-
Notifications
You must be signed in to change notification settings - Fork 2
/
comb_stoch_run_cqr.py
202 lines (151 loc) · 10.3 KB
/
comb_stoch_run_cqr.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
import copy
import time
import argparse
import pandas as pd
from QR import * # NN architecture to learn quantiles
from CQR import *
from utils import * # import-export methods
from Dataset import *
from TrainQR_multiquantile import *
# Reminder: we save results in the folder of the first variable in the pair
# for the sake of reproducibility we fix the seeds
torch.manual_seed(0)
np.random.seed(0)
parser = argparse.ArgumentParser()
parser.add_argument("--model_dim", default=2, type=int, help="Dimension of the model")
parser.add_argument("--model_prefix", default="MRH", type=str, help="Prefix of the model name")
parser.add_argument("--n_epochs", default=500, type=int, help="Nb of training epochs for QR")
parser.add_argument("--n_hidden", default=20, type=int, help="Nb of hidden nodes per layer")
parser.add_argument("--batch_size", default=512, type=int, help="Batch size")
parser.add_argument("--lr", default=0.0005, type=float, help="Learning rate")
parser.add_argument("--qr_training_flag", default=True, type=eval, help="training flag")
parser.add_argument("--comb_calibr_flag", default=True, type=eval, help="do combined calibration")
parser.add_argument("--xavier_flag", default=False, type=eval, help="Xavier random weights initialization")
parser.add_argument("--scheduler_flag", default=False, type=eval, help="scheduler flag")
parser.add_argument("--opt", default="Adam", type=str, help="Optimizer")
parser.add_argument("--dropout_rate", default=0.1, type=float, help="Drop-out rate")
parser.add_argument("--alpha", default=0.1, type=float, help="quantiles significance level")
parser.add_argument("--comb_idx", default=0, type=int, help="Identifier of the combination of properteies to monitor")
args = parser.parse_args()
print(args)
model_name = args.model_prefix+str(args.model_dim)
comb_pairs = []
for i in range(args.model_dim):
for j in range(i+1, args.model_dim):
comb_pairs.append((i,j))
print('Combination pairs = ', comb_pairs)
prop_idxs = comb_pairs[args.comb_idx]
print("Model name = ", model_name, "Model dim = ", args.model_dim)
trainset_fn, calibrset_fn, testset_fn, ds_details = import_filenames_w_dim(model_name, args.model_dim)
n_train_states, n_cal_states, n_test_states, cal_hist_size, test_hist_size = ds_details
print("qr_training_flag = ", args.qr_training_flag)
print("comb_calibr_flag = ", args.comb_calibr_flag)
quantiles = np.array([args.alpha/2, 0.5, 1-args.alpha/2]) # LB, MEDIAN, UB
nb_quantiles = len(quantiles)
print(f"Property idxs = {prop_idxs}")
idx_str1 = f'CQR_#{prop_idxs[0]}_Dropout{args.dropout_rate}_multiout_opt=_{args.n_hidden}hidden_{args.n_epochs}epochs_{nb_quantiles}quantiles_3layers_alpha{args.alpha}_lr{args.lr}'
idx_str2 = f'CQR_#{prop_idxs[1]}_Dropout{args.dropout_rate}_multiout_opt=_{args.n_hidden}hidden_{args.n_epochs}epochs_{nb_quantiles}quantiles_3layers_alpha{args.alpha}_lr{args.lr}'
# import data
dataset = Dataset(property_idx=args.comb_idx, comb_flag=True, trainset_fn=trainset_fn, testset_fn=testset_fn,
calibrset_fn=calibrset_fn, alpha=args.alpha, n_train_states=n_train_states, n_cal_states=n_cal_states,
n_test_states=n_test_states, hist_size=cal_hist_size, test_hist_size=test_hist_size)
eqr_width = dataset.load_data()
if args.comb_calibr_flag:
'''
Conjunction of CPI
'''
