-
Notifications
You must be signed in to change notification settings - Fork 17
/
AR.py
186 lines (152 loc) · 7.02 KB
/
AR.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
from typing import Dict
import torchinfo
import tqdm
from ...DataFactory import TSData
from ...Exptools import EarlyStoppingTorch
from .. import BaseMethod
import numpy as np
import torch
from torch import nn, optim
from torch.utils.data import DataLoader
from ...DataFactory.TorchDataSet import PredictWindow
class ARLinear(nn.Module):
def __init__(self, p) -> None:
super().__init__()
self.p = p
self.ar = nn.Linear(p, 1)
def forward(self, x):
return self.ar(x)
class AR(BaseMethod):
def __init__(self, params:dict) -> None:
super().__init__()
self.__anomaly_score = None
self.cuda = True
if self.cuda == True and torch.cuda.is_available():
self.device = torch.device("cuda")
print("=== Using CUDA ===")
else:
if self.cuda == True and not torch.cuda.is_available():
print("=== CUDA is unavailable ===")
self.device = torch.device("cpu")
print("=== Using CPU ===")
self.p = params["p"]
self.batch_size = params["batch_size"]
self.model = ARLinear(self.p).to(self.device)
self.epochs = params["epochs"]
learning_rate = params["lr"]
self.optimizer = optim.Adam(self.model.parameters(), lr=learning_rate)
self.scheduler = optim.lr_scheduler.StepLR(self.optimizer, step_size=5, gamma=0.75)
self.loss = nn.MSELoss()
self.save_path = None
self.early_stopping = EarlyStoppingTorch(save_path=self.save_path, patience=3)
def train_valid_phase(self, tsTrain: TSData):
train_loader = DataLoader(
dataset=PredictWindow.UTSOneByOneDataset(tsTrain, "train", window_size=self.p),
batch_size=self.batch_size,
shuffle=True
)
valid_loader = DataLoader(
dataset=PredictWindow.UTSOneByOneDataset(tsTrain, "valid", window_size=self.p),
batch_size=self.batch_size,
shuffle=False
)
for epoch in range(1, self.epochs + 1):
self.model.train(mode=True)
avg_loss = 0
loop = tqdm.tqdm(enumerate(train_loader),total=len(train_loader),leave=True)
for idx, (x, target) in loop:
x, target = x.to(self.device), target.to(self.device)
self.optimizer.zero_grad()
output = self.model(x)
loss = self.loss(output, target)
loss.backward()
self.optimizer.step()
avg_loss += loss.cpu().item()
loop.set_description(f'Training Epoch [{epoch}/{self.epochs}]')
loop.set_postfix(loss=loss.item(), avg_loss=avg_loss/(idx+1))
self.model.eval()
avg_loss = 0
loop = tqdm.tqdm(enumerate(valid_loader),total=len(valid_loader),leave=True)
with torch.no_grad():
for idx, (x, target) in loop:
x, target = x.to(self.device), target.to(self.device)
output = self.model(x)
loss = self.loss(output, target)
avg_loss += loss.cpu().item()
loop.set_description(f'Validation Epoch [{epoch}/{self.epochs}]')
loop.set_postfix(loss=loss.item(), avg_loss=avg_loss/(idx+1))
valid_loss = avg_loss/max(len(valid_loader), 1)
self.scheduler.step()
self.early_stopping(valid_loss, self.model)
if self.early_stopping.early_stop:
print(" Early stopping<<<")
break
def train_valid_phase_all_in_one(self, tsTrains: Dict[str, TSData]):
train_loader = DataLoader(
dataset=PredictWindow.UTSAllInOneDataset(tsTrains, "train", window_size=self.p),
batch_size=self.batch_size,
shuffle=True
)
valid_loader = DataLoader(
dataset=PredictWindow.UTSAllInOneDataset(tsTrains, "valid", window_size=self.p),
batch_size=self.batch_size,
shuffle=False
)
for epoch in range(1, self.epochs + 1):
self.model.train(mode=True)
avg_loss = 0
loop = tqdm.tqdm(enumerate(train_loader),total=len(train_loader),leave=True)
for idx, (x, target) in loop:
x, target = x.to(self.device), target.to(self.device)
self.optimizer.zero_grad()
output = self.model(x)
loss = self.loss(output, target)
loss.backward()
self.optimizer.step()
avg_loss += loss.cpu().item()
loop.set_description(f'Training Epoch [{epoch}/{self.epochs}]')
loop.set_postfix(loss=loss.item(), avg_loss=avg_loss/(idx+1))
self.model.eval()
avg_loss = 0
loop = tqdm.tqdm(enumerate(valid_loader),total=len(valid_loader),leave=True)
with torch.no_grad():
for idx, (x, target) in loop:
x, target = x.to(self.device), target.to(self.device)
output = self.model(x)
loss = self.loss(output, target)
avg_loss += loss.cpu().item()
loop.set_description(f'Validation Epoch [{epoch}/{self.epochs}]')
loop.set_postfix(loss=loss.item(), avg_loss=avg_loss/(idx+1))
valid_loss = avg_loss/max(len(valid_loader), 1)
self.scheduler.step()
self.early_stopping(valid_loss, self.model)
if self.early_stopping.early_stop:
print(" Early stopping<<<")
break
def test_phase(self, tsData: TSData):
test_loader = DataLoader(
dataset=PredictWindow.UTSOneByOneDataset(tsData, "test", window_size=self.p),
batch_size=self.batch_size,
shuffle=False
)
self.model.eval()
scores = []
loop = tqdm.tqdm(enumerate(test_loader),total=len(test_loader),leave=True)
with torch.no_grad():
for idx, (x, target) in loop:
x, target = x.to(self.device), target.to(self.device)
output = self.model(x)
# loss = self.loss(output, target)
mse = torch.sub(output, target).pow(2)
scores.append(mse.cpu())
loop.set_description(f'Testing: ')
scores = torch.cat(scores, dim=0)
scores = scores.numpy().flatten()
assert scores.ndim == 1
self.__anomaly_score = scores
def anomaly_score(self) -> np.ndarray:
return self.__anomaly_score
def param_statistic(self, save_file):
model_stats = torchinfo.summary(self.model, (self.batch_size, self.p), verbose=0)
with open(save_file, 'w') as f:
f.write(str(model_stats))