Dacon 온도추정 경진대회
회귀분석 프로젝트
회귀분석 프로젝트 : 온도 추정 경진대회
목차
데이터 설명 분석과정 데이터 전처리 KFold 검증
Lasso, Ridge, EN 정규화
전체 데이터를 활용한 Y18모델 +) 머신러닝 활용
목표 성능
test MSE 3.5 이하로 만들기 -> 3.94
최종순위 20등 이내 -> 40위
데이터 설명
대전지역에서 측정한 실내외 19곳의 센서데이터와 주변 지역의 기상청 공공데이터를 semi-비식별화하여 제공합니다.
센서는 온도를 측정하였습니다.
모든 데이터는 시간 순으로 정렬 되어 있으며 10분 단위 데이터 입니다.
예측 대상(target variable)은 Y18입니다.
train.csv - 30일 간의 기상청 데이터 (X00~X39) 및 센서데이터 (Y00~Y17)
- 이후 3일 간의 기상청 데이터 (X00~X39) 및 센서데이터 (Y18)test.csv - train.csv 기간 이후 80일 간의 기상청 데이터 (X00~X39)sample_submission.csv - 제출 양식 예시
각 위치에서 기온, 누적강수량, 풍속, 풍향, 해면기압, 현지기압, 일사량, 습도 모두 측정
분석과정
id : 시간(10분 단위)X00 ~ X39 : 기상청 관측 데이터'X00','X07','X28','X31','X32' : 기온"X01","X06","X22","X27","X29" : 현지기압 (실제 위치에서의 기압)"X02","X03","X18","X24","X26" : 풍속"X04","X10","X21","X36","X39" : 일일 누적강수량"X05","X08","X09","X23","X33" : 해면기압 (높이를 같다고 가정했을 때 기압)"X11","X34" : 일일 누적일사량 - X14, X16, X19는 고장"X12","X20","X30","X37","X38" : 습도'Y00','Y01','Y02','Y03','Y04','Y05','Y06','Y07','Y08','Y09','Y10','Y11','Y12','Y13','Y14','Y15','Y16','Y17' : 센서(기온)
필요한 패키지 import
1 2 3 4 5 6 7 from sklearn.model_selection import KFoldfrom sklearn.ensemble import RandomForestRegressorfrom sklearn.model_selection import GridSearchCVfrom sklearn.model_selection import train_test_splitfrom sklearn.metrics import mean_squared_error, accuracy_score, make_scorerfrom sklearn.linear_model import Lasso, Ridge%matplotlib inline
데이터 불러오기
1 2 df= pd.read_csv('datas/train.csv' , encoding='utf-8' ) df
id
X00
X01
X02
X03
X04
X05
X06
X07
X08
...
Y09
Y10
Y11
Y12
Y13
Y14
Y15
Y16
Y17
Y18
0
0
9.7
988.8
1.2
0.6
0.0
1009.3
989.6
12.2
1009.9
...
7.0
7.5
7.0
9.0
10.0
9.5
9.0
8.0
9.0
NaN
1
1
9.3
988.9
1.7
1.9
0.0
1009.3
989.6
12.1
1010.0
...
6.5
7.5
7.0
8.5
10.0
9.5
9.0
7.5
9.0
NaN
2
2
9.4
989.0
1.1
2.3
0.0
1009.2
989.7
12.1
1010.1
...
6.5
7.5
6.5
8.0
9.5
9.5
8.5
7.5
8.5
NaN
3
3
9.4
988.9
1.5
0.7
0.0
1009.2
989.6
12.0
1010.0
...
6.0
7.0
6.0
8.0
9.5
9.0
8.5
7.5
8.5
NaN
4
4
9.2
988.9
0.8
1.7
0.0
1009.2
989.7
12.0
1010.1
...
6.0
7.0
6.0
7.5
9.5
9.0
8.5
7.5
8.5
NaN
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4747
4747
19.9
987.6
0.9
0.8
0.0
1006.9
987.7
21.7
1007.5
...
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
21.5
4748
4748
19.9
987.6
0.5
0.7
0.0
1006.8
987.7
21.6
1007.5
...
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
21.5
4749
4749
19.7
987.7
0.9
0.6
0.0
1006.9
987.6
21.4
1007.4
...
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
21.5
4750
4750
19.4
987.7
0.9
0.8
0.0
1006.9
987.8
21.3
1007.6
...
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
21.5
4751
4751
19.1
987.6
1.0
0.3
0.0
1006.8
987.8
21.2
1007.5
...
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
NaN
21.0
4752 rows × 60 columns
데이터 종류에 따라 나누기
1 2 3 4 5 6 7 8 9 temperature_name = ['X00' ,'X07' ,'X28' ,'X31' ,'X32' ] localpress_name = ["X01" ,"X06" ,"X22" ,"X27" ,"X29" ] speed_name = ["X02" ,"X03" ,"X18" ,"X24" ,"X26" ] water_name = ["X04" ,"X10" ,"X21" ,"X36" ,"X39" ] press_name = ["X05" ,"X08" ,"X09" ,"X23" ,"X33" ] sun_name = ["X11" ,"X34" ] humidity_name = ["X12" ,"X20" ,"X30" ,"X37" ,"X38" ] direction_name = ["X13" ,"X15" ,"X17" ,"X25" ,"X35" ] sensor_name = ['Y00' ,'Y01' ,'Y02' ,'Y03' ,'Y04' ,'Y05' ,'Y06' ,'Y07' ,'Y08' ,'Y09' ,'Y10' ,'Y11' ,'Y12' ,'Y13' ,'Y14' ,'Y15' ,'Y16' ,'Y17' ]
시간 컬럼 추가
1 2 hour= pd.Series((df.index%144 /6 ).astype(int)) df['hour' ] = hour
데이터 전처리
풍향
북동(0), 남동(1), 남서(2), 북서(3)로 카테고리 변수 로 변경
강수량
현재 데이터는 누적강수량 데이터로 측정시각 이전에 내린 강수량의 합이다.
현재 온도에 미래의 강수량이 영향을 주지 못하므로, 온도 측정 시각의 누적강수량에서 이전 온도 측정 시각의 누적강수량을 빼서 계산하였다.
실수형 데이터로 예측모델을 만든 결과보다 카테고리형 데이터로 예측모델을 만들었을 경우 성능이 더 좋아 카테고리 변수 로 바꾸었다.
강수량에 따라 0, 1, 2, 3으로 나눈 모델과 강수가 있었을 경우 1, 없었을 경우 0으로 나눈 모델 중 후자가 더 성능이 좋았다.
1 2 3 4 5 6 7 8 9 10 for x in direction_name: df[x] = df[x].apply(lambda x: 0 if x < 90 else 1 if x < 180 else 2 if x < 270 else 3 ) for x in water_name: for i in range(1 , len(x)): df[x].loc[i] = df[x].loc[i] - df[x].loc[i-1 ] for x in water_name: df[x] = df[x].apply(lambda x: 0 if x == 0 else 1 )
이전 30일의 데이터 - Y18이 없음
1 df1.dropna(axis=1 , inplace=True )
id
X00
X01
X02
X03
X04
X05
X06
X07
X08
...
Y09
Y10
Y11
Y12
Y13
Y14
Y15
Y16
Y17
hour
0
0
9.7
988.8
1.2
0.6
0
1009.3
989.6
12.2
1009.9
...
7.0
7.5
7.0
9.0
10.0
9.5
9.0
8.0
9.0
0
1
1
9.3
988.9
1.7
1.9
0
1009.3
989.6
12.1
1010.0
...
6.5
7.5
7.0
8.5
10.0
9.5
9.0
7.5
9.0
0
2
2
9.4
989.0
1.1
2.3
0
1009.2
989.7
12.1
1010.1
...
6.5
7.5
6.5
8.0
9.5
9.5
8.5
7.5
8.5
0
3
3
9.4
988.9
1.5
0.7
0
1009.2
989.6
12.0
1010.0
...
6.0
7.0
6.0
8.0
9.5
9.0
8.5
7.5
8.5
0
4
4
9.2
988.9
0.8
1.7
0
1009.2
989.7
12.0
1010.1
...
6.0
7.0
6.0
7.5
9.5
9.0
8.5
7.5
8.5
0
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4315
4315
19.5
987.8
1.7
0.4
0
1007.8
988.0
21.3
1007.8
...
19.5
20.0
19.0
20.5
20.5
20.0
20.0
20.5
20.0
23
4316
4316
19.3
987.7
2.1
0.9
0
1007.8
988.1
21.3
1008.0
...
19.5
20.0
19.0
20.5
20.5
20.0
19.5
20.5
19.5
23
4317
4317
19.5
987.8
0.9
1.3
0
1007.8
988.3
21.2
1008.1
...
19.5
20.0
19.0
20.0
20.0
19.5
19.5
20.0
19.0
23
4318
4318
20.0
987.8
1.4
0.5
0
1007.8
988.3
21.1
1008.1
...
19.5
20.0
18.5
20.0
20.0
19.5
19.5
20.0
19.0
23
4319
4319
20.1
987.8
1.2
1.2
0
1007.8
988.2
20.9
1008.1
...
19.5
20.0
18.5
19.5
19.5
19.5
19.5
19.5
19.0
23
4320 rows × 60 columns
이전 30일의 기상청 데이터(X00 ~ X39)
1 2 3 X_30s = df1.drop(columns=sensor_name) X_30s.drop(columns=['X14' , 'X16' , 'X19' , 'id' ], inplace=True ) X_30s
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X31
X32
X33
X34
X35
X36
X37
X38
X39
hour
0
9.7
988.8
1.2
0.6
0
1009.3
989.6
12.2
1009.9
1009.8
...
8.2
10.7
1010.1
0.00
2
0
77.2
62.6
0
0
1
9.3
988.9
1.7
1.9
0
1009.3
989.6
12.1
1010.0
1009.9
...
8.3
10.3
1010.1
0.00
2
0
77.3
63.5
0
0
2
9.4
989.0
1.1
2.3
0
1009.2
989.7
12.1
1010.1
1010.1
...
8.0
9.7
1010.0
0.00
2
0
77.3
63.9
0
0
3
9.4
988.9
1.5
0.7
0
1009.2
989.6
12.0
1010.0
1010.0
...
7.7
9.4
1010.1
0.00
2
0
77.5
64.5
0
0
4
9.2
988.9
0.8
1.7
0
1009.2
989.7
12.0
1010.1
1010.0
...
7.4
9.4
1010.1
0.00
1
0
78.0
65.0
0
0
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4315
19.5
987.8
1.7
0.4
0
1007.8
988.0
21.3
1007.8
1007.5
...
18.8
19.4
1008.2
16.68
3
0
82.3
74.3
0
23
4316
19.3
987.7
2.1
0.9
0
1007.8
988.1
21.3
1008.0
1007.4
...
18.6
18.5
1008.4
16.68
3
0
82.4
74.8
0
23
4317
19.5
987.8
0.9
1.3
0
1007.8
988.3
21.2
1008.1
1007.5
...
18.4
19.1
1008.4
16.68
3
0
82.8
75.4
0
23
4318
20.0
987.8
1.4
0.5
0
1007.8
988.3
21.1
1008.1
1007.5
...
18.2
19.0
1008.4
16.68
3
0
82.8
75.8
0
23
4319
20.1
987.8
1.2
1.2
0
1007.8
988.2
20.9
1008.1
1007.5
...