# Load the pre-trained models specific to each of the two properties (qr1 and qr2)
qr1 = TrainQR(model_name, dataset, idx = idx_str1, cal_hist_size = cal_hist_size, test_hist_size = test_hist_size, quantiles = quantiles, opt = args.opt, n_hidden = args.n_hidden, xavier_flag = args.xavier_flag, scheduler_flag = args.scheduler_flag, drop_out_rate = args.dropout_rate)
qr1.load_model(args.n_epochs)
qr2 = TrainQR(model_name, dataset, idx = idx_str2, cal_hist_size = cal_hist_size, test_hist_size = test_hist_size, quantiles = quantiles, opt = args.opt, n_hidden = args.n_hidden, xavier_flag = args.xavier_flag, scheduler_flag = args.scheduler_flag, drop_out_rate = args.dropout_rate)
qr2.load_model(args.n_epochs)
print(f"--------Property idxs = {prop_idxs}")
# Obtain CQR intervals given the trained QR
cqr = CQR(dataset.X_cal, dataset.R_cal, (qr1.qr_model,qr2.qr_model), test_hist_size = test_hist_size, cal_hist_size = cal_hist_size, comb_flag= True)
cpi_test = cqr.get_cpi(dataset.X_test, pi_flag = False)
cqr.plot_comb_errorbars(dataset.R_test, cpi_test, "predictive intervals", qr1.results_path, extra_info=f'pred_interval_comb{args.comb_idx}_pair={prop_idxs}')
cpi_coverage, cpi_efficiency = cqr.get_coverage_efficiency(dataset.R_test, cpi_test)
print("cpi_coverage = ", cpi_coverage, "cpi_efficiency = ", cpi_efficiency)
cpi_correct, cpi_uncertain, cpi_wrong, cpi_fp = cqr.compute_accuracy_and_uncertainty(cpi_test, dataset.L_test)
print("cpi_correct = ", cpi_correct, "cpi_uncertain = ", cpi_uncertain, "cpi_wrong = ", cpi_wrong, "cpi_fp = ", cpi_fp)
dataset1 = Dataset(property_idx=prop_idxs[0], comb_flag=False, trainset_fn=trainset_fn, testset_fn=testset_fn,
calibrset_fn=calibrset_fn, alpha=args.alpha, n_train_states=n_train_states, n_cal_states=n_cal_states,
n_test_states=n_test_states, hist_size=cal_hist_size, test_hist_size=test_hist_size)
dataset1.load_data()
dataset2 = Dataset(property_idx=prop_idxs[1], comb_flag=False, trainset_fn=trainset_fn, testset_fn=testset_fn,
calibrset_fn=calibrset_fn, alpha=args.alpha, n_train_states=n_train_states, n_cal_states=n_cal_states,
n_test_states=n_test_states, hist_size=cal_hist_size, test_hist_size=test_hist_size)
dataset2.load_data()
print("-------------Statical guarantees of the union")
cqr1 = CQR(dataset1.X_cal, dataset1.R_cal, qr1.qr_model, test_hist_size = test_hist_size, cal_hist_size = cal_hist_size, comb_flag= False)
cpi1 = cqr1.get_cpi(dataset1.X_test)
cqr2 = CQR(dataset2.X_cal, dataset2.R_cal, qr2.qr_model, test_hist_size = test_hist_size, cal_hist_size = cal_hist_size, comb_flag= False)
cpi2 = cqr2.get_cpi(dataset2.X_test)
union_coverage, union_efficiency = cqr.get_coverage_efficiency_coupled(dataset.R_test, cpi1, cpi2)
print("union_cpi_coverage = ", union_coverage, "union_cpi_efficiency = ", union_efficiency)
results_list = ["Id1 = ", idx_str1,"\nId2 = ", idx_str2, "\n", "\n Quantiles = ", str(quantiles),"\n tau = ", str(cqr.tau),
"\n",
"\n cpi_correct = ", str(cpi_correct), "\n cpi_uncertain = ", str(cpi_uncertain), "\n cpi_wrong = ", str(cpi_wrong), "\n cpi_fp = ", str(cpi_fp),
"\n cpi_coverage = ", str(cpi_coverage), "\n cpi_efficiency = ", str(cpi_efficiency),
"\n",
"\n union_cpi_coverage = ", str(union_coverage), "\n union_cpi_efficiency = ", str(union_efficiency)]
save_results_to_file(results_list, qr1.results_path, extra_info=f'_comb{args.comb_idx}_pair={prop_idxs}')
print(qr1.results_path)
d = {model_name:['MIN', 'UNION'],'correct': [cpi_correct, '-'],
'uncertain': [cpi_uncertain, '-'],
'wrong':[cpi_wrong, '-'], 'FP':[cpi_fp, '-'],
'coverage':[cpi_coverage, union_coverage],
