/
TSP.py
265 lines (182 loc) · 9.91 KB
/
TSP.py
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
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
# -*- coding: utf-8 -*-
"""
Created on Wed Jun 29 09:57:54 2016
@author: d_floriello
time series website ---> weather:
http://archivio-meteo.distile.it/tabelle-dati-archivio-meteo/
"""
## Time Series Analysis of PUN in Python
import statsmodels
import statsmodels.api as sm
#from statsmodels.tsa import stattools
#from statsmodels.graphics import tsaplots
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import Functions_for_TSP
import statsmodels.tsa.arima_model
data = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2010.xlsx")
import temp
pun = data["PUN"]
pun = data["CSUD"]
dates = temp.dates(data[data.columns[0]])
prova = pd.to_datetime(dates)
df = pd.DataFrame(pun)
df = df.set_index(prova)
dec = sm.tsa.seasonal_decompose(df, freq=24)
dec.plot()
min_season = np.array(dec.seasonal.ix[0:24])
plt.plot(min_season)
D,Y = temp.create_dataset(data, "ven")
Y = np.array(Y)
acf, Q, P, = statsmodels.tsa.stattools.acf(Y, nlags = 48, qstat = True)
statsmodels.graphics.tsaplots.plot_acf(Y, lags=1000)
per = statsmodels.tsa.stattools.periodogram(Y)
plt.plot(per)
S_per = pd.Series(per)
S_per.describe()
peaks = Functions_for_TSP.find_peaks(per, 10)
FE = Functions_for_TSP.fourierExtrapolation(Y, n_predict = 24)
fitted_FE = FE[0:8736]
diff = Y - fitted_FE
np.mean(diff)
np.var(diff)
#RMSE = np.sqrt(np.mean(diff**2))
sp_y = Functions_for_TSP.Signum_Process(Y)
sp_f = Functions_for_TSP.Signum_Process(fitted_FE)
sp_p = sp_y * sp_f
perc_err_andamento = sp_p[sp_p <= 0].size/sp_p.size
aic = statsmodels.tsa.stattools.arma_order_select_ic(Y, ic = 'aic')
#D2 = D.convert_objects(convert_numeric=True)
D3 = D[D.columns[0:215]].convert_objects(convert_numeric=True)
#Dnp = float(D.as_matrix(columns=D.columns[0:215]))
ardata = pd.DataFrame(data[data.columns[[2,3,7,10,11,18,20]]])
ardata = ardata.convert_objects(convert_numeric=True)
ardata = ardata.set_index(prova)
slovsviz = ardata['SLOV'] - ardata['SVIZ']
slovfran = ardata['SLOV'] - ardata['FRAN']
austfran = ardata['AUST'] - ardata['FRAN']
