SVM - Práctica de regresión con SVR

Índice de contenido

python

import pandas as pd
from sklearn.svm import SVR
from sklearn.model_selection import GridSearchCV, train_test_split

python

df = pd.read_csv("C:/Users/user/Desktop/energydata_complete.csv", parse_dates = ["date"], infer_datetime_format =True)
df

	date	Appliances	lights	T1	RH_1	T2	RH_2	T3	RH_3	T4	...	T9	RH_9	T_out	Press_mm_hg	RH_out	Windspeed	Visibility	Tdewpoint	rv1	rv2
0	2016-01-11 17:00:00	60	30	19.890000	47.596667	19.200000	44.790000	19.790000	44.730000	19.000000	...	17.033333	45.5300	6.600000	733.5	92.000000	7.000000	63.000000	5.300000	13.275433	13.275433
1	2016-01-11 17:10:00	60	30	19.890000	46.693333	19.200000	44.722500	19.790000	44.790000	19.000000	...	17.066667	45.5600	6.483333	733.6	92.000000	6.666667	59.166667	5.200000	18.606195	18.606195
2	2016-01-11 17:20:00	50	30	19.890000	46.300000	19.200000	44.626667	19.790000	44.933333	18.926667	...	17.000000	45.5000	6.366667	733.7	92.000000	6.333333	55.333333	5.100000	28.642668	28.642668
3	2016-01-11 17:30:00	50	40	19.890000	46.066667	19.200000	44.590000	19.790000	45.000000	18.890000	...	17.000000	45.4000	6.250000	733.8	92.000000	6.000000	51.500000	5.000000	45.410389	45.410389
4	2016-01-11 17:40:00	60	40	19.890000	46.333333	19.200000	44.530000	19.790000	45.000000	18.890000	...	17.000000	45.4000	6.133333	733.9	92.000000	5.666667	47.666667	4.900000	10.084097	10.084097
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
19730	2016-05-27 17:20:00	100	0	25.566667	46.560000	25.890000	42.025714	27.200000	41.163333	24.700000	...	23.200000	46.7900	22.733333	755.2	55.666667	3.333333	23.666667	13.333333	43.096812	43.096812
19731	2016-05-27 17:30:00	90	0	25.500000	46.500000	25.754000	42.080000	27.133333	41.223333	24.700000	...	23.200000	46.7900	22.600000	755.2	56.000000	3.500000	24.500000	13.300000	49.282940	49.282940
19732	2016-05-27 17:40:00	270	10	25.500000	46.596667	25.628571	42.768571	27.050000	41.690000	24.700000	...	23.200000	46.7900	22.466667	755.2	56.333333	3.666667	25.333333	13.266667	29.199117	29.199117
19733	2016-05-27 17:50:00	420	10	25.500000	46.990000	25.414000	43.036000	26.890000	41.290000	24.700000	...	23.200000	46.8175	22.333333	755.2	56.666667	3.833333	26.166667	13.233333	6.322784	6.322784
19734	2016-05-27 18:00:00	430	10	25.500000	46.600000	25.264286	42.971429	26.823333	41.156667	24.700000	...	23.200000	46.8450	22.200000	755.2	57.000000	4.000000	27.000000	13.200000	34.118851	34.118851

19735 rows × 29 columns

python

df.set_index("date", inplace=True)

python

df

	Appliances	lights	T1	RH_1	T2	RH_2	T3	RH_3	T4	RH_4	...	T9	RH_9	T_out	Press_mm_hg	RH_out	Windspeed	Visibility	Tdewpoint	rv1	rv2
date
2016-01-11 17:00:00	60	30	19.890000	47.596667	19.200000	44.790000	19.790000	44.730000	19.000000	45.566667	...	17.033333	45.5300	6.600000	733.5	92.000000	7.000000	63.000000	5.300000	13.275433	13.275433
2016-01-11 17:10:00	60	30	19.890000	46.693333	19.200000	44.722500	19.790000	44.790000	19.000000	45.992500	...	17.066667	45.5600	6.483333	733.6	92.000000	6.666667	59.166667	5.200000	18.606195	18.606195
2016-01-11 17:20:00	50	30	19.890000	46.300000	19.200000	44.626667	19.790000	44.933333	18.926667	45.890000	...	17.000000	45.5000	6.366667	733.7	92.000000	6.333333	55.333333	5.100000	28.642668	28.642668
2016-01-11 17:30:00	50	40	19.890000	46.066667	19.200000	44.590000	19.790000	45.000000	18.890000	45.723333	...	17.000000	45.4000	6.250000	733.8	92.000000	6.000000	51.500000	5.000000	45.410389	45.410389
2016-01-11 17:40:00	60	40	19.890000	46.333333	19.200000	44.530000	19.790000	45.000000	18.890000	45.530000	...	17.000000	45.4000	6.133333	733.9	92.000000	5.666667	47.666667	4.900000	10.084097	10.084097
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
2016-05-27 17:20:00	100	0	25.566667	46.560000	25.890000	42.025714	27.200000	41.163333	24.700000	45.590000	...	23.200000	46.7900	22.733333	755.2	55.666667	3.333333	23.666667	13.333333	43.096812	43.096812
2016-05-27 17:30:00	90	0	25.500000	46.500000	25.754000	42.080000	27.133333	41.223333	24.700000	45.590000	...	23.200000	46.7900	22.600000	755.2	56.000000	3.500000	24.500000	13.300000	49.282940	49.282940
2016-05-27 17:40:00	270	10	25.500000	46.596667	25.628571	42.768571	27.050000	41.690000	24.700000	45.730000	...	23.200000	46.7900	22.466667	755.2	56.333333	3.666667	25.333333	13.266667	29.199117	29.199117
2016-05-27 17:50:00	420	10	25.500000	46.990000	25.414000	43.036000	26.890000	41.290000	24.700000	45.790000	...	23.200000	46.8175	22.333333	755.2	56.666667	3.833333	26.166667	13.233333	6.322784	6.322784
2016-05-27 18:00:00	430	10	25.500000	46.600000	25.264286	42.971429	26.823333	41.156667	24.700000	45.963333	...	23.200000	46.8450	22.200000	755.2	57.000000	4.000000	27.000000	13.200000	34.118851	34.118851