'efficiency': [cpi_efficiency,union_efficiency],
}
df = pd.DataFrame(data=d)
print('Table of results:\n ',df)
out_tables_path = f"out/tables/{args.model_prefix}"
os.makedirs(out_tables_path, exist_ok=True)
df.to_csv(out_tables_path+f"/{model_name}_{prop_idxs}_conj_results.csv", index=False)
else: # train the CQR over the combined property
'''
CQR trained over the combined propery
'''
idx_str12 = f'CQR_#{prop_idxs[0]}{prop_idxs[1]}_Dropout{args.dropout_rate}_multiout_opt=_{args.n_hidden}hidden_{args.n_epochs}epochs_{nb_quantiles}quantiles_3layers_alpha{args.alpha}_lr{args.lr}'
qr12 = TrainQR(model_name, dataset, idx = idx_str12, cal_hist_size = cal_hist_size, test_hist_size = test_hist_size, quantiles = quantiles, opt = args.opt, n_hidden = args.n_hidden, xavier_flag = args.xavier_flag, scheduler_flag = args.scheduler_flag, drop_out_rate = args.dropout_rate)
qr12.initialize()
if args.qr_training_flag:
start_time = time.time()
qr12.train(args.n_epochs, args.batch_size, args.lr)
end_time = time.time()-start_time
qr12.save_model()
print(f'Training time for {model_name}-#{prop_idxs} with {args.n_epochs} epochs = {end_time}')
else:
qr12.load_model(args.n_epochs)
# Obtain CQR intervals given the trained QR
cqr12 = CQR(dataset.X_cal, dataset.R_cal, qr12.qr_model, test_hist_size = test_hist_size, cal_hist_size = cal_hist_size)
cpi_test, pi_test = cqr12.get_cpi(dataset.X_test, pi_flag = True)
print("shape: ", cpi_test.shape, pi_test.shape)
pi_coverage, pi_efficiency = cqr12.get_coverage_efficiency(dataset.R_test, pi_test)
print("pi_coverage = ", pi_coverage, "pi_efficiency = ", pi_efficiency)
pi_correct, pi_uncertain, pi_wrong, pi_fp = cqr12.compute_accuracy_and_uncertainty(pi_test, dataset.L_test)
print("pi_correct = ", pi_correct, "pi_uncertain = ", pi_uncertain, "pi_wrong = ", pi_wrong, "pi_fp = ", pi_fp)
cpi_coverage, cpi_efficiency = cqr12.get_coverage_efficiency(dataset.R_test, cpi_test)
print("cpi_coverage = ", cpi_coverage, "cpi_efficiency = ", cpi_efficiency)
cpi_correct, cpi_uncertain, cpi_wrong, cpi_fp = cqr12.compute_accuracy_and_uncertainty(cpi_test, dataset.L_test)
print("cpi_correct = ", cpi_correct, "cpi_uncertain = ", cpi_uncertain, "cpi_wrong = ", cpi_wrong, "cpi_fp = ", cpi_fp)
cqr12.plot_errorbars(dataset.R_test, pi_test, cpi_test, "predictive intervals", qr12.results_path, 'pred_interval')
results_list = ["Id = ", idx_str12, "\n", "\n Quantiles = ", str(quantiles), "\n tau = ", str(cqr12.tau), "\n",
"\n pi_correct = ", str(pi_correct), "\n pi_uncertain = ", str(pi_uncertain), "\n pi_wrong = ", str(pi_wrong),"\n pi_fp = ", str(pi_fp),"\n pi_coverage = ", str(pi_coverage), "\n pi_efficiency = ", str(pi_efficiency),
"\n",
"\n eqr_width = ", str(eqr_width),
"\n",
"\n cpi_correct = ", str(cpi_correct), "\n cpi_uncertain = ", str(cpi_uncertain), "\n cpi_wrong = ", str(cpi_wrong),"\n cpi_fp = ", str(cpi_fp),"\n cpi_coverage = ", str(cpi_coverage), "\n cpi_efficiency = ", str(cpi_efficiency)]
save_results_to_file(results_list, qr12.results_path)
print(qr12.results_path)
d = {model_name:['QR', 'CQR'],'correct': [pi_correct, cpi_correct],
'uncertain': [pi_uncertain, cpi_uncertain],
'wrong':[pi_wrong,cpi_wrong], 'FP':[pi_fp,pi_fp],
'coverage':[pi_coverage, pi_coverage],
'efficiency': [pi_efficiency, cpi_efficiency],
'EQR width': [eqr_width, '-']}
df = pd.DataFrame(data=d)
print('Table of results:\n ',df)
out_tables_path = f"out/tables/{args.model_prefix}"
os.makedirs(out_tables_path, exist_ok=True)
df.to_csv(out_tables_path+f"/{model_name}_{prop_idxs}_results.csv", index=False)