-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathexecute_models.py
213 lines (170 loc) · 16.7 KB
/
execute_models.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
import pandas as pd
import numpy as np
import datetime as dt
import os
import warnings
from tqdm import tqdm
from models import (HeterocedasticQuantileRegression,
QuantileRegressionAveraging,
AdaptiveConformalInference,
WidthAdaptiveConformalInference,
WidthAdaptiveConformalInference_2)
warnings.filterwarnings(action='ignore')
np.random.seed(123)
df = pd.read_csv("Data//df_2024.csv")
df['full_date'] = pd.to_datetime(df.full_date)
df['mean_pred'] = df[['pred1', 'pred2', 'pred3']].mean(axis=1)
df['std_pred'] = df[['pred1', 'pred2', 'pred3']].std(axis=1, ddof=0)
date_test = dt.datetime(2024, 1, 1)
idx_date_test = df[pd.to_datetime(df.date) == date_test].head(1).index.values[0]
idx_first_date_qr = idx_date_test - 6*30*24
m = 3
alphas = [0.2, 0.1]
for alpha in alphas:
if not os.path.isfile(f"Data//df_quantile_regression_alpha_{alpha}.csv"):
# Heterocedastic Quantile Regression
preds_inf_hqr, preds_sup_hqr = HeterocedasticQuantileRegression(idx_first_date_qr, df, alpha, by_hour=False, save_coefs=True)
df.loc[df.tail(len(preds_inf_hqr)).index, 'preds_inf_hqr'] = preds_inf_hqr
df.loc[df.tail(len(preds_sup_hqr)).index, 'preds_sup_hqr'] = preds_sup_hqr
# # Heterocedastic Quantile Regression by hour
# preds_inf_hqr_by_hour, preds_sup_hqr_by_hour = HeterocedasticQuantileRegression(idx_first_date_qr, df, alpha, by_hour=True)
# df.loc[df.tail(len(preds_inf_hqr_by_hour)).index, 'preds_inf_hqr_by_hour'] = preds_inf_hqr_by_hour
# df.loc[df.tail(len(preds_sup_hqr_by_hour)).index, 'preds_sup_hqr_by_hour'] = preds_sup_hqr_by_hour
# Heterocedastic Quantile Regression with varying weights
preds_inf_hqr_w, preds_sup_hqr_w = HeterocedasticQuantileRegression(idx_first_date_qr, df, alpha, by_hour=False, varying_weights=True, num_preds=m)
df.loc[df.tail(len(preds_inf_hqr_w)).index, 'preds_inf_hqr_w'] = preds_inf_hqr_w
df.loc[df.tail(len(preds_sup_hqr_w)).index, 'preds_sup_hqr_w'] = preds_sup_hqr_w
# # Heterocedastic Quantile Regression with varying weights by hour
# preds_inf_hqr_w_by_hour, preds_sup_hqr_w_by_hour = HeterocedasticQuantileRegression(idx_first_date_qr, df, alpha, by_hour=True, varying_weights=True, num_preds=m)
# df.loc[df.tail(len(preds_inf_hqr_w_by_hour)).index, 'preds_inf_hqr_w_by_hour'] = preds_inf_hqr_w_by_hour
# df.loc[df.tail(len(preds_sup_hqr_w_by_hour)).index, 'preds_sup_hqr_w_by_hour'] = preds_sup_hqr_w_by_hour
# Quantile Regression Averaging
preds_inf_qra, preds_sup_qra = QuantileRegressionAveraging(idx_first_date_qr, df, alpha, m=m, by_hour=False)
df.loc[df.tail(len(preds_inf_qra)).index, 'preds_inf_qra'] = preds_inf_qra