19735 rows × 28 columns

python

df.dtypes

date           datetime64[ns]
Appliances              int64
lights                  int64
T1                    float64
RH_1                  float64
T2                    float64
RH_2                  float64
T3                    float64
RH_3                  float64
T4                    float64
RH_4                  float64
T5                    float64
RH_5                  float64
T6                    float64
RH_6                  float64
T7                    float64
RH_7                  float64
T8                    float64
RH_8                  float64
T9                    float64
RH_9                  float64
T_out                 float64
Press_mm_hg           float64
RH_out                float64
Windspeed             float64
Visibility            float64
Tdewpoint             float64
rv1                   float64
rv2                   float64
dtype: object

python

regressor = SVR()

python

y = df.Appliances

python

X = df.drop("Appliances", axis=1)

python

X.columns

Index(['lights', 'T1', 'RH_1', 'T2', 'RH_2', 'T3', 'RH_3', 'T4', 'RH_4', 'T5',
       'RH_5', 'T6', 'RH_6', 'T7', 'RH_7', 'T8', 'RH_8', 'T9', 'RH_9', 'T_out',
       'Press_mm_hg', 'RH_out', 'Windspeed', 'Visibility', 'Tdewpoint', 'rv1',
       'rv2'],
      dtype='object')

python

X_train, X_test, y_train, y_test = train_test_split(X, y, 
                                                    random_state = 42,
                                                   test_size = 0.15)

python

paramgrid = {"kernel": ["linear", "poly", "rbf", "sigmoid"],
            "C": [0.2, 0.4, 0.6, 0.8, 1]}

python

gs = GridSearchCV(estimator = regressor,
                 param_grid = paramgrid,
                 scoring = None,
                 cv = None,
                 refit = True)

python

gs.fit(X_train, y_train)

GridSearchCV(estimator=SVR(),
             param_grid={'C': [0.2, 0.4, 0.6, 0.8, 1],
                         'kernel': ['linear', 'poly', 'rbf', 'sigmoid']})

python

pd.DataFrame(gs.cv_results_)