## strana perfetta correlazione tra FRAN, SVIZ, SLOV e AUST.
aic = statsmodels.tsa.stattools.arma_order_select_ic(ardata["PUN"], ic = ['aic', 'bic'])
arma_y = statsmodels.tsa.arima_model.ARIMA(endog=ardata["PUN"], order=[4,1,2],
exog=ardata[ardata.columns[[2,4]]].as_matrix()).fit(trend = 'c',
method = 'mle', solver = 'newton', maxiter = 100)
#### when ARMA-like models are used, DO NOT use the DataFrame coming from create_dataset:
#### ARMA-like methods require only the time series (the lags are computed internally)
#### the DataFrame from create_dataset has to be used only in NN-like methods
arma_y = statsmodels.tsa.arima_model.ARIMA(endog=ardata["PUN"], order=[4,1,2]).fit(trend = 'c', method = 'mle', maxiter = 100)
arma_y.resid
RMSE = np.sqrt(np.mean(arma_y.resid**2))
arma_y.forecast(steps=24)
### provo modello PUN, ORA, GIORNO, HOLIDAY
vac = temp.add_holidays(dates) ## <-
ad = temp.associate_days(data[data.columns[1]], 'ven')
yd = temp.generate_days(data[data.columns[1]], 'ven')
anglesd = np.array([temp.convert_day_to_angle(v) for v in yd]) ## <-
ora = np.sin(np.array(data[data.columns[1]])*np.pi/24) ## <-
arg = {'holiday' : vac, 'day' : anglesd, 'ora' : ora}
arg = pd.DataFrame(arg)
arfit = statsmodels.tsa.arima_model.ARIMA(endog=ardata["PUN"], order=[4,1,2],exog = arg.as_matrix()).fit(trend = 'c', method = 'mle', maxiter = 100)
rmse_fit = Functions_for_TSP.RMSE(arfit.resid) ## 7.7520042757584031
trainset = list(range(8016))
testset = list(range(8016,8760))
artfit = statsmodels.tsa.arima_model.ARIMA(endog=ardata["PUN"].ix[trainset], order=[4,1,2],exog = arg.ix[trainset].as_matrix()).fit(trend = 'c', method = 'mle', maxiter = 100)
art_forecast = artfit.forecast(steps = 744, exog = arg.ix[testset].as_matrix())
### http://statsmodels.sourceforge.net/devel/generated/statsmodels.tsa.arima_model.ARMAResults.html ####
#RMSE(art_forecast[1]-ardata["PUN"].ix[testset])
############################################################
############################################################
###### prova su tutti i dataset ############################
############################################################
data2 = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2011.xlsx")
data3 = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2012.xlsx")
data4 = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2013.xlsx")
data5 = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2014.xlsx")
data6 = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2015.xlsx")
vector_dates = np.concatenate([np.array(data[data.columns[0]]), np.array(data2[data2.columns[0]]),
np.array(data3[data3.columns[0]]),
np.array(data4[data4.columns[0]]), np.array(data5[data5.columns[0]]),
np.array(data6[data6.columns[0]])])
vector_ore = np.concatenate([np.array(data[data.columns[1]]), np.array(data2[data2.columns[1]]),
np.array(data3[data3.columns[1]]),
np.array(data4[data4.columns[1]]), np.array(data5[data5.columns[1]]),
np.array(data6[data6.columns[1]])])
pun = np.concatenate([np.array(data[data.columns[2]]), np.array(data2[data2.columns[2]]),
np.array(data3[data3.columns[2]]),
np.array(data4[data4.columns[2]]), np.array(data5[data5.columns[2]]),
np.array(data6[data6.columns[2]])])
global_dates = temp.dates(pd.Series(vector_dates))
vac_glob = temp.add_holidays(global_dates) ###
all_days = temp.generate_days(vector_ore, 'ven')
aad = np.array([temp.convert_day_to_angle(v) for v in all_days]) ## <-
aore = np.sin(np.array(vector_ore)*np.pi/24) ## <-
all_dict= {'holiday' : vac_glob, 'day' : aad, 'ora' : aore}
tot_data = pd.DataFrame(all_dict)
aicg = statsmodels.tsa.stattools.arma_order_select_ic(pun, ic = ['aic','bic'])