df.loc[df.tail(len(preds_sup_qra)).index, 'preds_sup_qra'] = preds_sup_qra
# # Quantile Regression Averaging by hour
# preds_inf_qra_by_hour, preds_sup_qra_by_hour = QuantileRegressionAveraging(idx_first_date_qr, df, alpha, m=m, by_hour=True)
# df.loc[df.tail(len(preds_inf_qra_by_hour)).index, 'preds_inf_qra_by_hour'] = preds_inf_qra_by_hour
# df.loc[df.tail(len(preds_sup_qra_by_hour)).index, 'preds_sup_qra_by_hour'] = preds_sup_qra_by_hour
# Save the dataframe (just in case)
df.to_csv(f"Data//df_quantile_regression_alpha_{alpha}.csv", index=False)
else:
# df = pd.read_csv(f"Data//df_quantile_regression_alpha_{alpha}.csv")
df = pd.read_csv(f"Data//df_final_alpha_{alpha}.csv")
df['full_date'] = pd.to_datetime(df.full_date)
df['length_hqr'] = df['preds_sup_hqr'] - df['preds_inf_hqr']
# df['length_hqr_by_hour'] = df['preds_sup_hqr_by_hour'] - df['preds_inf_hqr_by_hour']
df['length_hqr_w'] = df['preds_sup_hqr_w'] - df['preds_inf_hqr_w']
# df['length_hqr_w_by_hour'] = df['preds_sup_hqr_w_by_hour'] - df['preds_inf_hqr_w_by_hour']
df['length_qra'] = df['preds_sup_qra'] - df['preds_inf_qra']
# df['length_qra_by_hour'] = df['preds_sup_qra_by_hour'] - df['preds_inf_qra_by_hour']
# Adaptive Conformal Inference over HQR
preds_inf_aci_hqr, preds_sup_aci_hqr = AdaptiveConformalInference(idx_date_test, df, alpha, gamma=0.02, method='hqr', by_hour=False, save_alphas=True)
df.loc[df.tail(len(preds_inf_aci_hqr)).index, 'preds_inf_aci_hqr'] = preds_inf_aci_hqr
df.loc[df.tail(len(preds_sup_aci_hqr)).index, 'preds_sup_aci_hqr'] = preds_sup_aci_hqr
df.loc[df.preds_inf_aci_hqr == -np.inf, 'preds_inf_aci_hqr'] = df[df.index < idx_date_test].preds_inf_hqr.min()
df.loc[df.preds_sup_aci_hqr == np.inf, 'preds_sup_aci_hqr'] = df[df.index < idx_date_test].preds_sup_hqr.max()
# # Adaptive Conformal Inference over HQR by hour
# preds_inf_aci_hqr_by_hour, preds_sup_aci_hqr_by_hour = AdaptiveConformalInference(idx_date_test, df, alpha, gamma=0.02, method='hqr', by_hour=True)
# df.loc[df.tail(len(preds_inf_aci_hqr_by_hour)).index, 'preds_inf_aci_hqr_by_hour'] = preds_inf_aci_hqr_by_hour
# df.loc[df.tail(len(preds_sup_aci_hqr_by_hour)).index, 'preds_sup_aci_hqr_by_hour'] = preds_sup_aci_hqr_by_hour
# df.loc[df.preds_inf_aci_hqr_by_hour == -np.inf, 'preds_inf_aci_hqr_by_hour'] = df[df.index < idx_date_test].preds_inf_hqr_by_hour.min()
# df.loc[df.preds_sup_aci_hqr_by_hour == np.inf, 'preds_sup_aci_hqr_by_hour'] = df[df.index < idx_date_test].preds_sup_hqr_by_hour.max()
# Adaptive Conformal Inference over HQR with varying weights
preds_inf_aci_hqr_w, preds_sup_aci_hqr_w = AdaptiveConformalInference(idx_date_test, df, alpha, gamma=0.02, method='hqr_w', by_hour=False)
df.loc[df.tail(len(preds_inf_aci_hqr_w)).index, 'preds_inf_aci_hqr_w'] = preds_inf_aci_hqr_w
