 1
2 0 1
3 0 1
4 0 1
... ... ...
10881 0 19
10882 0 19
10883 0 19
10884 0 19
10885 0 19
[10886 rows x 2 columns]
holiday hour
0 0 0
1 0 1
2 0 2
3 0 3
4 0 4
... ... ...
10881 0 19
10882 0 20
10883 0 21
10884 0 22
10885 0 23
[10886 rows x 2 columns]
workingday weather
0 0 1
1 0 1
2 0 1
3 0 1
4 0 1
... ... ...
10881 1 1
10882 1 1
10883 1 1
10884 1 1
10885 1 1
[10886 rows x 2 columns]
workingday temp
0 0 9.84
1 0 9.02
2 0 9.02
3 0 9.84
4 0 9.84
... ... ...
10881 1 15.58
10882 1 14.76
10883 1 13.94
10884 1 13.94
10885 1 13.12
[10886 rows x 2 columns]
workingday humidity
0 0 81
1 0 80
2 0 80
3 0 75
4 0 75
... ... ...
10881 1 50
10882 1 57
10883 1 61
10884 1 61
10885 1 66
[10886 rows x 2 columns]
workingday windspeed
0 0 0.0000
1 0 0.0000
2 0 0.0000
3 0 0.0000
4 0 0.0000
... ... ...
10881 1 26.0027
10882 1 15.0013
10883 1 15.0013
10884 1 6.0032
10885 1 8.9981
[10886 rows x 2 columns]
workingday year
0 0 2011
1 0 2011
2 0 2011
3 0 2011
4 0 2011
... ... ...
10881 1 2012
10882 1 2012
10883 1 2012
10884 1 2012
10885 1 2012
[10886 rows x 2 columns]
workingday month
0 0 1
1 0 1
2 0 1
3 0 1
4 0 1
... ... ...
10881 1 12
10882 1 12
10883 1 12
10884 1 12
10885 1 12
[10886 rows x 2 columns]
workingday day
0 0 1
1 0 1
2 0 1
3 0 1
4 0 1
... ... ...
10881 1 19
10882 1 19
10883 1 19
10884 1 19
10885 1 19
[10886 rows x 2 columns]
workingday hour
0 0 0
1 0 1
2 0 2
3 0 3
4 0 4
... ... ...
10881 1 19
10882 1 20
10883 1 21
10884 1 22
10885 1 23
[10886 rows x 2 columns]
weather temp
0 1 9.84
1 1 9.02
2 1 9.02
3 1 9.84
4 1 9.84
... ... ...
10881 1 15.58
10882 1 14.76
10883 1 13.94
10884 1 13.94
10885 1 13.12
[10886 rows x 2 columns]
weather humidity
0 1 81
1 1 80
2 1 80
3 1 75
4 1 75
... ... ...
10881 1 50
10882 1 57
10883 1 61
10884 1 61
10885 1 66
[10886 rows x 2 columns]
weather windspeed
0 1 0.0000
1 1 0.0000
2 1 0.0000
3 1 0.0000
4 1 0.0000
... ... ...
10881 1 26.0027
10882 1 15.0013
10883 1 15.0013
10884 1 6.0032
10885 1 8.9981
[10886 rows x 2 columns]
weather year
0 1 2011
1 1 2011
2 1 2011
3 1 2011
4 1 2011
... ... ...
10881 1 2012
10882 1 2012
10883 1 2012
10884 1 2012
10885 1 2012
[10886 rows x 2 columns]
weather month
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1
... ... ...
10881 1 12
10882 1 12
10883 1 12
10884 1 12
10885 1 12
[10886 rows x 2 columns]
weather day
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1
... ... ...
10881 1 19
10882 1 19
10883 1 19
10884 1 19
10885 1 19
[10886 rows x 2 columns]
weather hour
0 1 0
1 1 1
2 1 2
3 1 3
4 1 4
... ... ...
10881 1 19
10882 1 20
10883 1 21
10884 1 22
10885 1 23
[10886 rows x 2 columns]
temp humidity
0 9.84 81
1 9.02 80
2 9.02 80
3 9.84 75
4 9.84 75
... ... ...