18.1
18.5
1008.3
16.68
3
0
83.5
76.9
0
23
4320 rows × 38 columns
이전 30일의 센서 데이터(Y00 ~ Y17)
1 2 sensor = df1[sensor_name] sensor
Y00
Y01
Y02
Y03
Y04
Y05
Y06
Y07
Y08
Y09
Y10
Y11
Y12
Y13
Y14
Y15
Y16
Y17
0
12.5
11.5
11.0
20.0
20.0
10.5
10.0
10.5
10.0
7.0
7.5
7.0
9.0
10.0
9.5
9.0
8.0
9.0
1
12.5
11.0
11.0
20.0
20.0
10.0
9.5
10.0
9.5
6.5
7.5
7.0
8.5
10.0
9.5
9.0
7.5
9.0
2
12.0
11.0
11.0
19.5
19.5
10.0
9.0
10.0
9.0
6.5
7.5
6.5
8.0
9.5
9.5
8.5
7.5
8.5
3
12.0
11.0
11.0
19.5
19.5
9.5
9.0
9.5
9.0
6.0
7.0
6.0
8.0
9.5
9.0
8.5
7.5
8.5
4
12.0
10.5
10.5
19.0
19.5
9.5
8.5
9.5
8.5
6.0
7.0
6.0
7.5
9.5
9.0
8.5
7.5
8.5
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4315
22.0
22.5
22.5
26.0
26.0
21.0
20.0
20.5
20.5
19.5
20.0
19.0
20.5
20.5
20.0
20.0
20.5
20.0
4316
22.0
22.5
22.5
26.0
25.5
21.0
20.0
20.5
20.5
19.5
20.0
19.0
20.5
20.5
20.0
19.5
20.5
19.5
4317
21.5
22.5
22.5
26.0
25.5
21.0
20.0
20.5
20.5
19.5
20.0
19.0
20.0
20.0
19.5
19.5
20.0
19.0
4318
21.5
22.0
22.0
25.5
25.5
20.5
20.0
20.0
20.0
19.5
20.0
18.5
20.0
20.0
19.5
19.5
20.0
19.0
4319
21.5
22.0
22.0
25.5
25.5
20.5
19.5
20.0
20.0
19.5
20.0
18.5
19.5
19.5
19.5
19.5
19.5
19.0
4320 rows × 18 columns
1 2 3 formulas = temperature_name + localpress_name + humidity_name + sun_name + speed_name formulas = ['scale({})' .format(x) for x in formulas] formulas
['scale(X00)',
'scale(X07)',
'scale(X28)',
'scale(X31)',
'scale(X32)',
'scale(X01)',
'scale(X06)',
'scale(X22)',
'scale(X27)',
'scale(X29)',
'scale(X12)',
'scale(X20)',
'scale(X30)',
'scale(X37)',
'scale(X38)',
'scale(X11)',
'scale(X34)',
'scale(X02)',
'scale(X03)',
'scale(X18)',
'scale(X24)',
'scale(X26)']1 2 3 4 formulas2 = water_name + direction_name formulas2.append('hour' ) formulas2 = ['C({})' .format(x) for x in formulas2] formulas2
['C(X04)',
'C(X10)',
'C(X21)',
'C(X36)',
'C(X39)',
'C(X13)',
'C(X15)',
'C(X17)',
'C(X25)',
'C(X35)',
'C(hour)']1 formulas = formulas + formulas2
OLS 식 , Y00 예측 모델만들기
스케일링, 카테고리 변수
Y00을 포함한 Y00_df 생성
1 2 a = '+' .join(formulas) a
'scale(X00)+scale(X07)+scale(X28)+scale(X31)+scale(X32)+scale(X01)+scale(X06)+scale(X22)+scale(X27)+scale(X29)+scale(X12)+scale(X20)+scale(X30)+scale(X37)+scale(X38)+scale(X11)+scale(X34)+scale(X02)+scale(X03)+scale(X18)+scale(X24)+scale(X26)+C(X04)+C(X10)+C(X21)+C(X36)+C(X39)+C(X13)+C(X15)+C(X17)+C(X25)+C(X35)+C(hour)'1 2 Y00_df = pd.concat([X_30s, sensor['Y00' ]], axis=1 ) Y00_df
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X32
X33
X34
X35
X36
X37
X38
X39
hour
Y00
0
9.7
988.8
1.2
0.6
0
1009.3
989.6
12.2
1009.9
1009.8
...
10.7
1010.1
0.00
2
0
77.2
62.6
0
0
12.5
1
9.3
988.9
1.7
1.9
0
1009.3
989.6
12.1
1010.0
1009.9
...
10.3
1010.1
0.00
2
0
77.3
63.5
0
0
12.5
2
9.4
989.0
1.1
2.3
0
1009.2
989.7
12.1
1010.1
1010.1
...
9.7
1010.0
0.00
2
0
77.3
63.9
0
0
12.0
3
9.4
988.9
1.5
0.7
0
1009.2
989.6
12.0
1010.0
1010.0
...
9.4
1010.1
0.00
2
0
77.5
64.5
0
0
12.0
4
9.2
988.9
0.8
1.7
0
1009.2
989.7
12.0
1010.1
1010.0
...
9.4
1010.1
0.00
1
0
78.0
65.0
0
0
12.0
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4315
19.5
987.8
1.7
0.4
0
1007.8
988.0
21.3
1007.8
1007.5
...
19.4
1008.2
16.68
3
0
82.3
74.3
0
23
22.0
4316
19.3
987.7
2.1
0.9
0
1007.8
988.1
21.3
1008.0
1007.4
...
18.5
1008.4
16.68
3
0
82.4
74.8
0
23
22.0
4317
19.5
987.8
0.9
1.3
0
1007.8
988.3
21.2
1008.1
1007.5
...
19.1
1008.4
16.68
3
0
82.8
75.4
0
23
21.5
4318
20.0
987.8
1.4
0.5
0
1007.8
988.3
21.1
1008.1
1007.5
...
19.0
1008.4
16.68
3
0
82.8
75.8
0
23
21.5
4319
20.1
987.8
1.2
1.2
0
1007.8
988.2
20.9
1008.1
1007.5
...
18.5
1008.3
16.68
3
0
83.5
76.9
0
23
21.5
4320 rows × 39 columns
Y00 OLS Report
1 2 3 model1 = sm.OLS.from_formula("Y00 ~ scale(X00)+scale(X07)+scale(X28)+scale(X31)+scale(X32)+scale(X01)+scale(X06)+scale(X22)+scale(X27)+scale(X29)+scale(X12)+scale(X20)+scale(X30)+scale(X37)+scale(X38)+scale(X11)+scale(X34)+scale(X02)+scale(X03)+scale(X18)+scale(X24)+scale(X26)+C(X04)+C(X10)+C(X21)+C(X36)+C(X39)+C(X13)+C(X15)+C(X17)+C(X25)+C(X35)+C(hour)" , Y00_df) result1 = model1.fit() print(result1.summary())
OLS Regression Results
==============================================================================
Dep. Variable: Y00 R-squared: 0.993
Model: OLS Adj. R-squared: 0.993
Method: Least Squares F-statistic: 8823.
Date: Sat, 18 Apr 2020 Prob (F-statistic): 0.00
Time: 14:39:02 Log-Likelihood: -1831.6
No. Observations: 4320 AIC: 3795.
Df Residuals: 4254 BIC: 4216.
Df Model: 65
Covariance Type: nonrobust
=================================================================================
coef std err t P>|t| [0.025 0.975]
---------------------------------------------------------------------------------
Intercept 21.9936 0.043 509.663 0.000 21.909 22.078
C(X04)[T.1] -0.1483 0.054 -2.724 0.006 -0.255 -0.042
C(X10)[T.1] 0.2583 0.071 3.646 0.000 0.119 0.397
C(X21)[T.1] -0.1453 0.069 -2.111 0.035 -0.280 -0.010
C(X36)[T.1] -0.0792 0.055 -1.429 0.153 -0.188 0.029
C(X39)[T.1] -0.0264 0.040 -0.656 0.512 -0.105 0.052
C(X13)[T.1] -0.0262 0.019 -1.399 0.162 -0.063 0.011
C(X13)[T.2] -0.0040 0.016 -0.258 0.796 -0.035 0.027
C(X13)[T.3] 0.0225 0.021 1.090 0.276 -0.018 0.063
C(X15)[T.1] -0.0090 0.020 -0.443 0.658 -0.049 0.031
C(X15)[T.2] 0.0408 0.019 2.138 0.033 0.003 0.078
C(X15)[T.3] 0.0247 0.021 1.172 0.241 -0.017 0.066
C(X17)[T.1] -0.0217 0.020 -1.112 0.266 -0.060 0.017
C(X17)[T.2] -0.0114 0.019 -0.603 0.547 -0.049 0.026
C(X17)[T.3] -0.0262 0.020 -1.317 0.188 -0.065 0.013
C(X25)[T.1] -0.0218 0.024 -0.897 0.370 -0.069 0.026
C(X25)[T.2] 0.0802 0.020 3.914 0.000 0.040 0.120
C(X25)[T.3] 0.0382 0.018 2.081 0.037 0.002 0.074
C(X35)[T.1] 0.0633 0.023 2.742 0.006 0.018 0.108
C(X35)[T.2] 0.0510 0.020 2.535 0.011 0.012 0.091
C(X35)[T.3] 0.0091 0.021 0.444 0.657 -0.031 0.049
C(hour)[T.1] -0.1131 0.040 -2.851 0.004 -0.191 -0.035
C(hour)[T.2] -0.2216 0.040 -5.543 0.000 -0.300 -0.143
C(hour)[T.3] -0.3348 0.040 -8.272 0.000 -0.414 -0.255
C(hour)[T.4] -0.4096 0.041 -9.969 0.000 -0.490 -0.329
C(hour)[T.5] -0.5497 0.042 -13.096 0.000 -0.632 -0.467
C(hour)[T.6] -0.9854 0.042 -23.581 0.000 -1.067 -0.903
C(hour)[T.7] -1.3939 0.042 -33.197 0.000 -1.476 -1.312
C(hour)[T.8] -1.5723 0.043 -36.336 0.000 -1.657 -1.487
C(hour)[T.9] -1.6007 0.046 -35.118 0.000 -1.690 -1.511
C(hour)[T.10] -1.5496 0.048 -32.255 0.000 -1.644 -1.455
C(hour)[T.11] -1.4606 0.050 -28.934 0.000 -1.560 -1.362
C(hour)[T.12] -1.3289 0.053 -24.914 0.000 -1.433 -1.224
C(hour)[T.13] -1.3072 0.055 -23.769 0.000 -1.415 -1.199
C(hour)[T.14] -1.1328 0.057 -20.037 0.000 -1.244 -1.022
C(hour)[T.15] -0.9958 0.059 -16.972 0.000 -1.111 -0.881
C(hour)[T.16] -0.8724 0.060 -14.451 0.000 -0.991 -0.754
C(hour)[T.17] -0.6048 0.061 -9.920 0.000 -0.724 -0.485
C(hour)[T.18] -0.2100 0.062 -3.395 0.001 -0.331 -0.089
C(hour)[T.19] 0.1856 0.062 2.993 0.003 0.064 0.307
C(hour)[T.20] 0.1702 0.061 2.781 0.005 0.050 0.290
C(hour)[T.21] -0.0616 0.062 -1.000 0.317 -0.182 0.059
C(hour)[T.22] -0.2288 0.063 -3.652 0.000 -0.352 -0.106
C(hour)[T.23] -0.4915 0.064 -7.712 0.000 -0.616 -0.367
scale(X00) -0.3816 0.042 -9.083 0.000 -0.464 -0.299
scale(X07) 3.9408 0.056 69.833 0.000 3.830 4.051
scale(X28) 0.6474 0.053 12.245 0.000 0.544 0.751
scale(X31) -0.0514 0.051 -0.999 0.318 -0.152 0.049
scale(X32) 0.1671 0.049 3.393 0.001 0.071 0.264
scale(X01) 0.1292 0.132 0.977 0.329 -0.130 0.388
scale(X06) -0.1829 0.138 -1.328 0.184 -0.453 0.087
scale(X22) 0.6216 0.142 4.392 0.000 0.344 0.899
scale(X27) -0.6094 0.102 -5.997 0.000 -0.809 -0.410
scale(X29) 0.0311 0.188 0.165 0.869 -0.338 0.401
scale(X12) -0.0350 0.036 -0.970 0.332 -0.106 0.036
scale(X20) 0.0359 0.028 1.278 0.201 -0.019 0.091
scale(X30) 0.1846 0.032 5.839 0.000 0.123 0.247
scale(X37) -0.0198 0.029 -0.672 0.502 -0.078 0.038
scale(X38) 0.1327 0.027 4.901 0.000 0.080 0.186
scale(X11) -0.4297 0.068 -6.304 0.000 -0.563 -0.296
scale(X34) 0.7449 0.070 10.587 0.000 0.607 0.883
scale(X02) 0.0242 0.008 3.171 0.002 0.009 0.039
scale(X03) 0.0249 0.008 3.133 0.002 0.009 0.041
scale(X18) 0.0107 0.009 1.140 0.254 -0.008 0.029
scale(X24) -0.0069 0.010 -0.687 0.492 -0.027 0.013
scale(X26) 0.0623 0.010 6.273 0.000 0.043 0.082
==============================================================================
Omnibus: 91.277 Durbin-Watson: 0.552
Prob(Omnibus): 0.000 Jarque-Bera (JB): 128.438
Skew: 0.243 Prob(JB): 1.29e-28
Kurtosis: 3.691 Cond. No. 132.
