-
Notifications
You must be signed in to change notification settings - Fork 7
/
Copy path'
174 lines (123 loc) · 6.11 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
import sys
import time
import torch
import pickle
import argparse
import preprocess_data
from model.model import Encoder, Decoder, EncDecAD
from pathlib import Path
import torch.nn as nn
def get_precision_recall(score, label, sz, beta=1.0, wsz=1):
# interval, max, ...
maximum = score.max()
th = torch.linspace(0, maximum, sz)
precision = []
recall = []
for i in range(len(th)):
anomaly = (score>th[i]).float()
idx = anomaly*2+label
tn = (idx==0).sum().item() # tn
fn = (idx==1).sum().item() # fn
fp = (idx==2).sum().item() # fp
tp = (idx==3).sum().item() # tp
p = tp/(tp+fp+1e-7)
r = tp/(tp+fn+1e-7)
if p!=0 and r!=0:
precision.append(p)
recall.append(r)
precision = torch.Tensor(precision)
recall = torch.Tensor(recall)
f1 = (1+beta**2)*torch.max((precision*recall).div(beta**2*precision+recall+1e-7))
return precision, recall, f1
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Argument Parser')
parser.add_argument('--data', type=str, default='ecg',
help='type of the dataset (ecg, gesture, power_demand, space_shuttle, respiration, nyc_taxi')
parser.add_argument('--filename', type=str, default='chfdb_chf13_45590.pkl',
help='filename of the dataset')
# parser.add_argument('--filename', type=str, default='xmitdb_x108_0.pkl',
# help='filename of the dataset')
parser.add_argument('--seqlen', type=int, default=16)
parser.add_argument('--epochs', type=int, default=100)
parser.add_argument('--lr', type=float, default=2e-4)
parser.add_argument('--ninp', type=int, default=2)
parser.add_argument('--nhid', type=int, default=64)
parser.add_argument('--clip', type=float, default=0.25)
parser.add_argument('--nlayers', type=int, default=2)
parser.add_argument('--dropout', type=float, default=0.25)
parser.add_argument('--h_dropout', type=float, default=0.25)
parser.add_argument('--log_interval', type=int, default=10)
parser.add_argument('--feedback', action='store_true')
parser.add_argument('--gated', action='store_true')
parser.add_argument('--verbose', action='store_true')
args = parser.parse_args()
device = torch.device('cuda')
# check whether if there is a trained file in saved folder
param_folder_name = 'nlayers:%d'%args.nlayers + '_nhid:%d'%args.nhid + ('_feedback:1' if args.feedback else '_feedback:0') + ('_gated:1' if args.gated else '_gated:0')
save_folder = Path('result', args.data, args.filename, param_folder_name)
if save_folder.joinpath('model_dictionary.pt').is_file() is not True:
print('There is no trained model in ')
print(str(save_folder))
sys.exit()
# if save_folder.joinpath('recall.pkl').is_file() is True:
# print('The precision, and recall were already calculated!')
# print(str(save_folder))
# sys.exit()
def get_batch(source, seqlen, i):
seqlen = min(seqlen, len(source)-i)
input = source[i:i+seqlen]
target_idx = torch.LongTensor(range(input.size(0)-1, -1, -1))
target = input.index_select(0, target_idx)
return input.requires_grad_().to(device), target.requires_grad_().to(device)
def evaluate(model, dataset):
model.eval()
total_loss = 0
start_time = time.time()
hidden = None
for nbatch, i in enumerate(range(0, dataset.size(0), args.seqlen)):
input, target = get_batch(dataset, args.seqlen, i)
output, hidden = model(input, hidden)
loss = criterion(output, target)
total_loss += loss.item()
return total_loss/(nbatch+1)
def get_anomaly_score(model, dataset, mean, cov):
assert(dataset.size(1)==1)
hidden = None
errors = []
outputs = []
for nbatch, i in enumerate(range(0, dataset.size(0), args.seqlen)):
input, target = get_batch(dataset, args.seqlen, i)
output, hidden = model(input, hidden) # input 8 1 2
output_idx = torch.arange(output.size(0)-1, -1, -1).to(device).long()
reverse_output = output.index_select(0, output_idx)
outputs.append(reverse_output)
error = output-target
errors.append(error)
hidden = hidden[0].detach(), hidden[1].detach()
outputs = torch.cat(outputs, 0).squeeze() # x by 2
errors = torch.cat(errors, 0).squeeze() # x by 2
xm = (errors - mean)
score = (xm).mm(cov.inverse())*xm
score = score.sum(1)
return outputs, score
# save
checkpoint = torch.load(str(save_folder.joinpath('model_dictionary.pt')))
mean = checkpoint['mean']
covariance = checkpoint['covariance']
TimeseriesData = preprocess_data.PickleDataLoad(data_type=args.data, filename=args.filename, augment=False)
gen_dataset = TimeseriesData.batchify(TimeseriesData.testData, 1)
gen_label = TimeseriesData.testLabel
encDecAD = EncDecAD(args.ninp, args.nhid, args.ninp, args.nlayers, dropout=args.dropout, h_dropout=args.h_dropout, feedback=args.feedback, gated=args.gated)
encDecAD.to(device)
encDecAD.load_state_dict(checkpoint['state_dict'])
criterion = torch.nn.MSELoss()
out_dataset, gen_score = get_anomaly_score(encDecAD, gen_dataset, mean, covariance)
pickle.dump(gen_dataset, open(str(save_folder.joinpath('gen_dataset.pkl')), 'wb'))
pickle.dump(out_dataset, open(str(save_folder.joinpath('out_dataset.pkl')), 'wb'))
pickle.dump(gen_score, open(str(save_folder.joinpath('scores.pkl')), 'wb'))
pickle.dump(gen_label, open(str(save_folder.joinpath('labels.pkl')), 'wb'))
# Get precision, recall
precision, recall, f1 = get_precision_recall(gen_score.cpu(), gen_label.cpu(), 1000, beta=1.0)
pickle.dump(precision, open(str(save_folder.joinpath('precision.pkl')), 'wb'))
pickle.dump(recall, open(str(save_folder.joinpath('recall.pkl')), 'wb'))
print(str(save_folder), f1)