10881 15.58 50
10882 14.76 57
10883 13.94 61
10884 13.94 61
10885 13.12 66
[10886 rows x 2 columns]
temp windspeed
0 9.84 0.0000
1 9.02 0.0000
2 9.02 0.0000
3 9.84 0.0000
4 9.84 0.0000
... ... ...
10881 15.58 26.0027
10882 14.76 15.0013
10883 13.94 15.0013
10884 13.94 6.0032
10885 13.12 8.9981
[10886 rows x 2 columns]
temp year
0 9.84 2011
1 9.02 2011
2 9.02 2011
3 9.84 2011
4 9.84 2011
... ... ...
10881 15.58 2012
10882 14.76 2012
10883 13.94 2012
10884 13.94 2012
10885 13.12 2012
[10886 rows x 2 columns]
temp month
0 9.84 1
1 9.02 1
2 9.02 1
3 9.84 1
4 9.84 1
... ... ...
10881 15.58 12
10882 14.76 12
10883 13.94 12
10884 13.94 12
10885 13.12 12
[10886 rows x 2 columns]
temp day
0 9.84 1
1 9.02 1
2 9.02 1
3 9.84 1
4 9.84 1
... ... ...
10881 15.58 19
10882 14.76 19
10883 13.94 19
10884 13.94 19
10885 13.12 19
[10886 rows x 2 columns]
temp hour
0 9.84 0
1 9.02 1
2 9.02 2
3 9.84 3
4 9.84 4
... ... ...
10881 15.58 19
10882 14.76 20
10883 13.94 21
10884 13.94 22
10885 13.12 23
[10886 rows x 2 columns]
humidity windspeed
0 81 0.0000
1 80 0.0000
2 80 0.0000
3 75 0.0000
4 75 0.0000
... ... ...
10881 50 26.0027
10882 57 15.0013
10883 61 15.0013
10884 61 6.0032
10885 66 8.9981
[10886 rows x 2 columns]
humidity year
0 81 2011
1 80 2011
2 80 2011
3 75 2011
4 75 2011
... ... ...
10881 50 2012
10882 57 2012
10883 61 2012
10884 61 2012
10885 66 2012
[10886 rows x 2 columns]
humidity month
0 81 1
1 80 1
2 80 1
3 75 1
4 75 1
... ... ...
10881 50 12
10882 57 12
10883 61 12
10884 61 12
10885 66 12
[10886 rows x 2 columns]
humidity day
0 81 1
1 80 1
2 80 1
3 75 1
4 75 1
... ... ...
10881 50 19
10882 57 19
10883 61 19
10884 61 19
10885 66 19
[10886 rows x 2 columns]
humidity hour
0 81 0
1 80 1
2 80 2
3 75 3
4 75 4
... ... ...
10881 50 19
10882 57 20
10883 61 21
10884 61 22
10885 66 23
[10886 rows x 2 columns]
windspeed year
0 0.0000 2011
1 0.0000 2011
2 0.0000 2011
3 0.0000 2011
4 0.0000 2011
... ... ...
10881 26.0027 2012
10882 15.0013 2012
10883 15.0013 2012
10884 6.0032 2012
10885 8.9981 2012
[10886 rows x 2 columns]
windspeed month
0 0.0000 1
1 0.0000 1
2 0.0000 1
3 0.0000 1
4 0.0000 1
... ... ...
10881 26.0027 12
10882 15.0013 12
10883 15.0013 12
10884 6.0032 12
10885 8.9981 12
[10886 rows x 2 columns]
windspeed day
0 0.0000 1
1 0.0000 1
2 0.0000 1
3 0.0000 1
4 0.0000 1
... ... ...
10881 26.0027 19
10882 15.0013 19
10883 15.0013 19
10884 6.0032 19
10885 8.9981 19
[10886 rows x 2 columns]
windspeed hour
0 0.0000 0
1 0.0000 1
2 0.0000 2
3 0.0000 3
4 0.0000 4
... ... ...
10881 26.0027 19
10882 15.0013 20
10883 15.0013 21
10884 6.0032 22
10885 8.9981 23
[10886 rows x 2 columns]
year month
0 2011 1
1 2011 1
2 2011 1
3 2011 1
4 2011 1
... ... ...