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.1 2 3 4 5 6 fig = plt.figure(figsize=(8 , 100 )) sm.graphics.plot_ccpr_grid(result1, fig=fig) fig.suptitle("" ) plt.tight_layout() plt.show()
KFold 검증
1 2 from sklearn.metrics import mean_squared_errorkfold = KFold(5 , shuffle=True , random_state=13 )
1 2 3 4 5 6 7 8 N = len(Y00_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y00_df.iloc[idx_train] dfX_test = Y00_df.iloc[idx_test]
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model0 = sm.OLS.from_formula("Y00 ~ 1 +" + a, data=dfX_train) result0 = model0.fit() pred = result0.predict(df_test) rss = ((df_test.Y00 - pred) ** 2 ).sum() tss = ((df_test.Y00 - df_test.Y00.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result0.rsquared, rsquared)) pred = result0.predict(dfX_test) rss = ((dfX_test.Y00 - pred) ** 2 ).sum() tss = ((dfX_test.Y00 - dfX_test.Y00.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.99272483, 검증 R2 = 0.99242741
학습 R2 = 0.99272483, 검증 R2 = 0.99215678
학습 R2 = 0.99272483, 검증 R2 = 0.99328929
학습 R2 = 0.99272483, 검증 R2 = 0.99281616
학습 R2 = 0.99272483, 검증 R2 = 0.99284578
0.99237369126495971 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y00' ], pred) plt.plot([0 , dfX_test['Y00' ].max()], [0 , dfX_test['Y00' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
나머지 값들도 KFold 검증
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y01_df = pd.concat([X_30s, sensor['Y01' ]], axis=1 ) N = len(Y01_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y01_df.iloc[idx_train] dfX_test = Y01_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model1 = sm.OLS.from_formula("Y01 ~ 1 +" + a, data=dfX_train) result1 = model1.fit() pred = result1.predict(df_test) rss = ((df_test.Y01 - pred) ** 2 ).sum() tss = ((df_test.Y01 - df_test.Y01.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result1.rsquared, rsquared)) pred = result1.predict(dfX_test) rss = ((dfX_test.Y01 - pred) ** 2 ).sum() tss = ((dfX_test.Y01 - dfX_test.Y01.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.98413713, 검증 R2 = 0.98378369
학습 R2 = 0.98413713, 검증 R2 = 0.98328775
학습 R2 = 0.98413713, 검증 R2 = 0.98360291
학습 R2 = 0.98413713, 검증 R2 = 0.98524484
학습 R2 = 0.98413713, 검증 R2 = 0.98458787
0.98127394273747541 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y01' ], pred) plt.plot([0 , dfX_test['Y01' ].max()], [0 , dfX_test['Y01' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y02_df = pd.concat([X_30s, sensor['Y02' ]], axis=1 ) N = len(Y02_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y02_df.iloc[idx_train] dfX_test = Y02_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model2 = sm.OLS.from_formula("Y02 ~ 1 +" + a, data=dfX_train) result2 = model2.fit() pred = result2.predict(df_test) rss = ((df_test.Y02 - pred) ** 2 ).sum() tss = ((df_test.Y02 - df_test.Y02.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result2.rsquared, rsquared)) pred = result2.predict(dfX_test) rss = ((dfX_test.Y02 - pred) ** 2 ).sum() tss = ((dfX_test.Y02 - dfX_test.Y02.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.98337319, 검증 R2 = 0.98398663
학습 R2 = 0.98337319, 검증 R2 = 0.98194591
학습 R2 = 0.98337319, 검증 R2 = 0.98332113
학습 R2 = 0.98337319, 검증 R2 = 0.98376697
학습 R2 = 0.98337319, 검증 R2 = 0.98364382
0.98137443644142671 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y02' ], pred) plt.plot([0 , dfX_test['Y02' ].max()], [0 , dfX_test['Y02' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y03_df = pd.concat([X_30s, sensor['Y03' ]], axis=1 ) N = len(Y03_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y03_df.iloc[idx_train] dfX_test = Y03_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model3 = sm.OLS.from_formula("Y03 ~ 1 +" + a, data=dfX_train) result3 = model3.fit() pred = result3.predict(df_test) rss = ((df_test.Y03 - pred) ** 2 ).sum() tss = ((df_test.Y03 - df_test.Y03.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result3.rsquared, rsquared)) pred = result3.predict(dfX_test) rss = ((dfX_test.Y03 - pred) ** 2 ).sum() tss = ((dfX_test.Y03 - dfX_test.Y03.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.94846634, 검증 R2 = 0.94657794
학습 R2 = 0.94846634, 검증 R2 = 0.94653673
학습 R2 = 0.94846634, 검증 R2 = 0.95014468
학습 R2 = 0.94846634, 검증 R2 = 0.95104291
학습 R2 = 0.94846634, 검증 R2 = 0.94797948
0.94733443657143681 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y03' ], pred) plt.plot([0 , dfX_test['Y03' ].max()], [0 , dfX_test['Y03' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 Y04_df = pd.concat([X_30s, sensor['Y04' ]], axis=1 ) N = len(Y04_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y04_df.iloc[idx_train] dfX_test = Y04_df.iloc[idx_test] scores = np.zeros(5 ) cv = KFold(5 , shuffle=True , random_state=13 ) for i, (idx_train, idx_test) in enumerate(cv.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model4 = sm.OLS.from_formula("Y04 ~ 1 +" + a, data=dfX_train) result4 = model4.fit() pred = result4.predict(df_test) rss = ((df_test.Y04 - pred) ** 2 ).sum() tss = ((df_test.Y04 - df_test.Y04.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result4.rsquared, rsquared)) pred = result4.predict(dfX_test) rss = ((dfX_test.Y04 - pred) ** 2 ).sum() tss = ((dfX_test.Y04 - dfX_test.Y04.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.94796568, 검증 R2 = 0.94734167
학습 R2 = 0.94796568, 검증 R2 = 0.94561380
학습 R2 = 0.94796568, 검증 R2 = 0.94708696
학습 R2 = 0.94796568, 검증 R2 = 0.95117340
학습 R2 = 0.94796568, 검증 R2 = 0.94848745
0.94661639675132481 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y04' ], pred) plt.plot([0 , dfX_test['Y04' ].max()], [0 , dfX_test['Y04' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y05_df = pd.concat([X_30s, sensor['Y05' ]], axis=1 ) N = len(Y05_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y05_df.iloc[idx_train] dfX_test = Y05_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model5 = sm.OLS.from_formula("Y05 ~ 1 +" + a, data=dfX_train) result5 = model5.fit() pred = result5.predict(df_test) rss = ((df_test.Y05 - pred) ** 2 ).sum() tss = ((df_test.Y05 - df_test.Y05.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result5.rsquared, rsquared)) pred = result5.predict(dfX_test) rss = ((dfX_test.Y05 - pred) ** 2 ).sum() tss = ((dfX_test.Y05 - dfX_test.Y05.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.94777908, 검증 R2 = 0.94299018
학습 R2 = 0.94777908, 검증 R2 = 0.94408928
학습 R2 = 0.94777908, 검증 R2 = 0.94572819
학습 R2 = 0.94777908, 검증 R2 = 0.94986888
학습 R2 = 0.94777908, 검증 R2 = 0.95540827
0.94060424247268121 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y05' ], pred) plt.plot([0 , dfX_test['Y05' ].max()], [0 , dfX_test['Y05' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y06_df = pd.concat([X_30s, sensor['Y06' ]], axis=1 ) N = len(Y06_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y06_df.iloc[idx_train] dfX_test = Y06_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model6 = sm.OLS.from_formula("Y06 ~ 1 +" + a, data=dfX_train) result6 = model6.fit() pred = result6.predict(df_test) rss = ((df_test.Y06 - pred) ** 2 ).sum() tss = ((df_test.Y06 - df_test.Y06.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result6.rsquared, rsquared)) pred = result6.predict(dfX_test) rss = ((dfX_test.Y06 - pred) ** 2 ).sum() tss = ((dfX_test.Y06 - dfX_test.Y06.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.93753514, 검증 R2 = 0.93838017
학습 R2 = 0.93753514, 검증 R2 = 0.93545692
학습 R2 = 0.93753514, 검증 R2 = 0.93884759
학습 R2 = 0.93753514, 검증 R2 = 0.93079795
학습 R2 = 0.93753514, 검증 R2 = 0.94346828
0.93303619367834021 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y06' ], pred) plt.plot([0 , dfX_test['Y06' ].max()], [0 , dfX_test['Y06' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y07_df = pd.concat([X_30s, sensor['Y07' ]], axis=1 ) N = len(Y07_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y07_df.iloc[idx_train] dfX_test = Y07_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model7 = sm.OLS.from_formula("Y07 ~ 1 +" + a, data=dfX_train) result7 = model7.fit() pred = result7.predict(df_test) rss = ((df_test.Y07 - pred) ** 2 ).sum() tss = ((df_test.Y07 - df_test.Y07.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result7.rsquared, rsquared)) pred = result7.predict(dfX_test) rss = ((dfX_test.Y07 - pred) ** 2 ).sum() tss = ((dfX_test.Y07 - dfX_test.Y07.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.89336104, 검증 R2 = 0.88463988
학습 R2 = 0.89336104, 검증 R2 = 0.89507160
학습 R2 = 0.89336104, 검증 R2 = 0.90083034
학습 R2 = 0.89336104, 검증 R2 = 0.87764882
학습 R2 = 0.89336104, 검증 R2 = 0.90739659
0.88674404417940641 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y07' ], pred) plt.plot([0 , dfX_test['Y07' ].max()], [0 , dfX_test['Y07' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y08_df = pd.concat([X_30s, sensor['Y08' ]], axis=1 ) N = len(Y08_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y08_df.iloc[idx_train] dfX_test = Y08_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model8 = sm.OLS.from_formula("Y08 ~ 1 +" + a, data=dfX_train) result8 = model8.fit() pred = result8.predict(df_test) rss = ((df_test.Y08 - pred) ** 2 ).sum() tss = ((df_test.Y08 - df_test.Y08.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result8.rsquared, rsquared)) pred = result8.predict(dfX_test) rss = ((dfX_test.Y08 - pred) ** 2 ).sum() tss = ((dfX_test.Y08 - dfX_test.Y08.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.93860214, 검증 R2 = 0.93855661
학습 R2 = 0.93860214, 검증 R2 = 0.93569780
학습 R2 = 0.93860214, 검증 R2 = 0.93912749
학습 R2 = 0.93860214, 검증 R2 = 0.93465834
학습 R2 = 0.93860214, 검증 R2 = 0.94418988
0.92779419794118151 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y08' ], pred) plt.plot([0 , dfX_test['Y08' ].max()], [0 , dfX_test['Y08' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y09_df = pd.concat([X_30s, sensor['Y09' ]], axis=1 ) N = len(Y09_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y09_df.iloc[idx_train] dfX_test = Y09_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model9 = sm.OLS.from_formula("Y09 ~ 1 +" + a, data=dfX_train) result9 = model9.fit() pred = result9.predict(df_test) rss = ((df_test.Y09 - pred) ** 2 ).sum() tss = ((df_test.Y09 - df_test.Y09.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result9.rsquared, rsquared)) pred = result9.predict(dfX_test) rss = ((dfX_test.Y09 - pred) ** 2 ).sum() tss = ((dfX_test.Y09 - dfX_test.Y09.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.97479786, 검증 R2 = 0.97486045
학습 R2 = 0.97479786, 검증 R2 = 0.97286052
학습 R2 = 0.97479786, 검증 R2 = 0.97321239
학습 R2 = 0.97479786, 검증 R2 = 0.97642863
학습 R2 = 0.97479786, 검증 R2 = 0.97640687