	mean_fit_time	std_fit_time	mean_score_time	std_score_time	param_C	param_kernel	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
0	40.752531	1.352108	2.373936	0.140572	0.2	linear	{'C': 0.2, 'kernel': 'linear'}	0.019653	0.010238	0.033227	0.043215	0.034625	0.028192	0.011728	5
1	18.326448	0.866918	2.413738	0.129136	0.2	poly	{'C': 0.2, 'kernel': 'poly'}	-0.118780	-0.115594	-0.115704	-0.108767	-0.113247	-0.114419	0.003328	13
2	19.781931	0.312770	2.794960	0.019261	0.2	rbf	{'C': 0.2, 'kernel': 'rbf'}	-0.129083	-0.125228	-0.126925	-0.118871	-0.123578	-0.124737	0.003455	15
3	21.603636	0.072545	3.587005	0.017782	0.2	sigmoid	{'C': 0.2, 'kernel': 'sigmoid'}	-0.137131	-0.132830	-0.135788	-0.127274	-0.131937	-0.132992	0.003429	20
4	41.414769	0.552810	1.692897	0.127614	0.4	linear	{'C': 0.4, 'kernel': 'linear'}	0.020779	0.011267	0.034262	0.044283	0.036474	0.029413	0.011818	4
5	18.166039	1.024226	2.379536	0.071229	0.4	poly	{'C': 0.4, 'kernel': 'poly'}	-0.111933	-0.109167	-0.108241	-0.102607	-0.106765	-0.107743	0.003071	10
6	23.986372	0.514323	3.469398	0.044869	0.4	rbf	{'C': 0.4, 'kernel': 'rbf'}	-0.119906	-0.116396	-0.116779	-0.109390	-0.113987	-0.115292	0.003500	14
7	24.208985	1.328853	3.954226	0.327962	0.4	sigmoid	{'C': 0.4, 'kernel': 'sigmoid'}	-0.135667	-0.131464	-0.134193	-0.125775	-0.130426	-0.131505	0.003422	19
8	55.259561	0.989018	1.674096	0.034429	0.6	linear	{'C': 0.6, 'kernel': 'linear'}	0.021643	0.011705	0.035039	0.045299	0.036830	0.030103	0.011921	3
9	16.357136	0.987940	2.138522	0.334483	0.6	poly	{'C': 0.6, 'kernel': 'poly'}	-0.107240	-0.104847	-0.102930	-0.098082	-0.102331	-0.103086	0.003033	8
10	19.806733	0.146611	2.820161	0.009880	0.6	rbf	{'C': 0.6, 'kernel': 'rbf'}	-0.115550	-0.112441	-0.111675	-0.105857	-0.109917	-0.111088	0.003188	12
11	22.374480	0.094818	3.743414	0.030038	0.6	sigmoid	{'C': 0.6, 'kernel': 'sigmoid'}	-0.134220	-0.130056	-0.132545	-0.124174	-0.128895	-0.129978	0.003448	18
12	64.584294	0.953269	1.651294	0.014635	0.8	linear	{'C': 0.8, 'kernel': 'linear'}	0.021894	0.011550	0.035023	0.046248	0.038090	0.030561	0.012323	2
13	15.359278	0.216740	1.953712	0.013808	0.8	poly	{'C': 0.8, 'kernel': 'poly'}	-0.103437	-0.101485	-0.098904	-0.094132	-0.098330	-0.099258	0.003152	7
14	19.679126	0.099504	2.811561	0.012094	0.8	rbf	{'C': 0.8, 'kernel': 'rbf'}	-0.112886	-0.109904	-0.109041	-0.103642	-0.107450	-0.108585	0.003039	11
15	22.378080	0.070287	3.765815	0.018436	0.8	sigmoid	{'C': 0.8, 'kernel': 'sigmoid'}	-0.132739	-0.128607	-0.130842	-0.122580	-0.127290	-0.128411	0.003462	17
16	75.441115	2.213122	1.738899	0.191793	1	linear	{'C': 1, 'kernel': 'linear'}	0.021846	0.011744	0.035227	0.046698	0.038177	0.030739	0.012410	1
17	15.253072	0.142732	1.980913	0.102427	1	poly	{'C': 1, 'kernel': 'poly'}	-0.100448	-0.098625	-0.095828	-0.090579	-0.095173	-0.096130	0.003367	6
18	19.578120	0.075016	2.799760	0.024994	1	rbf	{'C': 1, 'kernel': 'rbf'}	-0.110324	-0.107222	-0.106138	-0.100981	-0.104646	-0.105862	0.003070	9
19	22.398681	0.108951	3.797617	0.055737	1	sigmoid	{'C': 1, 'kernel': 'sigmoid'}	-0.131244	-0.127154	-0.129186	-0.121045	-0.125743	-0.126875	0.003458	16

python

gs.best_params_

{'C': 1, 'kernel': 'linear'}

python

gs.best_score_

0.0307386556178449

python

best_model = gs.best_estimator_

python

y_pred = best_model.predict(X_test)

python

pd.DataFrame({"y_test": y_test, "y_pred": y_pred})

	y_test	y_pred
date
2016-03-14 01:40:00	40	47.425845
2016-01-30 20:00:00	90	144.780909
2016-03-15 03:00:00	50	43.357670
2016-04-20 10:10:00	50	72.324751
2016-03-13 08:10:00	70	40.150215
...	...	...
2016-04-25 13:30:00	50	82.308629
2016-04-09 07:20:00	30	52.122389
2016-02-12 05:40:00	60	48.252534
2016-05-01 06:10:00	60	58.874192
2016-04-07 16:00:00	50	79.774316

2961 rows × 2 columns

python

best_model.score(X_test, y_test)

0.05339127348678119

python

best_model.get_params()

{'C': 1,
 'cache_size': 200,
 'coef0': 0.0,
 'degree': 3,
 'epsilon': 0.1,
 'gamma': 'scale',
 'kernel': 'linear',
 'max_iter': -1,
 'shrinking': True,
 'tol': 0.001,
 'verbose': False}