10881 2012 12
10882 2012 12
10883 2012 12
10884 2012 12
10885 2012 12
[10886 rows x 2 columns]
year day
0 2011 1
1 2011 1
2 2011 1
3 2011 1
4 2011 1
... ... ...
10881 2012 19
10882 2012 19
10883 2012 19
10884 2012 19
10885 2012 19
[10886 rows x 2 columns]
year hour
0 2011 0
1 2011 1
2 2011 2
3 2011 3
4 2011 4
... ... ...
10881 2012 19
10882 2012 20
10883 2012 21
10884 2012 22
10885 2012 23
[10886 rows x 2 columns]
month day
0 1 1
1 1 1
2 1 1
3 1 1
4 1 1
... ... ...
10881 12 19
10882 12 19
10883 12 19
10884 12 19
10885 12 19
[10886 rows x 2 columns]
month hour
0 1 0
1 1 1
2 1 2
3 1 3
4 1 4
... ... ...
10881 12 19
10882 12 20
10883 12 21
10884 12 22
10885 12 23
[10886 rows x 2 columns]
day hour
0 1 0
1 1 1
2 1 2
3 1 3
4 1 4
... ... ...
10881 19 19
10882 19 20
10883 19 21
10884 19 22
10885 19 23
[10886 rows x 2 columns]
In [29]:
for c in combinations(f, 2):
X_train, X_test, y_train, y_test = train_test_split(df[list(c)], df['count'], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
train_score = model.score(X_train, y_train)
test_score = model.score(X_test, y_test)
print(str(c))
print("train : ", train_score)
print("test : ", test_score)
print("="*20)
('season', 'holiday')
train : 0.05845819571136379
test : 0.06801029389571611
====================
('season', 'workingday')
train : 0.06119690546536827
test : 0.06356368452321925
====================
('season', 'weather')
train : 0.07576082202281542
test : 0.08724412927488545
====================
('season', 'temp')
train : 0.23284427316676115
test : 0.2126150527730073
====================
('season', 'humidity')
train : 0.28060963515764825
test : 0.2553308376517509
====================
('season', 'windspeed')
train : 0.09695952795617624
test : 0.0870184127764726
====================
('season', 'year')
train : 0.12481734997659888
test : 0.1419045806133954
====================
('season', 'month')
train : 0.07110487449658509
test : 0.07775741473009401
====================
('season', 'day')
train : 0.0730837414840454
test : 0.0508598813046125
====================
('season', 'hour')
train : 0.6119506571560571
test : 0.603707014633353
====================
('holiday', 'workingday')
train : 6.254614111333989e-05
test : -7.213416623663704e-05
====================
('holiday', 'weather')
train : 0.018203745414681616
test : 0.017385163764621958
====================
('holiday', 'temp')
train : 0.16992644518592026
test : 0.18105496093542683
====================
('holiday', 'humidity')
train : 0.13910601748660945
test : 0.1188547039149569
====================
('holiday', 'windspeed')
train : 0.019516032040894027
test : 0.01644872706450229
====================
('holiday', 'year')
train : 0.07140907585286749
test : 0.05987754943200241
====================
('holiday', 'month')
train : 0.07170470686238317
test : 0.07658428013922136
====================
('holiday', 'day')
train : 0.007090447830976676
test : -0.002347600544694739
====================
('holiday', 'hour')
train : 0.5203690714412798
test : 0.5147726988065502
====================
('workingday', 'weather')
train : 0.017652970990177486
test : 0.018520313691839374
====================
('workingday', 'temp')
train : 0.17701773948619048
test : 0.1656941296996629
====================
('workingday', 'humidity')
train : 0.1458885900358774
test : 0.12107148034198867
====================
('workingday', 'windspeed')
train : 0.025180747003014092
test : 0.009864892264444092
====================
('workingday', 'year')
train : 0.06463100782531617