0.97116467890424011 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y09' ], pred) plt.plot([0 , dfX_test['Y09' ].max()], [0 , dfX_test['Y09' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y10_df = pd.concat([X_30s, sensor['Y10' ]], axis=1 ) N = len(Y10_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y10_df.iloc[idx_train] dfX_test = Y10_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model10 = sm.OLS.from_formula("Y10 ~ 1 +" + a, data=dfX_train) result10 = model10.fit() pred = result10.predict(df_test) rss = ((df_test.Y10 - pred) ** 2 ).sum() tss = ((df_test.Y10 - df_test.Y10.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result10.rsquared, rsquared)) pred = result10.predict(dfX_test) rss = ((dfX_test.Y10 - pred) ** 2 ).sum() tss = ((dfX_test.Y10 - dfX_test.Y10.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.97589977, 검증 R2 = 0.97627704
학습 R2 = 0.97589977, 검증 R2 = 0.97351250
학습 R2 = 0.97589977, 검증 R2 = 0.97526803
학습 R2 = 0.97589977, 검증 R2 = 0.97882271
학습 R2 = 0.97589977, 검증 R2 = 0.97525676
0.97104769247239091 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y10' ], pred) plt.plot([0 , dfX_test['Y10' ].max()], [0 , dfX_test['Y10' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y11_df = pd.concat([X_30s, sensor['Y11' ]], axis=1 ) N = len(Y11_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y11_df.iloc[idx_train] dfX_test = Y11_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model11 = sm.OLS.from_formula("Y11 ~ 1 +" + a, data=dfX_train) result11 = model11.fit() pred = result11.predict(df_test) rss = ((df_test.Y11 - pred) ** 2 ).sum() tss = ((df_test.Y11 - df_test.Y11.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result11.rsquared, rsquared)) pred = result11.predict(dfX_test) rss = ((dfX_test.Y11 - pred) ** 2 ).sum() tss = ((dfX_test.Y11 - dfX_test.Y11.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.97119857, 검증 R2 = 0.97285008
학습 R2 = 0.97119857, 검증 R2 = 0.96713121
학습 R2 = 0.97119857, 검증 R2 = 0.96954841
학습 R2 = 0.97119857, 검증 R2 = 0.97423509
학습 R2 = 0.97119857, 검증 R2 = 0.97184550
0.96513652251464711 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y11' ], pred) plt.plot([0 , dfX_test['Y11' ].max()], [0 , dfX_test['Y11' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y12_df = pd.concat([X_30s, sensor['Y12' ]], axis=1 ) N = len(Y12_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y12_df.iloc[idx_train] dfX_test = Y12_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model12 = sm.OLS.from_formula("Y12 ~ 1 +" + a, data=dfX_train) result12 = model12.fit() pred = result12.predict(df_test) rss = ((df_test.Y12 - pred) ** 2 ).sum() tss = ((df_test.Y12 - df_test.Y12.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result12.rsquared, rsquared)) pred = result12.predict(dfX_test) rss = ((dfX_test.Y12 - pred) ** 2 ).sum() tss = ((dfX_test.Y12 - dfX_test.Y12.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.93043914, 검증 R2 = 0.92730076
학습 R2 = 0.93043914, 검증 R2 = 0.92779232
학습 R2 = 0.93043914, 검증 R2 = 0.93719237
학습 R2 = 0.93043914, 검증 R2 = 0.92444176
학습 R2 = 0.93043914, 검증 R2 = 0.93435359
0.92585305411287161 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y12' ], pred) plt.plot([0 , dfX_test['Y12' ].max()], [0 , dfX_test['Y12' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y13_df = pd.concat([X_30s, sensor['Y13' ]], axis=1 ) N = len(Y13_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y13_df.iloc[idx_train] dfX_test = Y13_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model13 = sm.OLS.from_formula("Y13 ~ 1 +" + a, data=dfX_train) result13 = model13.fit() pred = result13.predict(df_test) rss = ((df_test.Y13 - pred) ** 2 ).sum() tss = ((df_test.Y13 - df_test.Y13.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result13.rsquared, rsquared)) pred = result13.predict(dfX_test) rss = ((dfX_test.Y13 - pred) ** 2 ).sum() tss = ((dfX_test.Y13 - dfX_test.Y13.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.97551170, 검증 R2 = 0.97691878
학습 R2 = 0.97551170, 검증 R2 = 0.97502043
학습 R2 = 0.97551170, 검증 R2 = 0.97650240
학습 R2 = 0.97551170, 검증 R2 = 0.97279838
학습 R2 = 0.97551170, 검증 R2 = 0.97615578
0.97230875704766321 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y13' ], pred) plt.plot([0 , dfX_test['Y13' ].max()], [0 , dfX_test['Y13' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y14_df = pd.concat([X_30s, sensor['Y14' ]], axis=1 ) N = len(Y14_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y14_df.iloc[idx_train] dfX_test = Y14_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model14 = sm.OLS.from_formula("Y14 ~ 1 +" + a, data=dfX_train) result14 = model14.fit() pred = result14.predict(df_test) rss = ((df_test.Y14 - pred) ** 2 ).sum() tss = ((df_test.Y14 - df_test.Y14.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result14.rsquared, rsquared)) pred = result14.predict(dfX_test) rss = ((dfX_test.Y14 - pred) ** 2 ).sum() tss = ((dfX_test.Y14 - dfX_test.Y14.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.95055229, 검증 R2 = 0.95174316
학습 R2 = 0.95055229, 검증 R2 = 0.95003596
학습 R2 = 0.95055229, 검증 R2 = 0.95233452
학습 R2 = 0.95055229, 검증 R2 = 0.94609319
학습 R2 = 0.95055229, 검증 R2 = 0.95207706
0.94339029247856451 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y14' ], pred) plt.plot([0 , dfX_test['Y14' ].max()], [0 , dfX_test['Y14' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y15_df = pd.concat([X_30s, sensor['Y15' ]], axis=1 ) N = len(Y15_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y15_df.iloc[idx_train] dfX_test = Y15_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model15 = sm.OLS.from_formula("Y15 ~ 1 +" + a, data=dfX_train) result15 = model15.fit() pred = result15.predict(df_test) rss = ((df_test.Y15 - pred) ** 2 ).sum() tss = ((df_test.Y15 - df_test.Y15.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result15.rsquared, rsquared)) pred = result15.predict(dfX_test) rss = ((dfX_test.Y15 - pred) ** 2 ).sum() tss = ((dfX_test.Y15 - dfX_test.Y15.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.97086836, 검증 R2 = 0.97222260
학습 R2 = 0.97086836, 검증 R2 = 0.97190213
학습 R2 = 0.97086836, 검증 R2 = 0.96797335
학습 R2 = 0.97086836, 검증 R2 = 0.97158117
학습 R2 = 0.97086836, 검증 R2 = 0.97070777
0.96736908948071921 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y15' ], pred) plt.plot([0 , dfX_test['Y15' ].max()], [0 , dfX_test['Y15' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y16_df = pd.concat([X_30s, sensor['Y16' ]], axis=1 ) N = len(Y16_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y16_df.iloc[idx_train] dfX_test = Y16_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model16 = sm.OLS.from_formula("Y16 ~ 1 +" + a, data=dfX_train) result16 = model16.fit() b pred = result16.predict(df_test) rss = ((df_test.Y16 - pred) ** 2 ).sum() tss = ((df_test.Y16 - df_test.Y16.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result16.rsquared, rsquared)) pred = result16.predict(dfX_test) rss = ((dfX_test.Y16 - pred) ** 2 ).sum() tss = ((dfX_test.Y16 - dfX_test.Y16.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.96814730, 검증 R2 = 0.96684993
학습 R2 = 0.96814730, 검증 R2 = 0.96568806
학습 R2 = 0.96814730, 검증 R2 = 0.96983776
학습 R2 = 0.96814730, 검증 R2 = 0.96852404
학습 R2 = 0.96814730, 검증 R2 = 0.96930078
0.96352848163535131 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y16' ], pred) plt.plot([0 , dfX_test['Y16' ].max()], [0 , dfX_test['Y16' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 Y17_df = pd.concat([X_30s, sensor['Y17' ]], axis=1 ) N = len(Y17_df) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y17_df.iloc[idx_train] dfX_test = Y17_df.iloc[idx_test] scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model17 = sm.OLS.from_formula("Y17 ~ 1 +" + a, data=dfX_train) result17 = model17.fit() pred = result17.predict(df_test) rss = ((df_test.Y17 - pred) ** 2 ).sum() tss = ((df_test.Y17 - df_test.Y17.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result17.rsquared, rsquared)) pred = result17.predict(dfX_test) rss = ((dfX_test.Y17 - pred) ** 2 ).sum() tss = ((dfX_test.Y17 - dfX_test.Y17.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared)
학습 R2 = 0.95399940, 검증 R2 = 0.95549371
학습 R2 = 0.95399940, 검증 R2 = 0.95233226
학습 R2 = 0.95399940, 검증 R2 = 0.95487363
학습 R2 = 0.95399940, 검증 R2 = 0.95520874
학습 R2 = 0.95399940, 검증 R2 = 0.95170376
0.94950623436259321 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y17' ], pred) plt.plot([0 , dfX_test['Y17' ].max()], [0 , dfX_test['Y17' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
모델을 활용한 이후 3일의 Y00~Y17 값 예측
1 2 3 df2 = df.loc[4320 :] df2.dropna(axis=1 , inplace=True ) df2
id
X00
X01
X02
X03
X04
X05
X06
X07
X08
...
X32
X33
X34
X35
X36
X37
X38
X39
Y18
hour
4320
4320
19.3
987.7
0.9
2.2
0
1007.7
988.0
20.8
1007.8
...
18.7
1008.3
0.00
3
0
84.0
77.6
0
20.5
0
4321
4321
19.0
987.6
1.9
2.2
0
1007.7
988.0
20.6
1007.8
...
19.0
1008.3
0.00
3
0
84.4
79.2
0
20.5
0
4322
4322
19.1
987.6
2.0
1.4
0
1007.8
988.1
20.5
1007.9
...
19.2
1008.3
0.00
3
0
84.9
79.9
0
20.5
0
4323
4323
19.2
987.7
1.8
1.5
0
1007.9
988.1
20.5
1007.9
...
19.3
1008.4
0.00
3
0
84.9
80.7
0
20.5
0
4324
4324
19.2
987.8
1.4
1.4
0
1007.9
988.1
20.5
1007.8
...
19.5
1008.4
0.00
3
0
84.9
80.9
0
20.5
0
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4747
4747
19.9
987.6
0.9
0.8
0
1006.9
987.7
21.7
1007.5
...
19.1
1007.5
22.16
2
0
82.3
58.6
1
21.5
23
4748
4748
19.9
987.6
0.5
0.7
0
1006.8
987.7
21.6
1007.5
...
19.2
1007.5
22.16
1
0
82.5
59.1
1
21.5
23
4749
4749
19.7
987.7
0.9
0.6
0
1006.9
987.6
21.4
1007.4
...
19.2
1007.5
22.16
2
0
83.0
58.9
1
21.5
23
4750
4750
19.4
987.7
0.9
0.8
0
1006.9
987.8
21.3
1007.6
...
19.3
1007.6
22.16
3
0
83.2
59.8
1
21.5
23
4751
4751
19.1
987.6
1.0
0.3
0
1006.8
987.8
21.2
1007.5
...
19.5
1007.7
22.16
1
0
84.0
59.5
1
21.0
23
432 rows × 43 columns
1 X_3s = df2.drop(columns=['X14' , 'X16' , 'X19' , 'Y18' , 'id' ], axis=1 )
이후 3일의 기상청데이터
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X31
X32
X33
X34
X35
X36
X37
X38
X39
hour
4320
19.3
987.7
0.9
2.2
0
1007.7
988.0
20.8
1007.8
1007.4
...
18.0
18.7
1008.3
0.00
3
0
84.0
77.6
0
0
4321
19.0
987.6
1.9
2.2
0
1007.7
988.0
20.6
1007.8
1007.4
...
17.7
19.0
1008.3
0.00
3
0
84.4
79.2
0
0
4322
19.1
987.6
2.0
1.4
0
1007.8
988.1
20.5
1007.9
1007.4
...
17.4
19.2
1008.3
0.00
3
0
84.9
79.9
0
0
4323
19.2
987.7
1.8
1.5
0
1007.9
988.1
20.5
1007.9
1007.5
...
17.5
19.3
1008.4
0.00
3
0
84.9
80.7
0
0
4324
19.2
987.8
1.4
1.4
0
1007.9
988.1
20.5
1007.8
1007.6
...
17.4
19.5
1008.4
0.00
3
0
84.9
80.9
0
0
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4747
19.9
987.6
0.9
0.8
0
1006.9
987.7
21.7
1007.5
1007.4
...
17.7
19.1
1007.5
22.16
2
0
82.3
58.6
1
23
4748
19.9
987.6
0.5
0.7
0
1006.8
987.7
21.6
1007.5
1007.4
...
17.7
19.2
1007.5
22.16
1
0
82.5
59.1
1
23
4749
19.7
987.7
0.9
0.6
0
1006.9
987.6
21.4
1007.4
1007.5
...
17.8
19.2
1007.5
22.16
2
0
83.0
58.9
1
23
4750
19.4
987.7
0.9
0.8
0
1006.9
987.8
21.3
1007.6
1007.5
...
