-
Notifications
You must be signed in to change notification settings - Fork 2
Expand file tree
/
Copy pathmain.py
More file actions
200 lines (154 loc) · 9.36 KB
/
main.py
File metadata and controls
200 lines (154 loc) · 9.36 KB
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
import numpy as np
import time
from load_dataset import load_dataset, print_info, preprocessing
# NVARk
from model.NVARk import NVARk
from model.emb_pars import SSR_parameters
# # different tasks
# from sklearn.svm import SVC
#internal imports
import utils
import tasks
import warnings
warnings.filterwarnings("ignore")
datasets_list = [ ##### ---univ---
# 'SwedishLeaf',
# 'CinCECGTorso',
#### ---multiv---
'JapaneseVowels',
# 'UWaveGestureLibrary'
]
"""global variables"""
# set to 'zero_padding' for matching the longest series in the dataset
# set to 'interpolate'
prepr_option = 'zero_padding' # 'none' / 'zero_padding' / 'interpolate'
experiment='SVM_NVARk' # SVM_NVARk, SVM_NVARk* , time_NVARk
random_iterations = 10
svm_C_list = np.logspace(-3, 3, 7)
solver = 'svd' # 'svd' or 'cholesky' ('cholesky' is used in the paper, is faster but can be unstable for matrices with high collinearity)
def main():
"""################# Data Loading ##########################################"""
for dataset_name in datasets_list:
TRAIN_x_raw, TRAIN_y_raw, TEST_x_raw, TEST_y_raw = load_dataset(dataset_name)
info = print_info(dataset_name, TRAIN_x_raw, TEST_x_raw, y=TRAIN_y_raw)
T_min_init = min(info[dataset_name+' train']['T_min'], info[dataset_name+' test']['T_min'])
if prepr_option=='zero_padding': T_max = max(info[dataset_name+' train']['T_max'], info[dataset_name+' test']['T_max'])
elif prepr_option=='interpolate' : T_max = 25
"""################# Preprocessing #################################"""
TRAIN_x, TRAIN_y, TEST_x, TEST_y = preprocessing(dataset_name, prepr_option,
TRAIN_x_raw, TRAIN_y_raw, TEST_x_raw, TEST_y_raw,
T_new=T_max, info=info)
info = print_info(dataset_name, TRAIN_x, TEST_x, y=TRAIN_y)
print('\n')
# convert datasets of panda series to a list of 2D numpy arrays (shape = [[N], T, D])
TRAIN_x_l = utils.pdSeriesDataFrame_to_listOfnpArray(TRAIN_x)
TEST_x_l = utils.pdSeriesDataFrame_to_listOfnpArray(TEST_x)
"""################# NVAR model ##########################################"""
# get embedding parameters
T_min = min(info[dataset_name+' train']['T_min'], info[dataset_name+' test']['T_min'])
filter_scale = 1/(20*T_min)
k_sqrt, s_sqrt, _, _, _ = SSR_parameters(TRAIN_x_l, T_min,
filter_data = True,
filter_scale = filter_scale,
plot=False)
k = k_sqrt; s = s_sqrt
if k*s >= T_min_init:
while k*s >= T_min_init:
k = k-1; s = s-1
params = {'k':k,
'n':2,
's':s,
'n_dim':75,
'lamb':None,
'gamma_mult':1}
""" individual steps to output K tr-tr """
# model = NVARk(**params, repr_mode='ridge', random_state=1, verbose_lvl=2, solver=solver)
# _ = model.sample_indices(TRAIN_x_l)
# R_nvar = model.compute_embedding(TRAIN_x_l) # list of 2D np arrays
# theta_repr = model.linear_readout(R_nvar) # 2D np arrays
# K = model.rbf_function(theta_repr) # 2D np arrays
""" RUNNING TIME: output matrices in one call and compute running time"""