df.loc[df.tail(len(preds_sup_aci_hqr_w)).index, 'preds_sup_aci_hqr_w'] = preds_sup_aci_hqr_w
df.loc[df.preds_inf_aci_hqr_w == -np.inf, 'preds_inf_aci_hqr_w'] = df[df.index < idx_date_test].preds_inf_hqr_w.min()
df.loc[df.preds_sup_aci_hqr_w == np.inf, 'preds_sup_aci_hqr_w'] = df[df.index < idx_date_test].preds_sup_hqr_w.max()
# # Adaptive Conformal Inference over HQR with varying weights by hour
# preds_inf_aci_hqr_w_by_hour, preds_sup_aci_hqr_w_by_hour = AdaptiveConformalInference(idx_date_test, df, alpha, gamma=0.02, method='hqr_w', by_hour=True)
# df.loc[df.tail(len(preds_inf_aci_hqr_w_by_hour)).index, 'preds_inf_aci_hqr_w_by_hour'] = preds_inf_aci_hqr_w_by_hour
# df.loc[df.tail(len(preds_sup_aci_hqr_w_by_hour)).index, 'preds_sup_aci_hqr_w_by_hour'] = preds_sup_aci_hqr_w_by_hour
# df.loc[df.preds_inf_aci_hqr_w_by_hour == -np.inf, 'preds_inf_aci_hqr_w_by_hour'] = df[df.index < idx_date_test].preds_inf_hqr_w_by_hour.min()
# df.loc[df.preds_sup_aci_hqr_w_by_hour == np.inf, 'preds_sup_aci_hqr_w_by_hour'] = df[df.index < idx_date_test].preds_sup_hqr_w_by_hour.max()
# Adaptive Conformal Inference over QRA
preds_inf_aci_qra, preds_sup_aci_qra = AdaptiveConformalInference(idx_date_test, df, alpha, gamma=0.02, method='qra', by_hour=False)
df.loc[df.tail(len(preds_inf_aci_qra)).index, 'preds_inf_aci_qra'] = preds_inf_aci_qra
df.loc[df.tail(len(preds_sup_aci_qra)).index, 'preds_sup_aci_qra'] = preds_sup_aci_qra
df.loc[df.preds_inf_aci_qra == -np.inf, 'preds_inf_aci_qra'] = df[df.index < idx_date_test].preds_inf_qra.min()
df.loc[df.preds_sup_aci_qra == np.inf, 'preds_sup_aci_qra'] = df[df.index < idx_date_test].preds_sup_qra.max()
# # Adaptive Conformal Inference over QRA by hour
# preds_inf_aci_qra_by_hour, preds_sup_aci_qra_by_hour = AdaptiveConformalInference(idx_date_test, df, alpha, gamma=0.02, method='qra', by_hour=True)
# df.loc[df.tail(len(preds_inf_aci_qra_by_hour)).index, 'preds_inf_aci_qra_by_hour'] = preds_inf_aci_qra_by_hour
# df.loc[df.tail(len(preds_sup_aci_qra_by_hour)).index, 'preds_sup_aci_qra_by_hour'] = preds_sup_aci_qra_by_hour
# df.loc[df.preds_inf_aci_qra_by_hour == -np.inf, 'preds_inf_aci_qra_by_hour'] = df[df.index < idx_date_test].preds_inf_qra_by_hour.min()
# df.loc[df.preds_sup_aci_qra_by_hour == np.inf, 'preds_sup_aci_qra_by_hour'] = df[df.index < idx_date_test].preds_sup_qra_by_hour.max()
# Width Adaptive Conformal Inference over HQR
preds_inf_waci_hqr, preds_sup_waci_hqr = WidthAdaptiveConformalInference(idx_date_test, df, alpha, gamma = 0.02, sigma=3, method='hqr', by_hour=False, save_alphas=True)
df.loc[df.tail(len(preds_inf_waci_hqr)).index, 'preds_inf_waci_hqr'] = preds_inf_waci_hqr
