手写梯度下降

import pandas as pd
import numpy as np

train=pd.read_csv('train.csv')

train.head()

	Date	stations	observation	0	1	2	3	4	5	6	...	14	15	16	17	18	19	20	21	22	23
0	2014/1/1	station	AMB_TEMP	14	14	14	13	12	12	12	...	22	22	21	19	17	16	15	15	15	15
1	2014/1/1	station	CH4	1.8	1.8	1.8	1.8	1.8	1.8	1.8	...	1.8	1.8	1.8	1.8	1.8	1.8	1.8	1.8	1.8	1.8
2	2014/1/1	station	CO	0.51	0.41	0.39	0.37	0.35	0.3	0.37	...	0.37	0.37	0.47	0.69	0.56	0.45	0.38	0.35	0.36	0.32
3	2014/1/1	station	NMHC	0.2	0.15	0.13	0.12	0.11	0.06	0.1	...	0.1	0.13	0.14	0.23	0.18	0.12	0.1	0.09	0.1	0.08
4	2014/1/1	station	NO	0.9	0.6	0.5	1.7	1.8	1.5	1.9	...	2.5	2.2	2.5	2.3	2.1	1.9	1.5	1.6	1.8	1.5

5 rows × 27 columns

train.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 4320 entries, 0 to 4319
Data columns (total 27 columns):
Date           4320 non-null object
stations       4320 non-null object
observation    4320 non-null object
0              4320 non-null object
1              4320 non-null object
2              4320 non-null object
3              4320 non-null object
4              4320 non-null object
5              4320 non-null object
6              4320 non-null object
7              4320 non-null object
8              4320 non-null object
9              4320 non-null object
10             4320 non-null object
11             4320 non-null object
12             4320 non-null object
13             4320 non-null object
14             4320 non-null object
15             4320 non-null object
16             4320 non-null object
17             4320 non-null object
18             4320 non-null object
19             4320 non-null object
20             4320 non-null object
21             4320 non-null object
22             4320 non-null object
23             4320 non-null object
dtypes: object(27)
memory usage: 911.3+ KB

train.observation.unique()

array(['AMB_TEMP', 'CH4', 'CO', 'NMHC', 'NO', 'NO2', 'NOx', 'O3', 'PM10',
       'PM2.5', 'RAINFALL', 'RH', 'SO2', 'THC', 'WD_HR', 'WIND_DIREC',
       'WIND_SPEED', 'WS_HR'], dtype=object)

train_PM=train[train.observation=='PM2.5']

train_PM.head()

	Date	stations	observation	0	1	2	3	4	5	6	...	14	15	16	17	18	19	20	21	22	23
9	2014/1/1	station	PM2.5	26	39	36	35	31	28	25	...	36	45	42	49	45	44	41	30	24	13
27	2014/1/2	station	PM2.5	21	23	30	30	22	18	13	...	53	43	43	45	46	32	16	19	22	26
45	2014/1/3	station	PM2.5	19	25	27	20	16	14	15	...	32	36	34	45	40	41	23	29	23	37
63	2014/1/4	station	PM2.5	27	27	14	20	22	24	26	...	62	55	56	67	78	83	90	75	85	82
81	2014/1/5	station	PM2.5	80	80	76	81	75	66	70	...	64	73	57	57	53	70	70	60	68	66

5 rows × 27 columns

PM_data=train_PM.iloc[:,3:]

PM_data.head()

	0	1	2	3	4	5	6	7	8	9	...	14	15	16	17	18	19	20	21	22	23
9	26	39	36	35	31	28	25	20	19	30	...	36	45	42	49	45	44	41	30	24	13
27	21	23	30	30	22	18	13	13	11	22	...	53	43	43	45	46	32	16	19	22	26
45	19	25	27	20	16	14	15	8	4	9	...	32	36	34	45	40	41	23	29	23	37
63	27	27	14	20	22	24	26	33	48	50	...	62	55	56	67	78	83	90	75	85	82
81	80	80	76	81	75	66	70	65	66	57	...	64	73	57	57	53	70	70	60	68	66

5 rows × 24 columns

PM_data=PM_data.apply(lambda x : x.astype('float'))