if experiment=='time_NVARk':
st_time = time.perf_counter()
model = NVARk(**params, repr_mode='ridge', random_state=1, verbose_lvl=2, solver=solver)
K_trtr = model.compute_Ktrtr(TRAIN_x_l)
K_tetr = model.compute_Ktetr(TEST_x_l, TRAIN_x_l)
end_time = time.perf_counter()
print('time = ', round(end_time - st_time,3), 's' )
""" NVARk GENERAL SETTING """
if experiment=='SVM_NVARk':
# mean over more iters
accuracy=[]
for i in range(1,random_iterations+1):
print(f'iteration {i}')
if i==1: verbose_lvl = 2
else: verbose_lvl = 0
model = NVARk(**params, repr_mode='ridge', random_state=i, verbose_lvl=verbose_lvl, readout_type='SVM', solver=solver)
K_trtr = model.compute_Ktrtr(TRAIN_x_l)
K_tetr = model.compute_Ktetr(TEST_x_l, TRAIN_x_l)
acc_test, acc_train, best_C = tasks.my_SVMopt_classifier(K_trtr, TRAIN_y,
K_tetr, TEST_y,
svm_C_list, i, n_folds=10, val_size=0.33,
verbose=False)
accuracy.append(acc_test)
print('accuracy = ', round(np.mean(accuracy),3) , ' +- ', round(np.std(accuracy),3))
""" NVARk* OPTIMIZED SETTING """
if experiment=='SVM_NVARk*':
""" optimize params via CV """
D = TRAIN_x_l[0].shape[1]
if D>1:
if T_min < 60:
k_list = list(set([1,2,3,4,k]))
s_list = list(set([1,2,3,4,s]))
elif T_min >= 60:
k_list = list(set([1,2,3,4,10,20,k]))
s_list = list(set([1,5,20,s]))
else:
if T_min < 400:
k_list = list(set([1,2,3,4,5,k]))
s_list = list(set([1,2,3,4,5,s]))
elif T_min >= 400:
k_list = list(set([1,2,3,4,5,10,20,k]))
s_list = list(set([1,5,10,20,s]))
n_dim_list = [75]
# optimize
model = NVARk(n=2, repr_mode='ridge', random_state=1, verbose_lvl=1, readout_type='SVM', gamma_mult=1, solver=solver)
model.optimize_params(TRAIN_x_l, TRAIN_y,
k_list, s_list, n_dim_list, svm_C_list,
n_folds=10, val_size=0.33, n_jobs=-1, random_state=1,
split='stratified')
accuracy=[]
for i in range(1,random_iterations+1):
print(f'iteration {i}')
model.random_state = i
# evaluate
# mean over more iters
if i==1:model.verbose_lvl = 2
else: model.verbose_lvl = 0
model.fit(TRAIN_x_l, TRAIN_y)
accuracy.append(model.score(TEST_x_l, TEST_y))
print('accuracy = ', round(np.mean(accuracy),3) , ' +- ', round(np.std(accuracy),3))
###### alternative loop in which params are optimized in each iteration with different seed
###### best embedding parameters and best SVM parameters are found for each terms sampling in NVARk
###### should lead to slightly better result
# accuracy=[]
# for i in range(1,random_iterations+1):
# print(f'iteration {i}')
# # optimize
# model = NVARk(n=2, repr_mode='ridge', random_state=i, verbose_lvl=1, readout_type='SVM', gamma_mult=1, solver=solver)
# st_time = time.perf_counter()
# model.optimize_params(TRAIN_x_l, TRAIN_y,
# k_list, s_list, n_dim_list, svm_C_list,
# n_folds=10, val_size=0.33, n_jobs=-1, random_state=i,
# split='stratified')
# opt_time = time.perf_counter()-st_time
# print(f'optimization time = {round(opt_time,3)}')
# # evaluate
# # mean over more iters
# if i==1:model.verbose_lvl = 2
# else: model.verbose_lvl = 0
# model.fit(TRAIN_x_l, TRAIN_y)
# accuracy.append(model.score(TEST_x_l, TEST_y))
# print('accuracy = ', round(np.mean(accuracy),3) , ' +- ', round(np.std(accuracy),3))
if __name__ == "__main__":
main()