df.loc[df.tail(len(preds_sup_waci_hqr)).index, 'preds_sup_waci_hqr'] = preds_sup_waci_hqr
df.loc[df.preds_inf_waci_hqr == -np.inf, 'preds_inf_waci_hqr'] = df[df.index < idx_date_test].preds_inf_hqr.min()
df.loc[df.preds_sup_waci_hqr == np.inf, 'preds_sup_waci_hqr'] = df[df.index < idx_date_test].preds_sup_hqr.max()
# # Width Adaptive Conformal Inference over HQR by hour
# preds_inf_waci_hqr_by_hour, preds_sup_waci_hqr_by_hour = WidthAdaptiveConformalInference(idx_date_test, df, alpha, gamma = 0.02, sigma=3, method='hqr', by_hour=True)
# df.loc[df.tail(len(preds_inf_waci_hqr_by_hour)).index, 'preds_inf_waci_hqr_by_hour'] = preds_inf_waci_hqr_by_hour
# df.loc[df.tail(len(preds_sup_waci_hqr_by_hour)).index, 'preds_sup_waci_hqr_by_hour'] = preds_sup_waci_hqr_by_hour
# df.loc[df.preds_inf_waci_hqr_by_hour == -np.inf, 'preds_inf_waci_hqr_by_hour'] = df[df.index < idx_date_test].preds_inf_hqr_by_hour.min()
# df.loc[df.preds_sup_waci_hqr_by_hour == np.inf, 'preds_sup_waci_hqr_by_hour'] = df[df.index < idx_date_test].preds_sup_hqr_by_hour.max()
# Width Adaptive Conformal Inference over HQR with varying weights
preds_inf_waci_hqr_w, preds_sup_waci_hqr_w = WidthAdaptiveConformalInference(idx_date_test, df, alpha, gamma = 0.02, sigma=3, method='hqr_w', by_hour=False)
df.loc[df.tail(len(preds_inf_waci_hqr_w)).index, 'preds_inf_waci_hqr_w'] = preds_inf_waci_hqr_w
df.loc[df.tail(len(preds_sup_waci_hqr_w)).index, 'preds_sup_waci_hqr_w'] = preds_sup_waci_hqr_w
df.loc[df.preds_inf_waci_hqr_w == -np.inf, 'preds_inf_waci_hqr_w'] = df[df.index < idx_date_test].preds_inf_hqr_w.min()
df.loc[df.preds_sup_waci_hqr_w == np.inf, 'preds_sup_waci_hqr_w'] = df[df.index < idx_date_test].preds_sup_hqr_w.max()
# # Width Adaptive Conformal Inference over HQR with varying weights by hour
# preds_inf_waci_hqr_w_by_hour, preds_sup_waci_hqr_w_by_hour = WidthAdaptiveConformalInference(idx_date_test, df, alpha, gamma = 0.02, sigma=3, method='hqr_w', by_hour=True)
# df.loc[df.tail(len(preds_inf_waci_hqr_w_by_hour)).index, 'preds_inf_waci_hqr_w_by_hour'] = preds_inf_waci_hqr_w_by_hour
# df.loc[df.tail(len(preds_sup_waci_hqr_w_by_hour)).index, 'preds_sup_waci_hqr_w_by_hour'] = preds_sup_waci_hqr_w_by_hour
# df.loc[df.preds_inf_waci_hqr_w_by_hour == -np.inf, 'preds_inf_waci_hqr_w_by_hour'] = df[df.index < idx_date_test].preds_inf_hqr_w_by_hour.min()
# df.loc[df.preds_sup_waci_hqr_w_by_hour == np.inf, 'preds_sup_waci_hqr_w_by_hour'] = df[df.index < idx_date_test].preds_sup_hqr_w_by_hour.max()
# Width Adaptive Conformal Inference over QRA
preds_inf_waci_qra, preds_sup_waci_qra = WidthAdaptiveConformalInference(idx_date_test, df, alpha, gamma = 0.02, sigma=3, method='qra', by_hour=False)
df.loc[df.tail(len(preds_inf_waci_qra)).index, 'preds_inf_waci_qra'] = preds_inf_waci_qra