PM_data.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 240 entries, 9 to 4311
Data columns (total 24 columns):
0     240 non-null float64
1     240 non-null float64
2     240 non-null float64
3     240 non-null float64
4     240 non-null float64
5     240 non-null float64
6     240 non-null float64
7     240 non-null float64
8     240 non-null float64
9     240 non-null float64
10    240 non-null float64
11    240 non-null float64
12    240 non-null float64
13    240 non-null float64
14    240 non-null float64
15    240 non-null float64
16    240 non-null float64
17    240 non-null float64
18    240 non-null float64
19    240 non-null float64
20    240 non-null float64
21    240 non-null float64
22    240 non-null float64
23    240 non-null float64
dtypes: float64(24)
memory usage: 46.9 KB

train_x=[]
train_y=[]
for i in range(15):
    x=PM_data.iloc[:,i:i+9]
    x.columns=np.arange(9)
    y=PM_data.iloc[:,i+9]
    y.columns=np.arange(1)
    train_x.append(x)
    train_y.append(y)

train_x=pd.concat(train_x,axis=0)
train_y=pd.concat(train_y,axis=0)

train_x.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 3600 entries, 9 to 4311
Data columns (total 9 columns):
0    3600 non-null float64
1    3600 non-null float64
2    3600 non-null float64
3    3600 non-null float64
4    3600 non-null float64
5    3600 non-null float64
6    3600 non-null float64
7    3600 non-null float64
8    3600 non-null float64
dtypes: float64(9)
memory usage: 281.2 KB

train_x.shape

(3600, 9)

train_y.shape

(3600,)

train_x=np.array(train_x)
train_y=np.array(train_y)

train_x=np.concatenate((np.ones((train_x.shape[0],1)),train_x),axis=1)

train_x.shape

(3600, 10)

w=np.zeros(train_x.shape[1])
lr=10
Iteration=10000
# sum_gra=np.zeros(train_x.shape[0])
sum_gra=np.zeros(train_x.shape[1])
for i in range(Iteration):
    y_new=np.dot(train_x,w)
    loss=y_new-train_y
    ##X^t * loss 为矩阵形式的梯度
    gra=np.dot(train_x.transpose(),loss)
    ###保存之前的梯度的平方
    sum_gra+=gra**2
    ada=np.sqrt(sum_gra)
    w=w-lr*gra/ada
w

array([ 2.15246702,  0.00728964, -0.04603067,  0.19941492, -0.20757788,
       -0.04384333,  0.46235285, -0.54329351,  0.01552538,  1.07716609])

df_test=pd.read_csv('test(1).csv')

df_test.shape

(4319, 11)

test_PM=df_test[df_test.AMB_TEMP=='PM2.5']
test_PM.head()

	id_0	AMB_TEMP	15	14	14.1	13	13.1	13.2	13.3	13.4	12
8	id_0	PM2.5	27	13	24	29	41	30	29	27	28
26	id_1	PM2.5	46	47	57	78	84	76	59	61	61
44	id_2	PM2.5	10	10	25	34	40	39	36	25	22
62	id_3	PM2.5	71	58	51	41	41	46	43	34	29
80	id_4	PM2.5	13	23	18	10	5	5	13	9	12

test_x=np.array(test_PM.iloc[:,2:],float)
 

test_x.shape

(240, 9)

test_x=np.concatenate([np.ones((test_x.shape[0],1)),test_x],axis=1)

test_x.shape

(240, 10)

y_pre=np.dot(test_x,w)

y_submit=pd.read_csv('sampleSubmission.csv')
y_submit.head()

	id	value
0	id_0	27.414421
1	id_1	61.555764
2	id_2	20.498032
3	id_3	29.534434
4	id_4	10.797670

y_submit.value=y_pre
y_submit.to_csv('sampleSubmission.csv',index=False)

y_real=pd.read_csv('answer.csv')
y_real.head()

	id	value
0	id_0	33
1	id_1	60
2	id_2	16
3	id_3	33
4	id_4	5

MAE

error_mae=abs(y_real.value-y_pre).sum()/y_real.shape[0]

error_mae

4.97442948413227

MSE

error_mse=sum((y_real.value-y_pre)**2)/y_real.shape[0]

error_mse

44.38203674794888

$R^2$

def r2_score(y_ture,y_predict):
    mse=sum((y_ture-y_predict)**2)
    variance=sum((y_ture-np.mean(y_predict))**2)
    return 1-mse/variance

error_r2=r2_score(y_real.value,y_pre)

error_r2

0.9094491525564087

使用sklearn

from sklearn import linear_model
reg=linear_model.LinearRegression()

reg.fit(train_x,train_y)

LinearRegression(copy_X=True, fit_intercept=True, n_jobs=None,
         normalize=False)

reg.coef_

array([ 0.        ,  0.00729085, -0.04603342,  0.1994169 , -0.20757678,
       -0.0438468 ,  0.46235542, -0.54329308,  0.01552321,  1.07716736])

y_prediction=reg.predict(test_x)

from sklearn.metrics import r2_score

r2_score(y_real.value,y_prediction)

0.90944810736946002

r2_score(y_real.value,y_pre)

0.90944807944600869