test : 0.0758191172864463
====================
('workingday', 'month')
train : 0.07721271823416076
test : 0.0669414583433603
====================
('workingday', 'day')
train : 0.00513734237636676
test : -0.0019346627158343122
====================
('workingday', 'hour')
train : 0.6538749506801984
test : 0.6533623225870554
====================
('weather', 'temp')
train : 0.19434607518171065
test : 0.17027741524867768
====================
('weather', 'humidity')
train : 0.15828235468992424
test : 0.10505369754010518
====================
('weather', 'windspeed')
train : 0.04406413642063167
test : 0.03658864299539133
====================
('weather', 'year')
train : 0.08756829553309142
test : 0.07928282501967343
====================
('weather', 'month')
train : 0.09587192759658014
test : 0.08846831294974555
====================
('weather', 'day')
train : 0.030841733528468973
test : 0.009799462464378794
====================
('weather', 'hour')
train : 0.5532384362164128
test : 0.5332867151807875
====================
('temp', 'humidity')
train : 0.3686949135390849
test : 0.24043607129139977
====================
('temp', 'windspeed')
train : 0.2590220131987515
test : 0.11573756231088816
====================
('temp', 'year')
train : 0.23753762757820374
test : 0.21815533604779336
====================
('temp', 'month')
train : 0.2880174693849482
test : 0.22970742981217374
====================
('temp', 'day')
train : 0.2663310001326936
test : 0.14210666955745732
====================
('temp', 'hour')
train : 0.679074285859746
test : 0.6037291739012631
====================
('humidity', 'windspeed')
train : 0.25819049064192934
test : 0.036960054751834504
====================
('humidity', 'year')
train : 0.20136505635494273
test : 0.18650522639089195
====================
('humidity', 'month')
train : 0.38909825925915065
test : 0.2496396608260928
====================
('humidity', 'day')
train : 0.3194661339783871
test : 0.06093953967038068
====================
('humidity', 'hour')
train : 0.6346694753826874
test : 0.5050059135565201
====================
('windspeed', 'year')
train : 0.09323464953999572
test : 0.08183108426857966
====================
('windspeed', 'month')
train : 0.13519972851366324
test : 0.06816860927886215
====================
('windspeed', 'day')
train : 0.06735231888334947
test : -0.012116863497556363
====================
('windspeed', 'hour')
train : 0.5454843269854666
test : 0.4981731533709608
====================
('year', 'month')
train : 0.14949221759923648
test : 0.13777955850772106
====================
('year', 'day')
train : 0.06481644985431456
test : 0.07503715410081335
====================
('year', 'hour')
train : 0.6219452481465291
test : 0.6155966763154723
====================
('month', 'day')
train : 0.10347692407809816
test : 0.0417316959498103
====================
('month', 'hour')
train : 0.6344925826836026
test : 0.6095869000632028
====================
('day', 'hour')
train : 0.5278468147569957
test : 0.47347898020353973
====================
In [30]:
for c in combinations(f, 2):
X_train, X_test, y_train, y_test = train_test_split(df[list(c)], df['count'], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
train_score = model.score(X_train, y_train)
test_score = model.score(X_test, y_test)
result = {"combination" : str(c), "train" : train_score, "test" : test_score}
all_result.append(result)
all_result
Out[30]:
[{'combination': "('season', 'holiday')",
'train': 0.06429541388996607,