df.loc[df.tail(len(preds_sup_waci_qra)).index, 'preds_sup_waci_qra'] = preds_sup_waci_qra
df.loc[df.preds_inf_waci_qra == -np.inf, 'preds_inf_waci_qra'] = df[df.index < idx_date_test].preds_inf_qra.min()
df.loc[df.preds_sup_waci_qra == np.inf, 'preds_sup_waci_qra'] = df[df.index < idx_date_test].preds_sup_qra.max()
# # Width Adaptive Conformal Inference over QRA by hour
# preds_inf_waci_qra_by_hour, preds_sup_waci_qra_by_hour = WidthAdaptiveConformalInference(idx_date_test, df, alpha, gamma = 0.02, sigma=3, method='qra', by_hour=True)
# df.loc[df.tail(len(preds_inf_waci_qra_by_hour)).index, 'preds_inf_waci_qra_by_hour'] = preds_inf_waci_qra_by_hour
# df.loc[df.tail(len(preds_sup_waci_qra_by_hour)).index, 'preds_sup_waci_qra_by_hour'] = preds_sup_waci_qra_by_hour
# df.loc[df.preds_inf_waci_qra_by_hour == -np.inf, 'preds_inf_waci_qra_by_hour'] = df[df.index < idx_date_test].preds_inf_qra_by_hour.min()
# df.loc[df.preds_sup_waci_qra_by_hour == np.inf, 'preds_sup_waci_qra_by_hour'] = df[df.index < idx_date_test].preds_sup_qra_by_hour.max()
# # Width Adaptive Conformal Inference 2 over HQR
# preds_inf_waci_2_hqr, preds_sup_waci_2_hqr = WidthAdaptiveConformalInference_2(idx_date_test, df, alpha, gamma=0.02, method='hqr', by_hour=False, save_alphas=True, rate=1.02)
# df.loc[df.tail(len(preds_inf_waci_2_hqr)).index, 'preds_inf_waci_2_hqr'] = preds_inf_waci_2_hqr
# df.loc[df.tail(len(preds_sup_waci_2_hqr)).index, 'preds_sup_waci_2_hqr'] = preds_sup_waci_2_hqr
# df.loc[df.preds_inf_waci_2_hqr == -np.inf, 'preds_inf_waci_2_hqr'] = df[df.index < idx_date_test].preds_inf_hqr.min()
# df.loc[df.preds_sup_waci_2_hqr == np.inf, 'preds_sup_waci_2_hqr'] = df[df.index < idx_date_test].preds_sup_hqr.max()
# # Width Adaptive Conformal Inference 2 over HQR by hour
# preds_inf_waci_2_hqr_by_hour, preds_sup_waci_2_hqr_by_hour = WidthAdaptiveConformalInference_2(idx_date_test, df, alpha, gamma=0.02, method='hqr', by_hour=True, rate=1.02)
# df.loc[df.tail(len(preds_inf_waci_2_hqr_by_hour)).index, 'preds_inf_waci_2_hqr_by_hour'] = preds_inf_waci_2_hqr_by_hour
# df.loc[df.tail(len(preds_sup_waci_2_hqr_by_hour)).index, 'preds_sup_waci_2_hqr_by_hour'] = preds_sup_waci_2_hqr_by_hour
# df.loc[df.preds_inf_waci_2_hqr_by_hour == -np.inf, 'preds_inf_waci_2_hqr_by_hour'] = df[df.index < idx_date_test].preds_inf_hqr_by_hour.min()
# df.loc[df.preds_sup_waci_2_hqr_by_hour == np.inf, 'preds_sup_waci_2_hqr_by_hour'] = df[df.index < idx_date_test].preds_sup_hqr_by_hour.max()
# # Width Adaptive Conformal Inference 2 over HQR with varying weights
# preds_inf_waci_2_hqr_w, preds_sup_waci_2_hqr_w = WidthAdaptiveConformalInference_2(idx_date_test, df, alpha, gamma=0.02, method='hqr_w', by_hour=False, rate=1.02)