17.7
19.3
1007.6
22.16
3
0
83.2
59.8
1
23
4751
19.1
987.6
1.0
0.3
0
1006.8
987.8
21.2
1007.5
1007.4
...
17.7
19.5
1007.7
22.16
1
0
84.0
59.5
1
23
432 rows × 38 columns
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 model0 = sm.OLS.from_formula("Y00 ~ 1 +" + a, data=Y00_df) model1 = sm.OLS.from_formula("Y01 ~ 1 +" + a, data=Y01_df) model2 = sm.OLS.from_formula("Y02 ~ 1 +" + a, data=Y02_df) model3 = sm.OLS.from_formula("Y03 ~ 1 +" + a, data=Y03_df) model4 = sm.OLS.from_formula("Y04 ~ 1 +" + a, data=Y04_df) model5 = sm.OLS.from_formula("Y05 ~ 1 +" + a, data=Y05_df) model6 = sm.OLS.from_formula("Y06 ~ 1 +" + a, data=Y06_df) model7 = sm.OLS.from_formula("Y07 ~ 1 +" + a, data=Y07_df) model8 = sm.OLS.from_formula("Y08 ~ 1 +" + a, data=Y08_df) model9 = sm.OLS.from_formula("Y09 ~ 1 +" + a, data=Y09_df) mode110 = sm.OLS.from_formula("Y10 ~ 1 +" + a, data=Y10_df) model11 = sm.OLS.from_formula("Y11 ~ 1 +" + a, data=Y11_df) model12 = sm.OLS.from_formula("Y12 ~ 1 +" + a, data=Y12_df) model13 = sm.OLS.from_formula("Y13 ~ 1 +" + a, data=Y13_df) model14 = sm.OLS.from_formula("Y14 ~ 1 +" + a, data=Y14_df) model15 = sm.OLS.from_formula("Y15 ~ 1 +" + a, data=Y15_df) model16 = sm.OLS.from_formula("Y16 ~ 1 +" + a, data=Y16_df) model17 = sm.OLS.from_formula("Y17 ~ 1 +" + a, data=Y17_df)
1 models = [model0, model1, model2, model3, model4, model5, model6, model7, model8, model9, model10, model11, model12, model13, model14, model15, model16, model17]
1 2 3 4 5 6 7 Y_pred = [] for model in models: result = model.fit() result = result.predict(X_3s).round(2 ) Y_pred.append(result) Y_pred
[4320 21.36
4321 21.25
4322 21.09
4323 21.03
4324 21.02
...
4747 22.56
4748 22.44
4749 22.21
4750 22.13
4751 22.02
Length: 432, dtype: float64, 4320 21.83
4321 21.55
4322 21.57
4323 21.35
4324 21.42
...
4747 22.25
4748 21.99
4749 22.06
4750 21.82
4751 21.43
Length: 432, dtype: float64, 4320 21.58
4321 21.42
4322 21.38
4323 21.30
4324 21.37
...
4747 22.03
4748 21.86
4749 21.73
4750 21.50
4751 21.34
Length: 432, dtype: float64, 4320 25.63
4321 25.49
4322 25.48
4323 25.24
4324 25.25
...
4747 27.11
4748 27.09
4749 27.02
4750 26.94
4751 26.84
Length: 432, dtype: float64, 4320 25.70
4321 25.52
4322 25.47
4323 25.24
4324 25.24
...
4747 27.18
4748 27.14
4749 27.07
4750 26.95
4751 26.84
Length: 432, dtype: float64, 4320 20.22
4321 19.86
4322 20.11
4323 20.02
4324 19.95
...
4747 19.51
4748 19.39
4749 19.28
4750 19.29
4751 18.73
Length: 432, dtype: float64, 4320 18.43
4321 18.65
4322 18.50
4323 18.72
4324 18.67
...
4747 17.86
4748 17.81
4749 17.76
4750 17.75
4751 17.81
Length: 432, dtype: float64, 4320 18.44
4321 19.08
4322 18.90
4323 19.42
4324 19.29
...
4747 18.67
4748 18.59
4749 18.41
4750 18.50
4751 18.89
Length: 432, dtype: float64, 4320 20.46
4321 19.56
4322 19.49
4323 19.72
4324 19.72
...
4747 18.71
4748 18.73
4749 18.45
4750 18.41
4751 17.68
Length: 432, dtype: float64, 4320 19.21
4321 18.63
4322 18.35
4323 18.45
4324 18.49
...
4747 17.77
4748 17.91
4749 17.91
4750 17.53
4751 17.39
Length: 432, dtype: float64, 4320 20.82
4321 19.94
4322 20.01
4323 19.85
4324 19.92
...
4747 19.15
4748 19.29
4749 19.85
4750 19.29
4751 19.14
Length: 432, dtype: float64, 4320 19.03
4321 18.60
4322 18.53
4323 18.48
4324 18.49
...
4747 17.62
4748 17.78
4749 17.58
4750 17.32
4751 17.21
Length: 432, dtype: float64, 4320 18.23
4321 18.58
4322 18.58
4323 18.69
4324 18.63
...
4747 18.96
4748 18.77
4749 18.65
4750 18.76
4751 18.71
Length: 432, dtype: float64, 4320 19.66
4321 19.08
4322 18.87
4323 18.87
4324 18.79
...
4747 19.37
4748 19.34
4749 19.27
4750 19.20
4751 18.74
Length: 432, dtype: float64, 4320 20.33
4321 19.07
4322 18.58
4323 18.80
4324 18.74
...
4747 19.84
4748 19.96
4749 19.91
4750 19.73
4751 18.81
Length: 432, dtype: float64, 4320 19.40
4321 18.95
4322 18.74
4323 18.78
4324 18.72
...
4747 19.04
4748 19.08
4749 19.07
4750 18.80
4751 18.68
Length: 432, dtype: float64, 4320 18.74
4321 18.63
4322 18.52
4323 18.57
4324 18.60
...
4747 17.95
4748 17.98
4749 17.88
4750 17.84
4751 17.69
Length: 432, dtype: float64, 4320 19.58
4321 18.57
4322 18.37
4323 18.78
4324 18.75
...
4747 17.98
4748 18.20
4749 18.21
4750 17.78
4751 17.44
Length: 432, dtype: float64]1 2 3 4 5 6 7 8 9 Y = [] for x in range(18 ): if x < 10 : result = 'Y0' + str(x) Y.append(result) elif x >= 10 : result = 'Y' + str(x) Y.append(result) Y
['Y00',
'Y01',
'Y02',
'Y03',
'Y04',
'Y05',
'Y06',
'Y07',
'Y08',
'Y09',
'Y10',
'Y11',
'Y12',
'Y13',
'Y14',
'Y15',
'Y16',
'Y17']1 2 3 4 5 6 Y_pred = pd.DataFrame(Y_pred) Y_pred = Y_pred.T Y_pred.columns = Y Y_pred['hour' ] = df2['hour' ] Y_pred['Y18' ] = df2['Y18' ] Y_pred
Y00
Y01
Y02
Y03
Y04
Y05
Y06
Y07
Y08
Y09
Y10
Y11
Y12
Y13
Y14
Y15
Y16
Y17
hour
Y18
4320
21.36
21.83
21.58
25.63
25.70
20.22
18.43
18.44
20.46
19.21
20.82
19.03
18.23
19.66
20.33
19.40
18.74
19.58
0
20.5
4321
21.25
21.55
21.42
25.49
25.52
19.86
18.65
19.08
19.56
18.63
19.94
18.60
18.58
19.08
19.07
18.95
18.63
18.57
0
20.5
4322
21.09
21.57
21.38
25.48
25.47
20.11
18.50
18.90
19.49
18.35
20.01
18.53
18.58
18.87
18.58
18.74
18.52
18.37
0
20.5
4323
21.03
21.35
21.30
25.24
25.24
20.02
18.72
19.42
19.72
18.45
19.85
18.48
18.69
18.87
18.80
18.78
18.57
18.78
0
20.5
4324
21.02
21.42
21.37
25.25
25.24
19.95
18.67
19.29
19.72
18.49
19.92
18.49
18.63
18.79
18.74
18.72
18.60
18.75
0
20.5
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4747
22.56
22.25
22.03
27.11
27.18
19.51
17.86
18.67
18.71
17.77
19.15
17.62
18.96
19.37
19.84
19.04
17.95
17.98
23
21.5
4748
22.44
21.99
21.86
27.09
27.14
19.39
17.81
18.59
18.73
17.91
19.29
17.78
18.77
19.34
19.96
19.08
17.98
18.20
23
21.5
4749
22.21
22.06
21.73
27.02
27.07
19.28
17.76
18.41
18.45
17.91
19.85
17.58
18.65
19.27
19.91
19.07
17.88
18.21
23
21.5
4750
22.13
21.82
21.50
26.94
26.95
19.29
17.75
18.50
18.41
17.53
19.29
17.32
18.76
19.20
19.73
18.80
17.84
17.78
23
21.5
4751
22.02
21.43
21.34
26.84
26.84
18.73
17.81
18.89
17.68
17.39
19.14
17.21
18.71
18.74
18.81
18.68
17.69
17.44
23
21.0
432 rows × 20 columns
1 2 3 4 5 b = list(Y_pred.columns) b.remove('Y18' ) b.remove('hour' ) b = ['scale({})' .format(x) for x in b] '+' .join(b)
'scale(Y00)+scale(Y01)+scale(Y02)+scale(Y03)+scale(Y04)+scale(Y05)+scale(Y06)+scale(Y07)+scale(Y08)+scale(Y09)+scale(Y10)+scale(Y11)+scale(Y12)+scale(Y13)+scale(Y14)+scale(Y15)+scale(Y16)+scale(Y17)'
Y00 ~ Y17로 Y18 예측 모델 만들기
1 2 3 model20 = sm.OLS.from_formula("Y18 ~ 1 +" + '+' .join(b), Y_pred) result20 = model20.fit() print(result20.summary())
OLS Regression Results
==============================================================================
Dep. Variable: Y18 R-squared: 0.956
Model: OLS Adj. R-squared: 0.954
Method: Least Squares F-statistic: 495.8
Date: Mon, 13 Apr 2020 Prob (F-statistic): 1.52e-266
Time: 15:30:41 Log-Likelihood: -733.14
No. Observations: 432 AIC: 1504.
Df Residuals: 413 BIC: 1582.
Df Model: 18
Covariance Type: nonrobust
==============================================================================
coef std err t P>|t| [0.025 0.975]
------------------------------------------------------------------------------
Intercept 26.6875 0.065 410.667 0.000 26.560 26.815
scale(Y00) 0.6176 0.838 0.737 0.462 -1.030 2.266
scale(Y01) 0.5445 3.285 0.166 0.868 -5.914 7.002
scale(Y02) 2.6717 3.208 0.833 0.405 -3.635 8.978
scale(Y03) -0.0450 4.271 -0.011 0.992 -8.440 8.350
scale(Y04) -0.0445 4.078 -0.011 0.991 -8.060 7.971
scale(Y05) -1.5686 0.463 -3.388 0.001 -2.479 -0.658
scale(Y06) 2.3657 1.095 2.160 0.031 0.213 4.518
scale(Y07) -4.3842 0.590 -7.429 0.000 -5.544 -3.224
scale(Y08) -6.4418 0.869 -7.411 0.000 -8.151 -4.733
scale(Y09) -20.4360 2.400 -8.517 0.000 -25.153 -15.719
scale(Y10) -5.2662 1.781 -2.957 0.003 -8.767 -1.765
scale(Y11) 14.8776 1.777 8.370 0.000 11.384 18.372
scale(Y12) 3.4542 1.059 3.262 0.001 1.373 5.536
scale(Y13) 5.8498 1.715 3.411 0.001 2.479 9.221
scale(Y14) -2.1079 0.742 -2.843 0.005 -3.566 -0.650
scale(Y15) 2.3095 0.881 2.622 0.009 0.578 4.041
scale(Y16) 2.0766 1.413 1.470 0.142 -0.700 4.854
scale(Y17) 11.3015 1.191 9.493 0.000 8.961 13.642
==============================================================================
Omnibus: 52.523 Durbin-Watson: 0.502
Prob(Omnibus): 0.000 Jarque-Bera (JB): 140.165
Skew: -0.589 Prob(JB): 3.66e-31
Kurtosis: 5.530 Cond. No. 372.