'test': 0.05439025396453023},
{'combination': "('season', 'workingday')",
'train': 0.06534159774564652,
'test': 0.05273019184958949},
{'combination': "('season', 'weather')",
'train': 0.08216391857162053,
'test': 0.07265795524806051},
{'combination': "('season', 'temp')",
'train': 0.2351771066164272,
'test': 0.20490765892966023},
{'combination': "('season', 'humidity')",
'train': 0.2856744999467695,
'test': 0.24563783000940131},
{'combination': "('season', 'windspeed')",
'train': 0.09543727422483361,
'test': 0.08810482783949436},
{'combination': "('season', 'year')",
'train': 0.13202699951491037,
'test': 0.12640899224946056},
{'combination': "('season', 'month')",
'train': 0.0741459578302236,
'test': 0.06991633720348411},
{'combination': "('season', 'day')",
'train': 0.07435037626015506,
'test': 0.04502838971546008},
{'combination': "('season', 'hour')",
'train': 0.6064060557910389,
'test': 0.6161882221308241},
{'combination': "('holiday', 'workingday')",
'train': 9.283914579183428e-05,
'test': -0.0006753884944799005},
{'combination': "('holiday', 'weather')",
'train': 0.01698704128836559,
'test': 0.020411523484866634},
{'combination': "('holiday', 'temp')",
'train': 0.17543847668694112,
'test': 0.16892160464232586},
{'combination': "('holiday', 'humidity')",
'train': 0.1519459899428811,
'test': 0.08329276861928858},
{'combination': "('holiday', 'windspeed')",
'train': 0.022829782512737373,
'test': 0.008072210959060855},
{'combination': "('holiday', 'year')",
'train': 0.07105795444215501,
'test': 0.06048052370486012},
{'combination': "('holiday', 'month')",
'train': 0.07507252836735323,
'test': 0.06951961287107877},
{'combination': "('holiday', 'day')",
'train': 0.0068259510156423175,
'test': 0.003024065717116664},
{'combination': "('holiday', 'hour')",
'train': 0.5228490974196865,
'test': 0.5080892804225323},
{'combination': "('workingday', 'weather')",
'train': 0.020828301449846043,
'test': 0.011411014014170995},
{'combination': "('workingday', 'temp')",
'train': 0.17502869489703998,
'test': 0.17364869335565458},
{'combination': "('workingday', 'humidity')",
'train': 0.14476689394463738,
'test': 0.12851029339806597},
{'combination': "('workingday', 'windspeed')",
'train': 0.023235429159391696,
'test': 0.01615716563294245},
{'combination': "('workingday', 'year')",
'train': 0.07277112893558291,
'test': 0.05661629910846899},
{'combination': "('workingday', 'month')",
'train': 0.07715409638606674,
'test': 0.06466835772871471},
{'combination': "('workingday', 'day')",
'train': 0.005231127635079602,
'test': -0.0036445773889950406},
{'combination': "('workingday', 'hour')",
'train': 0.65163346772907,
'test': 0.65980453848609},
{'combination': "('weather', 'temp')",
'train': 0.1952024653027693,
'test': 0.16654321420746143},
{'combination': "('weather', 'humidity')",
'train': 0.15658685847221743,
'test': 0.10298867496824105},
{'combination': "('weather', 'windspeed')",
'train': 0.0474964097877556,
'test': 0.02771650824026939},
{'combination': "('weather', 'year')",
'train': 0.09116637560345542,
'test': 0.07078163179731245},
{'combination': "('weather', 'month')",
'train': 0.09422581958616993,
'test': 0.09301492937102684},
{'combination': "('weather', 'day')",
'train': 0.02716329585565125,
'test': 0.019863988688087142},
{'combination': "('weather', 'hour')",
'train': 0.5558689455886267,
'test': 0.5285109716235332},
{'combination': "('temp', 'humidity')",
'train': 0.37535892235591173,
'test': 0.23880588782898804},