tot_model = statsmodels.tsa.arima_model.ARIMA(endog=pun, order=[4,1,2],exog = tot_data.as_matrix()).fit(trend = 'c', method = 'mle', maxiter = 100)
dataf = pd.read_excel("C:/Users/d_floriello/Documents/PUN/Anno 2016_04.xlsx")
datesf = temp.dates(dataf[dataf.columns[0]])
vacf = temp.add_holidays(datesf) ## <-
adf = temp.associate_days(dataf[dataf.columns[1]], 'ven')
ydf = temp.generate_days(dataf[dataf.columns[1]], 'ven')
anglesdf = np.array([temp.convert_day_to_angle(v) for v in ydf]) ## <-
oraf = np.sin(np.array(dataf[dataf.columns[1]])*np.pi/24) ## <-
argf = {'holiday' : vacf, 'day' : anglesdf, 'ora' : oraf}
argf = pd.DataFrame(argf)
forecast_04_16 = tot_model.forecast(steps = 2903, exog = argf.as_matrix())
predicted_pun_04_16 = forecast_04_16[1]
pundf = pd.DataFrame(pun)
pundf = pundf.set_index(pd.to_datetime(global_dates))
dec_pun = sm.tsa.seasonal_decompose(pundf, freq=24)
dec_pun.plot()
min_per_seasonality = np.array(dec_pun.seasonal.ix[0:24])
plt.plot(min_per_seasonality)
diff_pred = predicted_pun_04_16 - dataf[dataf.columns[2]]
spp = Functions_for_TSP.Signum_Process(predicted_pun_04_16)
asp = Functions_for_TSP.Signum_Process(dataf[dataf.columns[2]])
ppp = spp*asp
ppp[ppp <= 0].size/ppp.size
## a little bit of tuning ##
params_d = [1,2,3,4,6,12,24]
for d in params_d:
print("model with d = ", d)
fit = statsmodels.tsa.arima_model.ARIMA(endog=pun, order=[4,d,2],exog = tot_data.as_matrix()).fit(trend = 'c', method = 'mle', maxiter = 100)
pred = tot_model.forecast(steps = 2903, exog = argf.as_matrix())
dd = pred[0] - dataf[dataf.columns[2]]
print("computed RMSE:", RMSE(dd))
spp = Functions_for_TSP.Signum_Process(pred[1])
asp = Functions_for_TSP.Signum_Process(dataf[dataf.columns[2]])
ppp = spp*asp
print("error on sign process:", ppp[ppp <= 0].size/ppp.size)
##### eliminate trend and or seasonality ###
### ref: http://www.analyticsvidhya.com/blog/2016/02/time-series-forecasting-codes-python/
### trying eliminate global seasonality:
pun_des = pun - np.repeat(min_per_seasonality, pun.size/24)
des_pun = sm.tsa.seasonal_decompose(pun_des, freq=24)
des_pun.plot()
params_d = [1,2]
for d in params_d:
print("model with d = ", d)
fit = statsmodels.tsa.arima_model.ARIMA(endog=pun_des, order=[24,d,24],exog = tot_data.as_matrix()).fit(trend = 'c', method = 'mle', maxiter = 100)
pred = tot_model.forecast(steps = 2903, exog = argf.as_matrix())
dd = pred[1] - dataf[dataf.columns[2]]
print("computed RMSE:", RMSE(dd))
spp = Functions_for_TSP.Signum_Process(pred[1])
asp = Functions_for_TSP.Signum_Process(dataf[dataf.columns[2]])
ppp = spp*asp
print("error on sign process:", ppp[ppp <= 0].size/ppp.size)
pun2 = pd.Series(pun_des)
pun2.rolling(center = True, window= 24).mean()
####################################################################
############### new test ###########################################
fit = statsmodels.tsa.arima_model.ARIMA(endog=pun, order=[24,2,24],exog = tot_data.as_matrix()).fit(trend = 'c', maxiter = 100)
fit_des = statsmodels.tsa.arima_model.ARIMA(endog=pun_des, order=[24,2,24],exog = tot_data.as_matrix()).fit(trend = 'c', method='css', maxiter = 1000)
fit.rmse = Functions_for_TSP.RMSE(fit.resid)
fit_des.rmse = Functions_for_TSP.RMSE(fit_des.resid)
pred_fit = fit.forecast(steps = 2903, exog = argf.as_matrix())
pred_fit_des = fit_des.forecast(steps = 2903, exog = argf.as_matrix())
asp = Functions_for_TSP.Signum_Process(dataf[dataf.columns[2]])
fsp = Functions_for_TSP.Signum_Process(pred_fit[0])
fdsp = Functions_for_TSP.Signum_Process(pred_fit_des[0])
###################################################################
########### ARIMA con meteo #######################################