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.1 2 3 4 5 6 7 8 N = len(Y_pred) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = Y_pred.iloc[idx_train] dfX_test = Y_pred.iloc[idx_test]
Y18 모델 KFold 검증
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 scores = np.zeros(5 ) for i, (idx_train, idx_test) in enumerate(kfold.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model20 = sm.OLS.from_formula("Y18 ~ 1 +" + '+' .join(b), data=dfX_train) result20 = model20.fit() pred = result20.predict(df_test) rss = ((df_test.Y18 - pred) ** 2 ).sum() tss = ((df_test.Y18 - df_test.Y18.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result20.rsquared, rsquared)) pred = result20.predict(dfX_test) rss = ((dfX_test.Y18 - pred) ** 2 ).sum() tss = ((dfX_test.Y18 - dfX_test.Y18.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared) print(mean_squared_error(dfX_test.Y18, pred))
학습 R2 = 0.95404730, 검증 R2 = 0.95265858
학습 R2 = 0.95404730, 검증 R2 = 0.93511920
학습 R2 = 0.95404730, 검증 R2 = 0.96495716
학습 R2 = 0.95404730, 검증 R2 = 0.96417472
학습 R2 = 0.95404730, 검증 R2 = 0.94263016
0.9504900638797171
1.82855707807725981 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y18' ], pred) plt.plot([0 , dfX_test['Y18' ].max()], [0 , dfX_test['Y18' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
accuracy_score와 mse가 좋게 나왔으나, 이를 활용해 구한 전체 데이터로 예측모델을 만들 경우 성능이 좋지 못함
다른 방법이 필요(Lasso, Ridge, EN)
Ridge, Lasso, EN 정규화를 모두 사용했을 때 Lasso가 가장 좋은 성능을 가짐
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 from sklearn.linear_model import Lassofrom sklearn.metrics import mean_squared_errorfrom sklearn.model_selection import cross_val_scoreX = Y_pred.drop(columns=['Y18' ]) y = Y_pred['Y18' ] alphas = np.logspace(-4 , 0 , 200 ) train_scores = [] test_scores = [] for alpha in alphas: model = Lasso(alpha=alpha) train_score = -mean_squared_error(y, model.fit(X, y).predict(X)) test_score = np.mean(cross_val_score(model, X, y, scoring="neg_mean_squared_error" , cv=5 )) train_scores.append(train_score) test_scores.append(test_score) optimal_alpha = alphas[np.argmax(test_scores)] optimal_score = np.max(test_scores) print(optimal_alpha) plt.plot(alphas, test_scores, "-" , label="검증 성능" ) plt.plot(alphas, train_scores, "--" , label="학습 성능" ) plt.axhline(optimal_score, linestyle=':' ) plt.axvline(optimal_alpha, linestyle=':' ) plt.scatter(optimal_alpha, optimal_score) plt.title("최적 정규화" ) plt.ylabel('성능' ) plt.xlabel('정규화 가중치' ) plt.legend() plt.show()
검증
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 from sklearn.linear_model import Lasso, RidgeX = dfX_train.drop(columns=['Y18' ]) y = dfX_train['Y18' ] for train_idx, test_idx in kfold.split(X): X_train, X_test = X.iloc[train_idx], X.iloc[test_idx] y_train, y_test = y.iloc[train_idx], y.iloc[test_idx] model18 = Lasso(0.172 ) model18.fit(X_train, y_train) pred = model18.predict(X_test) rss = ((y_test - pred) ** 2 ).sum() tss = ((y_test - y_test.mean())** 2 ).sum() rsquared = 1 - rss / tss print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(model18.score(X_test, y_test), rsquared)) pred = model18.predict(dfX_test.drop(columns=['Y18' ])) rss = ((dfX_test.Y18 - pred) ** 2 ).sum() tss = ((dfX_test.Y18 - dfX_test.Y18.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared) print(mean_squared_error(dfX_test.Y18, pred))
학습 R2 = 0.92285981, 검증 R2 = 0.92285981
학습 R2 = 0.89027532, 검증 R2 = 0.89027532
학습 R2 = 0.94315430, 검증 R2 = 0.94315430
학습 R2 = 0.93493313, 검증 R2 = 0.93493313
학습 R2 = 0.90241774, 검증 R2 = 0.90241774
0.9127293186631104
3.22317972052117161 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y18' ], pred) plt.plot([0 , dfX_test['Y18' ].max()], [0 , dfX_test['Y18' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
1 model18.fit(Y_pred.drop(columns=['Y18' ]), Y_pred['Y18' ])
Lasso(alpha=0.172, copy_X=True, fit_intercept=True, max_iter=1000,
normalize=False, positive=False, precompute=False, random_state=None,
selection='cyclic', tol=0.0001, warm_start=False)1 2 3 c = sensor_name c.append('hour' ) c
['Y00',
'Y01',
'Y02',
'Y03',
'Y04',
'Y05',
'Y06',
'Y07',
'Y08',
'Y09',
'Y10',
'Y11',
'Y12',
'Y13',
'Y14',
'Y15',
'Y16',
'Y17',
'hour']1 result20.predict(df1[c]).round(2 )
0 11.30
1 13.29
2 10.72
3 12.32
4 11.70
...
4315 22.28
4316 21.57
4317 19.64
4318 19.37
4319 18.30
Length: 4320, dtype: float641 2 X_30s['Y18' ] = result20.predict(df1[c]).round(2 ) X_30s
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X32
X33
X34
X35
X36
X37
X38
X39
hour
Y18
0
9.7
988.8
1.2
0.6
0
1009.3
989.6
12.2
1009.9
1009.8
...
10.7
1010.1
0.00
2
0
77.2
62.6
0
0
11.30
1
9.3
988.9
1.7
1.9
0
1009.3
989.6
12.1
1010.0
1009.9
...
10.3
1010.1
0.00
2
0
77.3
63.5
0
0
13.29
2
9.4
989.0
1.1
2.3
0
1009.2
989.7
12.1
1010.1
1010.1
...
9.7
1010.0
0.00
2
0
77.3
63.9
0
0
10.72
3
9.4
988.9
1.5
0.7
0
1009.2
989.6
12.0
1010.0
1010.0
...
9.4
1010.1
0.00
2
0
77.5
64.5
0
0
12.32
4
9.2
988.9
0.8
1.7
0
1009.2
989.7
12.0
1010.1
1010.0
...
9.4
1010.1
0.00
1
0
78.0
65.0
0
0
11.70
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4315
19.5
987.8
1.7
0.4
0
1007.8
988.0
21.3
1007.8
1007.5
...
19.4
1008.2
16.68
3
0
82.3
74.3
0
23
22.28
4316
19.3
987.7
2.1
0.9
0
1007.8
988.1
21.3
1008.0
1007.4
...
18.5
1008.4
16.68
3
0
82.4
74.8
0
23
21.57
4317
19.5
987.8
0.9
1.3
0
1007.8
988.3
21.2
1008.1
1007.5
...
19.1
1008.4
16.68
3
0
82.8
75.4
0
23
19.64
4318
20.0
987.8
1.4
0.5
0
1007.8
988.3
21.1
1008.1
1007.5
...
19.0
1008.4
16.68
3
0
82.8
75.8
0
23
19.37
4319
20.1
987.8
1.2
1.2
0
1007.8
988.2
20.9
1008.1
1007.5
...
18.5
1008.3
16.68
3
0
83.5
76.9
0
23
18.30
4320 rows × 39 columns
이전 30일의 Y18 예측
1 2 X_30s['Y18' ] = model18.predict(df1[c]).round(2 ) X_30s
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X32
X33
X34
X35
X36
X37
X38
X39
hour
Y18
0
9.7
988.8
1.2
0.6
0
1009.3
989.6
12.2
1009.9
1009.8
...
10.7
1010.1
0.00
2
0
77.2
62.6
0
0
10.42
1
9.3
988.9
1.7
1.9
0
1009.3
989.6
12.1
1010.0
1009.9
...
10.3
1010.1
0.00
2
0
77.3
63.5
0
0
10.41
2
9.4
989.0
1.1
2.3
0
1009.2
989.7
12.1
1010.1
1010.1
...
9.7
1010.0
0.00
2
0
77.3
63.9
0
0
9.81
3
9.4
988.9
1.5
0.7
0
1009.2
989.6
12.0
1010.0
1010.0
...
9.4
1010.1
0.00
2
0
77.5
64.5
0
0
9.94
4
9.2
988.9
0.8
1.7
0
1009.2
989.7
12.0
1010.1
1010.0
...
9.4
1010.1
0.00
1
0
78.0
65.0
0
0
9.56
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4315
19.5
987.8
1.7
0.4
0
1007.8
988.0
21.3
1007.8
1007.5
...
19.4
1008.2
16.68
3
0
82.3
74.3
0
23
21.12
4316
19.3
987.7
2.1
0.9
0
1007.8
988.1
21.3
1008.0
1007.4
...
18.5
1008.4
16.68
3
0
82.4
74.8
0
23
20.91
4317
19.5
987.8
0.9
1.3
0
1007.8
988.3
21.2
1008.1
1007.5
...
19.1
1008.4
16.68
3
0
82.8
75.4
0
23
20.65
4318
20.0
987.8
1.4
0.5
0
1007.8
988.3
21.1
1008.1
1007.5
...
19.0
1008.4
16.68
3
0
82.8
75.8
0
23
20.66
4319
20.1
987.8
1.2
1.2
0
1007.8
988.2
20.9
1008.1
1007.5
...
18.5
1008.3
16.68
3
0
83.5
76.9
0
23
20.40
4320 rows × 39 columns
1 2 X_3s['Y18' ] = df2['Y18' ] X_3s
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X32
X33
X34
X35
X36
X37
X38
X39
hour
Y18
4320
19.3
987.7
0.9
2.2
0
1007.7
988.0
20.8
1007.8
1007.4
...
18.7
1008.3
0.00
3
0
84.0
77.6
0
0
20.5
4321
19.0
987.6
1.9
2.2
0
1007.7
988.0
20.6
1007.8
1007.4
...
19.0
1008.3
0.00
3
0
84.4
79.2
0
0
20.5
4322
19.1
987.6
2.0
1.4
0
1007.8
988.1
20.5
1007.9
1007.4
...
19.2
1008.3
0.00
3
0
84.9
79.9
0
0
20.5
4323
19.2
987.7
1.8
1.5
0
1007.9
988.1
20.5
1007.9
1007.5
...
19.3
1008.4
0.00
3
0
84.9
80.7
0
0
20.5
4324
19.2
987.8
1.4
1.4
0
1007.9
988.1
20.5
1007.8
1007.6
...
19.5
1008.4
0.00
3
0
84.9
80.9
0
0
20.5
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4747
19.9
987.6
0.9
0.8
0
1006.9
987.7
21.7
1007.5
1007.4
...
19.1
1007.5
22.16
2
0
82.3
58.6
1
23
21.5
4748
19.9
987.6
0.5
0.7
0
1006.8
987.7
21.6
1007.5
1007.4
...
19.2
1007.5
22.16
1
0
82.5
59.1
1
23
21.5
4749
19.7
987.7
0.9
0.6
0
1006.9
987.6
21.4
1007.4
1007.5
...
19.2
1007.5
22.16
2
0
83.0
58.9
1
23
21.5
4750
19.4
987.7
0.9
0.8
0
1006.9
987.8
21.3
1007.6
1007.5
...
19.3
1007.6
22.16
3
0
83.2
59.8
1
23
21.5
4751
19.1
987.6
1.0
0.3
0
1006.8
987.8
21.2
1007.5
1007.4
...
19.5
1007.7
22.16
1
0
84.0
59.5
1
23
21.0
432 rows × 39 columns
전체 X데이터로 Y18 예측모델 만들기
1 2 all_X = pd.concat([X_30s, X_3s]) all_X
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X32
X33
X34
X35
X36
X37
X38
X39
hour
Y18
0
9.7
988.8
1.2
0.6
0
1009.3
989.6
12.2
1009.9
1009.8
...
10.7
1010.1
0.00
2
0
77.2
62.6
0
0
10.42
1
9.3
988.9
1.7
1.9
0
1009.3
989.6
12.1
1010.0
1009.9
...
10.3
1010.1
0.00
2
0
77.3
63.5
0
0
10.41
2
9.4
989.0
1.1
2.3
0
1009.2
989.7
12.1
1010.1
1010.1
...
9.7
1010.0
0.00
2
0
77.3
63.9
0
0
9.81
3
9.4
988.9
1.5
0.7
0
1009.2
989.6
12.0
1010.0
1010.0
...