{'combination': "('temp', 'windspeed')",
'train': 0.25899025839398493,
'test': 0.12629183224336316},
{'combination': "('temp', 'year')",
'train': 0.2356678407766435,
'test': 0.22465200787573958},
{'combination': "('temp', 'month')",
'train': 0.28378494720565084,
'test': 0.2352749739844313},
{'combination': "('temp', 'day')",
'train': 0.2698911501769191,
'test': 0.13544182234706081},
{'combination': "('temp', 'hour')",
'train': 0.6904494294728518,
'test': 0.5739633601701017},
{'combination': "('humidity', 'windspeed')",
'train': 0.2705701220348399,
'test': 0.01345498635788167},
{'combination': "('humidity', 'year')",
'train': 0.20736843182661158,
'test': 0.17373936030535997},
{'combination': "('humidity', 'month')",
'train': 0.386774099530133,
'test': 0.2711809725326212},
{'combination': "('humidity', 'day')",
'train': 0.31456450734972696,
'test': 0.06626831878009554},
{'combination': "('humidity', 'hour')",
'train': 0.6484746860145716,
'test': 0.4599222799053242},
{'combination': "('windspeed', 'year')",
'train': 0.094354903191749,
'test': 0.0771786967680903},
{'combination': "('windspeed', 'month')",
'train': 0.13161596556156407,
'test': 0.08218792527707564},
{'combination': "('windspeed', 'day')",
'train': 0.07047871647514425,
'test': -0.018641448646160574},
{'combination': "('windspeed', 'hour')",
'train': 0.553397149647815,
'test': 0.48190568850884374},
{'combination': "('year', 'month')",
'train': 0.14710504454494133,
'test': 0.1424721899303245},
{'combination': "('year', 'day')",
'train': 0.07342932178624018,
'test': 0.05346903169933215},
{'combination': "('year', 'hour')",
'train': 0.6208178463861078,
'test': 0.6179964590935694},
{'combination': "('month', 'day')",
'train': 0.09841937927428457,
'test': 0.05207547071903362},
{'combination': "('month', 'hour')",
'train': 0.6386774092912783,
'test': 0.6014056338728897},
{'combination': "('day', 'hour')",
'train': 0.5235915799173942,
'test': 0.48925313013863636}]
In [37]:
# test 를 기준으로 오름차순으로 정렬
result_df = pd.DataFrame(all_result).sort_values(by="test")
# test 의 점수가 가장 높은 5가지를 확인
result_df.tail()
Out[37]:
| combination | train | test | |
|---|---|---|---|
| 9 | ('season', 'hour') | 0.606406 | 0.616188 |
| 51 | ('year', 'hour') | 0.620818 | 0.617996 |
| 81 | ('workingday', 'hour') | 0.657728 | 0.644720 |
| 136 | ('workingday', 'hour') | 0.657899 | 0.644887 |
| 26 | ('workingday', 'hour') | 0.651633 | 0.659805 |
In [38]:
fe = ['season', 'holiday', 'workingday', 'weather', 'temp',
'humidity', 'windspeed', 'year', 'month', 'day', 'hour']
# 가장 좋은 조합
best_f = ["workingday", "hour"]
In [41]:
all_result = []
for f in fe:
# 가장 좋은 조합을 확인하기 위해
# feature 하나씩 추가해서 확인
best_f.append(f)
# print(best_f)
X_train, X_test, y_train, y_test = train_test_split(df[best_f], df['count'], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
train_s = model.score(X_train, y_train)
test_s = model.score(X_test, y_test)
result = {"combin" : best_f.copy(), "train" : train_s, "test" : test_s}
all_result.append(result)
# 다음 feature를 넣어서 확인하기 위해 마지막에 넣은 feature 제거
best_f.pop()
# print(best_f)
all_result
Out[41]:
[{'combin': ['workingday', 'hour', 'season'],
'train': 0.7580913458160565,
'test': 0.7440944945103845},
{'combin': ['workingday', 'hour', 'holiday'],
'train': 0.6650019150696596,
'test': 0.6345567260841918},
{'combin': ['workingday', 'hour', 'workingday'],
'train': 0.6590593073097475,
'test': 0.6431016423118261},
{'combin': ['workingday', 'hour', 'weather'],