# df.loc[df.tail(len(preds_inf_waci_2_hqr_w)).index, 'preds_inf_waci_2_hqr_w'] = preds_inf_waci_2_hqr_w
# df.loc[df.tail(len(preds_sup_waci_2_hqr_w)).index, 'preds_sup_waci_2_hqr_w'] = preds_sup_waci_2_hqr_w
# df.loc[df.preds_inf_waci_2_hqr_w == -np.inf, 'preds_inf_waci_2_hqr_w'] = df[df.index < idx_date_test].preds_inf_hqr_w.min()
# df.loc[df.preds_sup_waci_2_hqr_w == np.inf, 'preds_sup_waci_2_hqr_w'] = df[df.index < idx_date_test].preds_sup_hqr_w.max()
# # Width Adaptive Conformal Inference 2 over HQR with varying weights by hour
# preds_inf_waci_2_hqr_w_by_hour, preds_sup_waci_2_hqr_w_by_hour = WidthAdaptiveConformalInference_2(idx_date_test, df, alpha, gamma=0.02, method='hqr_w', by_hour=True, rate=1.02)
# df.loc[df.tail(len(preds_inf_waci_2_hqr_w_by_hour)).index, 'preds_inf_waci_2_hqr_w_by_hour'] = preds_inf_waci_2_hqr_w_by_hour
# df.loc[df.tail(len(preds_sup_waci_2_hqr_w_by_hour)).index, 'preds_sup_waci_2_hqr_w_by_hour'] = preds_sup_waci_2_hqr_w_by_hour
# df.loc[df.preds_inf_waci_2_hqr_w_by_hour == -np.inf, 'preds_inf_waci_2_hqr_w_by_hour'] = df[df.index < idx_date_test].preds_inf_hqr_w_by_hour.min()
# df.loc[df.preds_sup_waci_2_hqr_w_by_hour == np.inf, 'preds_sup_waci_2_hqr_w_by_hour'] = df[df.index < idx_date_test].preds_sup_hqr_w_by_hour.max()
# # Width Adaptive Conformal Inference 2 over QRA
# preds_inf_waci_2_qra, preds_sup_waci_2_qra = WidthAdaptiveConformalInference_2(idx_date_test, df, alpha, gamma=0.02, method='qra', by_hour=False, rate=1.02)
# df.loc[df.tail(len(preds_inf_waci_2_qra)).index, 'preds_inf_waci_2_qra'] = preds_inf_waci_2_qra
# df.loc[df.tail(len(preds_sup_waci_2_qra)).index, 'preds_sup_waci_2_qra'] = preds_sup_waci_2_qra
# df.loc[df.preds_inf_waci_2_qra == -np.inf, 'preds_inf_waci_2_qra'] = df[df.index < idx_date_test].preds_inf_qra.min()
# df.loc[df.preds_sup_waci_2_qra == np.inf, 'preds_sup_waci_2_qra'] = df[df.index < idx_date_test].preds_sup_qra.max()
# # Width Adaptive Conformal Inference 2 over QRA by hour
# preds_inf_waci_2_qra_by_hour, preds_sup_waci_2_qra_by_hour = WidthAdaptiveConformalInference_2(idx_date_test, df, alpha, gamma=0.02, method='qra', by_hour=True, rate=1.02)
# df.loc[df.tail(len(preds_inf_waci_2_qra_by_hour)).index, 'preds_inf_waci_2_qra_by_hour'] = preds_inf_waci_2_qra_by_hour
# df.loc[df.tail(len(preds_sup_waci_2_qra_by_hour)).index, 'preds_sup_waci_2_qra_by_hour'] = preds_sup_waci_2_qra_by_hour
# df.loc[df.preds_inf_waci_2_qra_by_hour == -np.inf, 'preds_inf_waci_2_qra_by_hour'] = df[df.index < idx_date_test].preds_inf_qra_by_hour.min()
# df.loc[df.preds_sup_waci_2_qra_by_hour == np.inf, 'preds_sup_waci_2_qra_by_hour'] = df[df.index < idx_date_test].preds_sup_qra_by_hour.max()
# Save dataframe
df.to_csv(f"Data//df_final_alpha_{alpha}.csv", index=False)