9.4
1010.1
0.00
2
0
77.5
64.5
0
0
9.94
4
9.2
988.9
0.8
1.7
0
1009.2
989.7
12.0
1010.1
1010.0
...
9.4
1010.1
0.00
1
0
78.0
65.0
0
0
9.56
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
4747
19.9
987.6
0.9
0.8
0
1006.9
987.7
21.7
1007.5
1007.4
...
19.1
1007.5
22.16
2
0
82.3
58.6
1
23
21.50
4748
19.9
987.6
0.5
0.7
0
1006.8
987.7
21.6
1007.5
1007.4
...
19.2
1007.5
22.16
1
0
82.5
59.1
1
23
21.50
4749
19.7
987.7
0.9
0.6
0
1006.9
987.6
21.4
1007.4
1007.5
...
19.2
1007.5
22.16
2
0
83.0
58.9
1
23
21.50
4750
19.4
987.7
0.9
0.8
0
1006.9
987.8
21.3
1007.6
1007.5
...
19.3
1007.6
22.16
3
0
83.2
59.8
1
23
21.50
4751
19.1
987.6
1.0
0.3
0
1006.8
987.8
21.2
1007.5
1007.4
...
19.5
1007.7
22.16
1
0
84.0
59.5
1
23
21.00
4752 rows × 39 columns
1 2 3 4 5 6 7 8 N = len(all_X) ratio = 0.7 np.random.seed(0 ) idx_train = np.random.choice(np.arange(N), np.int(ratio * N)) idx_test = list(set(np.arange(N)).difference(idx_train)) dfX_train = all_X.iloc[idx_train] dfX_test = all_X.iloc[idx_test]
Y18 OLS report
1 2 3 model18 = sm.OLS.from_formula("Y18 ~ 1 +" + a, data=dfX_train) result18 = model18.fit() print(result18.summary())
OLS Regression Results
==============================================================================
Dep. Variable: Y18 R-squared: 0.967
Model: OLS Adj. R-squared: 0.966
Method: Least Squares F-statistic: 1472.
Date: Sat, 18 Apr 2020 Prob (F-statistic): 0.00
Time: 02:27:53 Log-Likelihood: -5817.7
No. Observations: 3326 AIC: 1.177e+04
Df Residuals: 3260 BIC: 1.217e+04
Df Model: 65
Covariance Type: nonrobust
=================================================================================
coef std err t P>|t| [0.025 0.975]
---------------------------------------------------------------------------------
Intercept 21.9494 0.179 122.879 0.000 21.599 22.300
C(X04)[T.1] 0.5592 0.250 2.237 0.025 0.069 1.049
C(X10)[T.1] 0.1419 0.281 0.506 0.613 -0.408 0.692
C(X21)[T.1] 0.8674 0.278 3.124 0.002 0.323 1.412
C(X36)[T.1] -2.5134 0.264 -9.514 0.000 -3.031 -1.995
C(X39)[T.1] -0.2722 0.160 -1.699 0.089 -0.586 0.042
C(X13)[T.1] 0.0005 0.081 0.006 0.996 -0.158 0.159
C(X13)[T.2] -0.0890 0.068 -1.306 0.192 -0.223 0.045
C(X13)[T.3] 0.1095 0.090 1.214 0.225 -0.067 0.286
C(X15)[T.1] -0.3793 0.087 -4.366 0.000 -0.550 -0.209
C(X15)[T.2] -0.3496 0.081 -4.331 0.000 -0.508 -0.191
C(X15)[T.3] -0.1822 0.090 -2.031 0.042 -0.358 -0.006
C(X17)[T.1] 0.1647 0.085 1.945 0.052 -0.001 0.331
C(X17)[T.2] 0.1234 0.082 1.499 0.134 -0.038 0.285
C(X17)[T.3] 0.0885 0.085 1.046 0.296 -0.077 0.255
C(X25)[T.1] -0.1689 0.108 -1.569 0.117 -0.380 0.042
C(X25)[T.2] 0.0230 0.089 0.258 0.796 -0.152 0.198
C(X25)[T.3] 0.1889 0.079 2.404 0.016 0.035 0.343
C(X35)[T.1] 0.2490 0.098 2.532 0.011 0.056 0.442
C(X35)[T.2] -0.1920 0.085 -2.259 0.024 -0.359 -0.025
C(X35)[T.3] 0.0621 0.087 0.714 0.475 -0.108 0.233
C(hour)[T.1] -0.0386 0.161 -0.240 0.810 -0.354 0.276
C(hour)[T.2] 0.1490 0.167 0.892 0.373 -0.179 0.477
C(hour)[T.3] 0.0931 0.167 0.559 0.576 -0.233 0.420
C(hour)[T.4] 0.1347 0.170 0.791 0.429 -0.199 0.469
C(hour)[T.5] 0.4355 0.172 2.530 0.011 0.098 0.773
C(hour)[T.6] 1.1144 0.173 6.443 0.000 0.775 1.454
C(hour)[T.7] 2.0508 0.177 11.578 0.000 1.703 2.398
C(hour)[T.8] 2.6423 0.180 14.706 0.000 2.290 2.995
C(hour)[T.9] 4.6882 0.190 24.630 0.000 4.315 5.061
C(hour)[T.10] 5.9639 0.200 29.816 0.000 5.572 6.356
C(hour)[T.11] 7.2187 0.213 33.969 0.000 6.802 7.635
C(hour)[T.12] 8.1562 0.223 36.513 0.000 7.718 8.594
C(hour)[T.13] 7.6495 0.235 32.535 0.000 7.189 8.111
C(hour)[T.14] 4.8837 0.238 20.509 0.000 4.417 5.351
C(hour)[T.15] 1.9181 0.251 7.651 0.000 1.427 2.410
C(hour)[T.16] 0.8868 0.255 3.475 0.001 0.386 1.387
C(hour)[T.17] 1.9417 0.260 7.473 0.000 1.432 2.451
C(hour)[T.18] 0.8412 0.268 3.144 0.002 0.317 1.366
C(hour)[T.19] 0.7748 0.270 2.869 0.004 0.245 1.304
C(hour)[T.20] 0.6640 0.271 2.453 0.014 0.133 1.195
C(hour)[T.21] 0.6215 0.277 2.241 0.025 0.078 1.165
C(hour)[T.22] 0.4593 0.274 1.674 0.094 -0.079 0.997
C(hour)[T.23] 0.5301 0.283 1.876 0.061 -0.024 1.084
scale(X00) 0.8557 0.184 4.663 0.000 0.496 1.216
scale(X07) 2.9758 0.237 12.559 0.000 2.511 3.440
scale(X28) -1.0887 0.219 -4.964 0.000 -1.519 -0.659
scale(X31) 2.3334 0.215 10.876 0.000 1.913 2.754
scale(X32) 0.3004 0.212 1.419 0.156 -0.115 0.715
scale(X01) 1.4245 0.554 2.570 0.010 0.338 2.511
scale(X06) -5.1961 0.586 -8.865 0.000 -6.345 -4.047
scale(X22) 0.9602 0.597 1.609 0.108 -0.210 2.130
scale(X27) 1.9049 0.440 4.329 0.000 1.042 2.768
scale(X29) 0.8797 0.792 1.110 0.267 -0.674 2.433
scale(X12) -0.0224 0.154 -0.145 0.884 -0.325 0.280
scale(X20) -0.1559 0.124 -1.257 0.209 -0.399 0.087
scale(X30) -0.3912 0.138 -2.831 0.005 -0.662 -0.120
scale(X37) 0.2628 0.125 2.097 0.036 0.017 0.509
scale(X38) -0.6327 0.115 -5.501 0.000 -0.858 -0.407
scale(X11) -1.1408 0.304 -3.750 0.000 -1.737 -0.544
scale(X34) 0.8955 0.312 2.870 0.004 0.284 1.507
scale(X02) 0.1300 0.033 3.974 0.000 0.066 0.194
scale(X03) -0.1836 0.033 -5.512 0.000 -0.249 -0.118
scale(X18) 0.0055 0.040 0.136 0.892 -0.073 0.084
scale(X24) -0.1218 0.045 -2.734 0.006 -0.209 -0.034
scale(X26) -0.1876 0.043 -4.407 0.000 -0.271 -0.104
==============================================================================
Omnibus: 188.844 Durbin-Watson: 2.019
Prob(Omnibus): 0.000 Jarque-Bera (JB): 516.990
Skew: -0.293 Prob(JB): 5.46e-113
Kurtosis: 4.841 Cond. No. 128.
==============================================================================
Warnings:
[1] Standard Errors assume that the covariance matrix of the errors is correctly specified.1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 scores = np.zeros(5 ) cv = KFold(5 , shuffle=True , random_state=13 ) for i, (idx_train, idx_test) in enumerate(cv.split(dfX_train)): df_train = dfX_train.iloc[idx_train] df_test = dfX_train.iloc[idx_test] model18 = sm.OLS.from_formula("Y18 ~ 1 +" + a, data=dfX_train) result18 = model18.fit() pred = result18.predict(df_test) rss = ((df_test.Y18 - pred) ** 2 ).sum() tss = ((df_test.Y18 - df_test.Y18.mean())** 2 ).sum() rsquared = 1 - rss / tss scores[i] = rsquared print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(result18.rsquared, rsquared)) pred = result18.predict(dfX_test) rss = ((dfX_test.Y18 - pred) ** 2 ).sum() tss = ((dfX_test.Y18 - dfX_test.Y18.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared) print(mean_squared_error(dfX_test.Y18, pred))
학습 R2 = 0.96704763, 검증 R2 = 0.96344232
학습 R2 = 0.96704763, 검증 R2 = 0.96605570
학습 R2 = 0.96704763, 검증 R2 = 0.96840262
학습 R2 = 0.96704763, 검증 R2 = 0.96691297
학습 R2 = 0.96704763, 검증 R2 = 0.96968124
0.9643708671481441
1.99137835827632361 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y18' ], pred) plt.plot([0 , dfX_test['Y18' ].max()], [0 , dfX_test['Y18' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
테스트 결과는?
Lasso는 어떨까?
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 from sklearn.linear_model import Lassofrom sklearn.metrics import mean_squared_errorfrom sklearn.model_selection import cross_val_scoreX = all_X.drop(columns=['Y18' ]) y = all_X['Y18' ] alphas = np.logspace(-4 , 0 , 200 ) train_scores = [] test_scores = [] for alpha in alphas: model = Lasso(alpha=alpha) train_score = -mean_squared_error(y, model.fit(X, y).predict(X)) test_score = np.mean(cross_val_score(model, X, y, scoring="neg_mean_squared_error" , cv=5 )) train_scores.append(train_score) test_scores.append(test_score) optimal_alpha = alphas[np.argmax(test_scores)] optimal_score = np.max(test_scores) print(optimal_alpha) plt.plot(alphas, test_scores, "-" , label="검증 성능" ) plt.plot(alphas, train_scores, "--" , label="학습 성능" ) plt.axhline(optimal_score, linestyle=':' ) plt.axvline(optimal_alpha, linestyle=':' ) plt.scatter(optimal_alpha, optimal_score) plt.title("최적 정규화" ) plt.ylabel('성능' ) plt.xlabel('정규화 가중치' ) plt.legend() plt.show()
0.01289890261253308{% asset_img final_blog_116_1.png %}
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 X = dfX_train.drop(columns=['Y18' ]) y = dfX_train['Y18' ] for train_idx, test_idx in kfold.split(X): X_train, X_test = X.iloc[train_idx], X.iloc[test_idx] y_train, y_test = y.iloc[train_idx], y.iloc[test_idx] model18 = Lasso(0.012 ) model18.fit(X_train, y_train) pred = model18.predict(X_test) rss = ((y_test - pred) ** 2 ).sum() tss = ((y_test - y_test.mean())** 2 ).sum() rsquared = 1 - rss / tss print("학습 R2 = {:.8f}, 검증 R2 = {:.8f}" .format(model18.score(X_test, y_test), rsquared)) pred = model18.predict(dfX_test.drop(columns=['Y18' ])) rss = ((dfX_test.Y18 - pred) ** 2 ).sum() tss = ((dfX_test.Y18 - dfX_test.Y18.mean())** 2 ).sum() rsquared = 1 - rss / tss print(rsquared) print(mean_squared_error(dfX_test.Y18, pred))
학습 R2 = 0.92249209, 검증 R2 = 0.92249209
학습 R2 = 0.91956967, 검증 R2 = 0.91956967
학습 R2 = 0.92019706, 검증 R2 = 0.92019706
학습 R2 = 0.92395915, 검증 R2 = 0.92395915
학습 R2 = 0.92177100, 검증 R2 = 0.92177100
0.9246531552968336
4.21127498472018851 2 3 4 5 6 7 plt.figure(figsize=(15 ,15 )) plt.scatter(dfX_test['Y18' ], pred) plt.plot([0 , dfX_test['Y18' ].max()], [0 , dfX_test['Y18' ].max()], "r" , lw=2 ) plt.xlabel("y_test" , fontsize=15 ) plt.ylabel("y_predict" , fontsize=15 ) plt.axis("equal" ) plt.show()
테스트 결과는?