'train': 0.6926684259802643,
'test': 0.6668174416876799},
{'combin': ['workingday', 'hour', 'temp'],
'train': 0.8323620614455098,
'test': 0.7295907373392445},
{'combin': ['workingday', 'hour', 'humidity'],
'train': 0.8004993632245669,
'test': 0.600456524550598},
{'combin': ['workingday', 'hour', 'windspeed'],
'train': 0.6990780034197865,
'test': 0.6266172439536393},
{'combin': ['workingday', 'hour', 'year'],
'train': 0.767761505426242,
'test': 0.7636787124282669},
{'combin': ['workingday', 'hour', 'month'],
'train': 0.7863534443871265,
'test': 0.7419456112733107},
{'combin': ['workingday', 'hour', 'day'],
'train': 0.6762676095319997,
'test': 0.5931861225822459},
{'combin': ['workingday', 'hour', 'hour'],
'train': 0.6560251705446849,
'test': 0.6500550217881965}]
In [42]:
# test 를 기준으로 오름차순으로 정렬
result_df = pd.DataFrame(all_result).sort_values(by="test")
# test 의 점수가 가장 높은 5가지를 확인
result_df.tail()
Out[42]:
| combin | train | test | |
|---|---|---|---|
| 3 | [workingday, hour, weather] | 0.692668 | 0.666817 |
| 4 | [workingday, hour, temp] | 0.832362 | 0.729591 |
| 8 | [workingday, hour, month] | 0.786353 | 0.741946 |
| 0 | [workingday, hour, season] | 0.758091 | 0.744094 |
| 7 | [workingday, hour, year] | 0.767762 | 0.763679 |
In [50]:
# 함수로 위 과정 처리
def features_select(feature, best_f):
all_result = []
for f in feature:
# 가장 좋은 조합을 확인하기 위해
# feature 하나씩 추가해서 확인
best_f.append(f)
# print(best_f)
X_train, X_test, y_train, y_test = train_test_split(df[best_f], df['count'], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
train_s = model.score(X_train, y_train)
test_s = model.score(X_test, y_test)
result = {"combin" : best_f.copy(), "train" : train_s, "test" : test_s}
all_result.append(result)
# 다음 feature를 넣어서 확인하기 위해 마지막에 넣은 feature 제거
best_f.pop()
# print(best_f)
return all_result;
In [51]:
fe = ['season', 'holiday', 'workingday', 'weather', 'temp',
'humidity', 'windspeed', 'month', 'day', 'hour']
# 가장 좋은 조합
best_f = ["workingday", "hour", "year"]
In [52]:
# test 를 기준으로 오름차순으로 정렬
result_df = pd.DataFrame(features_select(fe, best_f)).sort_values(by="test")
# test 의 점수가 가장 높은 5가지를 확인
result_df.tail()
Out[52]:
| combin | train | test | |
|---|---|---|---|
| 2 | [workingday, hour, year, workingday] | 0.764556 | 0.771759 |
| 3 | [workingday, hour, year, weather] | 0.802760 | 0.778615 |
| 4 | [workingday, hour, year, temp] | 0.927649 | 0.848878 |
| 0 | [workingday, hour, year, season] | 0.874880 | 0.861009 |
| 7 | [workingday, hour, year, month] | 0.906467 | 0.881726 |
In [ ]:
In [ ]:
In [53]:
# 이제 후진 선택법을 사용해 볼 것임
fe = ['season', 'holiday', 'workingday', 'weather', 'temp',
'humidity', 'windspeed', 'month', 'day', 'hour', "workingday", "hour", "year"]
# best_f = []
In [55]:
all_result = []
for c in combinations(fe, len(fe)-1):
# feature 에서 전체 컬럼의 수 -1 을 해서 하나씩 빼가면서
# 테스트를 해보겠다는 뜻
target = list(c)
# print(target)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
train_s = model.score(X_train, y_train)
test_s = model.score(X_test, y_test)
dropped = set(fe) - set(target)
result = {"dropped":dropped, "train":train_s, "test":test_s}
all_result.append(result)
In [57]:
result_df = pd.DataFrame(all_result).sort_values(by="test")
result_df
Out[57]:
| dropped | train | test | |
|---|---|---|---|
| 5 | {month} | 0.992111 | 0.930813 |
| 4 | {day} | 0.991652 | 0.932279 |
| 0 | {year} | 0.991575 | 0.936426 |
| 6 | {windspeed} | 0.992017 | 0.936628 |
| 1 | {} | 0.991685 | 0.937244 |
| 8 | {temp} | 0.991496 | 0.937614 |
| 11 | {holiday} | 0.991583 | 0.939131 |
| 10 | {} | 0.991901 | 0.939687 |
| 12 | {season} | 0.991995 | 0.940164 |
| 9 | {weather} | 0.991278 | 0.941420 |
| 3 | {} | 0.991178 | 0.942275 |
| 7 | {humidity} | 0.991742 | 0.943053 |
| 2 | {} | 0.991287 | 0.944364 |
In [ ]:
In [ ]:
In [58]:
fe = ['season', 'holiday', 'workingday', 'weather', 'temp',
'humidity', 'windspeed', 'month', 'day', 'hour', "workingday", "hour", "year"]
In [60]:
from sklearn.feature_selection import RFE
model = RandomForestRegressor()
rfe = RFE( estimator = model )
rfe.fit(X_train, y_train)
rfe_df = pd.DataFrame()
rfe_df["ranking"] = rfe.ranking_
rfe_df["features"] = X_train.columns
rfe_df
Out[60]:
| ranking | features | |
|---|---|---|
| 0 | 5 | season |
| 1 | 7 | holiday |
| 2 | 1 | workingday |
| 3 | 4 | weather |
| 4 | 1 | temp |
| 5 | 2 | atemp |
| 6 | 1 | humidity |
| 7 | 6 | windspeed |
| 8 | 1 | year |
| 9 | 1 | month |
| 10 | 3 | day |
| 11 | 1 | hour |
In [62]:
# ranking 이 높을수록 결과 값에 영향을 크게 미치는
# 중요한 컬럼들
rfe_df.sort_values(by="ranking")
Out[62]:
| ranking | features | |
|---|---|---|
| 2 | 1 | workingday |
| 4 | 1 | temp |
| 6 | 1 | humidity |
| 8 | 1 | year |
| 9 | 1 | month |
| 11 | 1 | hour |
| 5 | 2 | atemp |
| 10 | 3 | day |
| 3 | 4 | weather |
| 0 | 5 | season |
| 7 | 6 | windspeed |
| 1 | 7 | holiday |
In [68]:
feature = ['workingday', 'temp', 'humidity', 'year', 'month', 'hour', 'atemp']
In [69]:
model = RandomForestRegressor()
model.fit(X_train, y_train)
print("train : ", model.score(X_train, y_train))
print("test : ", model.score(X_test, y_test))
# 모든 컬럼을 사용했을때 평가점수
train : 0.9918053687355339 test : 0.9402065652970045
In [70]:
X_train.columns
Out[70]:
Index(['season', 'holiday', 'workingday', 'weather', 'temp', 'atemp',
'humidity', 'windspeed', 'year', 'month', 'day', 'hour'],
dtype='object')
In [72]:
X_tr, X_te, y_tr, y_te = train_test_split(df[feature], df[l], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_tr, y_tr)
print("train : ", model.score(X_tr, y_tr))
print("test : ", model.score(X_te, y_te))
# 결과에 영향을 크게 주는 컬럼들을 사용했을때 평가 점수
train : 0.9895003195998577 test : 0.9279685991163032
In [ ]:
In [ ]:
In [74]:
# embed 방식 변수 선택법을 확인해볼 것임
model = RandomForestRegressor()
model.fit(X_train, y_train)
print("train : ", model.score(X_train, y_train))
print("test : ", model.score(X_test, y_test))
print(model.feature_importances_)
print(X_train.columns)
train : 0.9917659298573632
test : 0.9393860178830495
[0.01127486 0.00352902 0.07385774 0.01356585 0.09019879 0.02357791
0.02972932 0.00938863 0.08657932 0.03626527 0.01343145 0.60860183]
Index(['season', 'holiday', 'workingday', 'weather', 'temp', 'atemp',
'humidity', 'windspeed', 'year', 'month', 'day', 'hour'],
dtype='object')
In [77]:
df_1 = pd.DataFrame()
# feature_importances 는 Tree 계열 알고리즘에서만 존재한다
# 이 방법이 embed 방법!!
df_1['rank'] = model.feature_importances_
df_1['feature'] = X_train.columns
df_1.sort_values(by="rank", ascending=False)
Out[77]:
| rank | feature | |
|---|---|---|
| 0 | 0.617593 | hour |
| 1 | 0.122718 | temp |
| 2 | 0.082577 | year |
| 3 | 0.071475 | workingday |
| 4 | 0.055804 | month |
| 5 | 0.049833 | humidity |
In [78]:
f = ['hour', 'temp', 'year', 'workingday', 'month', 'humidity']
X_train, X_test, y_train, y_test = train_test_split(df[f], df[l], test_size=0.3)
model = RandomForestRegressor()
model.fit(X_train, y_train)
print("train : ", model.score(X_train, y_train))
print("test : ", model.score(X_test, y_test))
train : 0.9889267902576386 test : 0.9297102519637512