시도해 본 모델들
각 측정영역 별 평균 모델(잘못된 방법)
OLS 모델
Lasso 모델
OLS + Lasso + OLS 모델
OLS + Lasso + Lasso 모델
+ 성능의 향상을 위해! 머신러닝 기법 적용
LGBM, RandomForestRegressor
RandomForestRegressor 성능이 제일 좋았음
RandomForestRegressor
1 2 3 4 X = all_X.drop(['Y18' ], axis=1 ) y = all_X['Y18' ] X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2 , random_state=13 )
1 rf_reg = RandomForestRegressor(random_state=13 , n_jobs=-1 , bootstrap=False , criterion='mse' )
GridSearchCV(cv=5, error_score=nan,
estimator=RandomForestRegressor(bootstrap=True, ccp_alpha=0.0,
criterion='mse', max_depth=None,
max_features='auto',
max_leaf_nodes=None,
max_samples=None,
min_impurity_decrease=0.0,
min_impurity_split=None,
min_samples_leaf=1,
min_samples_split=2,
min_weight_fraction_leaf=0.0,
n_estimators=100, n_jobs=-1,
oob_score=False, random_state=13,
verbose=0, warm_start=False),
iid='deprecated', n_jobs=-1,
param_grid={'max_depth': [6, 8, 10, 12, 16, 20, 24],
'min_samples_leaf': [8, 12],
'min_samples_split': [8, 12],
'n_estimators': [50, 100, 150, 200]},
pre_dispatch='2*n_jobs', refit=True, return_train_score=True,
scoring=make_scorer(MSE, greater_is_better=False), verbose=0)1 2 3 4 5 6 7 8 9 10 params = { 'max_depth' : [10 , 15 , 20 , 25 , 30 , 35 , 40 , 45 , 50 , 55 , 60 , 70 ], 'max_features' : [10 , 20 , 30 , 35 , 40 , 45 , 50 , 55 ], 'n_estimators' : [50 , 100 , 150 , 200 ], 'min_samples_leaf' : [8 , 12 ], 'min_samples_split' : [8 , 12 ], } grid_cv = GridSearchCV(rf_reg, param_grid=params, cv=5 , return_train_score=True , n_jobs=-1 ) grid_cv.fit(X_train, y_train)
GridSearchCV(cv=5, error_score=nan,
estimator=RandomForestRegressor(bootstrap=False, ccp_alpha=0.0,
criterion='mse', max_depth=None,
max_features='auto',
max_leaf_nodes=None,
max_samples=None,
min_impurity_decrease=0.0,
min_impurity_split=None,
min_samples_leaf=1,
min_samples_split=2,
min_weight_fraction_leaf=0.0,
n_estimators=100, n_jobs=-1,
oob_score=False, random_state=13,
verbose=0, warm_start=False),
iid='deprecated', n_jobs=-1,
param_grid={'max_depth': [10, 15, 20, 25, 30, 35, 40, 45, 50, 55,
60, 70],
'max_features': [10, 20, 30, 35, 40, 45, 50, 55],
'min_samples_leaf': [8, 12],
'min_samples_split': [8, 12],
'n_estimators': [50, 100, 150, 200]},
pre_dispatch='2*n_jobs', refit=True, return_train_score=True,
scoring=None, verbose=0)1 2 cv_results_df = pd.DataFrame(grid_cv.cv_results_) cv_results_df.columns
Index(['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time',
'param_max_depth', 'param_max_features', 'param_min_samples_leaf',
'param_min_samples_split', 'param_n_estimators', 'params',
'split0_test_score', 'split1_test_score', 'split2_test_score',
'split3_test_score', 'split4_test_score', 'mean_test_score',
'std_test_score', 'rank_test_score', 'split0_train_score',
'split1_train_score', 'split2_train_score', 'split3_train_score',
'split4_train_score', 'mean_train_score', 'std_train_score'],
dtype='object')1 2 cols = ['rank_test_score' , 'mean_test_score' , 'param_n_estimators' , 'param_max_depth' ] cv_results_df[cols].sort_values('rank_test_score' )
rank_test_score
mean_test_score
param_n_estimators
param_max_depth
1047
1
0.930728
200
50
1431
1
0.930728
200
70
919
1
0.930728
200
45
1303
1
0.930728
200
60
915
1
0.930728
200
45
...
...
...
...
...
730
1532
NaN
150
35
731
1533
NaN
200
35
732
1534
NaN
50
35
1022
1535
NaN
150
45
1535
1536
NaN
200
70
1536 rows × 4 columns
{'max_depth': 25,
'max_features': 20,
'min_samples_leaf': 8,
'min_samples_split': 8,
'n_estimators': 200}0.93072765543321641 2 3 4 5 6 rf_reg_best = grid_cv.best_estimator_ rf_reg_best.fit(X_train, y_train) pred = rf_reg_best.predict(X_test) print(mean_squared_error(y_test, pred))
3.5975712904104613
전체 모델 만들어 Test 데이터에 적용
1 2 3 4 5 6 7 8 9 params = { 'max_depth' : [6 , 8 , 10 , 12 , 16 , 20 , 24 ], 'n_estimators' : [50 , 100 , 150 , 200 ], 'min_samples_leaf' : [8 , 12 ], 'min_samples_split' : [8 , 12 ] } grid_cv = GridSearchCV(rf_reg, param_grid=params, cv=5 , return_train_score=True , n_jobs=-1 ) grid_cv.fit(X, y)
GridSearchCV(cv=5, error_score=nan,
estimator=RandomForestRegressor(bootstrap=True, ccp_alpha=0.0,
criterion='mse', max_depth=None,
max_features='auto',
max_leaf_nodes=None,
max_samples=None,
min_impurity_decrease=0.0,
min_impurity_split=None,
min_samples_leaf=1,
min_samples_split=2,
min_weight_fraction_leaf=0.0,
n_estimators=100, n_jobs=-1,
oob_score=False, random_state=13,
verbose=0, warm_start=False),
iid='deprecated', n_jobs=-1,
param_grid={'max_depth': [6, 8, 10, 12, 16, 20, 24],
'min_samples_leaf': [8, 12],
'min_samples_split': [8, 12],
'n_estimators': [50, 100, 150, 200]},
pre_dispatch='2*n_jobs', refit=True, return_train_score=True,
scoring=None, verbose=0){'max_depth': 16,
'min_samples_leaf': 8,
'min_samples_split': 8,
'n_estimators': 100}0.95584246885630081 2 cv_results_df = pd.DataFrame(grid_cv.cv_results_) cv_results_df.columns
Index(['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time',
'param_max_depth', 'param_min_samples_leaf', 'param_min_samples_split',
'param_n_estimators', 'params', 'split0_test_score',
'split1_test_score', 'split2_test_score', 'split3_test_score',
'split4_test_score', 'mean_test_score', 'std_test_score',
'rank_test_score', 'split0_train_score', 'split1_train_score',
'split2_train_score', 'split3_train_score', 'split4_train_score',
'mean_train_score', 'std_train_score'],
dtype='object')1 2 cols = ['rank_test_score' , 'mean_test_score' , 'param_n_estimators' , 'param_max_depth' ] cv_results_df[cols].sort_values('rank_test_score' )
rank_test_score
mean_test_score
param_n_estimators
param_max_depth
65
1
0.955842
100
16
69
1
0.955842
100
16
81
3
0.955842
100
20
101
3
0.955842
100
24
85
3
0.955842
100
20
...
...
...
...
...
11
107
0.947681
200
6
8
109
0.947502
50
6
12
109
0.947502
50
6
4
111
0.947374
50
6
0
111
0.947374
50
6
112 rows × 4 columns
1 2 rf_reg_best = grid_cv.best_estimator_ rf_reg_best.fit(X, y)
RandomForestRegressor(bootstrap=False, ccp_alpha=0.0, criterion='mse',
max_depth=25, max_features=20, max_leaf_nodes=None,
max_samples=None, min_impurity_decrease=0.0,
min_impurity_split=None, min_samples_leaf=8,
min_samples_split=8, min_weight_fraction_leaf=0.0,
n_estimators=200, n_jobs=-1, oob_score=False,
random_state=13, verbose=0, warm_start=False)1 test = pd.read_csv('datas/test.csv' , encoding='utf-8' )
1 2 hour= pd.Series((test.index%144 /6 ).astype(int)) test['hour' ] = list(hour)
1 test1 = test.drop(columns=['X14' , 'X16' , 'X19' , 'id' ], axis=1 )
1 2 3 for x in direction_name: test1[x] = test1[x].apply(lambda x: 0 if x < 90 else 1 if x < 180 else 2 if x < 270 else 3 )
1 2 3 for x in water_name: for i in range(1 , len(x)): test1[x].loc[i] = test1[x].loc[i] - test1[x].loc[i-1 ]
1 2 for x in water_name: test1[x] = test1[x].apply(lambda x: 0 if x == 0 else 1 )
TEST 데이터
X00
X01
X02
X03
X04
X05
X06
X07
X08
X09
...
X31
X32
X33
X34
X35
X36
X37
X38
X39
hour
0
18.7
987.4
1.5
1.3
0
1006.8
987.7
21.1
1007.5
1007.2
...
17.5
19.2
1007.6
0.00
2
0
84.0
59.9
0
0
1
18.8
987.4
1.1
1.2
0
1006.7
987.7
21.2
1007.5
1007.2
...
17.4
19.1
1007.5
0.00
1
0
84.0
60.5
0
0
2
19.0
987.4
1.3
1.2
0
1006.6
987.6
21.2
1007.4
1007.2
...
17.2
19.0
1007.5
0.00
2
0
85.0
60.8
0
0
3
18.7
987.4
2.3
0.8
0
1006.6
987.6
21.1
1007.4
1007.2
...
17.2
18.8
1007.4
0.00
2
0
85.8
61.2
0
0
4
18.4
987.4
1.1
0.7
0
1006.7
987.7
20.9
1007.5
1007.2
...
17.4
18.9
1007.5
0.00
1
0
87.3
61.7
0
0
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
...
11515
25.7
993.2
3.9
3.8
0
1011.7
993.8
26.3
1013.3
1012.6
...
23.6
24.4
1013.0
19.56
1
0
79.7
67.5
1
23
11516
25.6
993.2
3.0
3.0
0
1011.7
993.8
26.3
1013.3
1012.6
...
25.3
24.4
1012.9
19.56
1
0
79.6
67.9
1
23
11517
25.6
993.2
3.8
2.5
0
1011.9
993.7
26.2
1013.2
1012.7
...
24.9
24.3
1012.9
19.56
2
0
79.2
68.3
1
23
11518
25.5
993.2
2.4
2.8
0
1011.8
993.7
26.1
1013.2
1012.6
...
25.3
24.2
1012.9
19.56
1
0
80.1
68.8
1
23
11519
25.4
993.2
3.0
3.5
0
1011.9
993.6
26.0
1013.1
1012.6
...
25.4
24.3
1012.9
19.56
2
0
80.2
68.9
1
23
11520 rows × 38 columns
1 2 answer = rf_reg_best.predict(test1).round(2 ) answer
array([19.62, 19.61, 19.61, ..., 26.06, 24.58, 24.46])
결과는?
{% asset_img rf.png %}
{% asset_img final.png %}
1 2 3 4 answer_df = pd.DataFrame(answer) answer_df.rename(columns={ 0 : 'Y18' }, inplace=True )
1 2 answer_df['id' ] = test['id' ] answer_df = answer_df[['id' , 'Y18' ]]
1 answer_df.to_csv('submission_0412.csv' , index=False , encoding='utf-8' )
id
Y18
0
4752
19.62
1
4753
19.61
2
4754
19.61
3
4755
18.68
4
4756
19.52
...
...
...
11515
16267
25.63
11516
16268
26.11
11517
16269
26.06
11518
16270
24.58
11519
16271
24.46
11520 rows × 2 columns
