/
train_sythentic.py
194 lines (154 loc) · 5.84 KB
/
train_sythentic.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
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn.manifold import TSNE
from pyod.models.iforest import IForest
from pyod.models.hbos import HBOS
from pyod.models.pca import PCA
from pyod.models.ocsvm import OCSVM
from pyod.models.lof import LOF
from pyod.models.cblof import CBLOF
from pyod.models.cof import COF
from pyod.models.knn import KNN
from pyod.models.sod import SOD
from pyod.models.ecod import ECOD
from pyod.models.deep_svdd import DeepSVDD
from pyod.models.loda import LODA
from pyod.models.copod import COPOD
from pyod.models.gmm import GMM
from sklearn.preprocessing import MinMaxScaler
import torch
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from numpy import percentile
from torch.utils.data.dataset import TensorDataset
from torch.utils.data.dataloader import DataLoader
from copulas.multivariate import VineCopula
from copulas.univariate import GaussianKDE
class Encoder(nn.Module):
def __init__(self, input_dim, hidden_dim) -> None:
super().__init__()
self.layers = nn.Sequential(
nn.Linear(input_dim, hidden_dim),
nn.GELU(),
nn.Linear(hidden_dim, hidden_dim),
nn.GELU(),
nn.Linear(hidden_dim, hidden_dim),
)
def forward(self, x):
return self.layers(x)
class Model(nn.Module):
def __init__(self) -> None:
super().__init__()
self.build_model()
self.optimizer = optim.Adam(
self.parameters(), lr=1e-3,
)
def build_model(self):
self.encoder = Encoder(
2, 128
)
self.out = nn.Linear(128, 1)
def forward(self, inputs):
hidden = self.encoder(inputs)
return self.out(hidden)
def train_step(self, inputs, outputs):
# score = self.calc_student_rec_loss(batch, flag='train')
logits = self(inputs)
loss = F.mse_loss(logits, outputs)
self.optimizer.zero_grad()
loss.backward()
self.optimizer.step()
return loss
def val_step(self, inputs, outputs):
logits = self(inputs)
loss = F.mse_loss(logits, outputs)
return loss
def decision_function(self, inputs, pseudo_labels):
logits = self(inputs).to('cpu')
pseudo_labels = torch.cat([pseudo_labels, logits], dim=-1)
std = pseudo_labels.std(dim=-1, keepdim=True)
return logits + std
# Define the number of inliers and outliers
n_samples = 320
outliers_fraction = 0.1
xx, yy = np.meshgrid(np.linspace(-7, 7, 100), np.linspace(-7, 7, 100))
n_inliers = int((1. - outliers_fraction) * n_samples)
n_outliers = int(outliers_fraction * n_samples)
ground_truth = np.zeros(n_samples, dtype=int)
ground_truth[-n_outliers:] = 1
random_state = np.random.RandomState(42)
# Data generation
X1 = 0.6 * np.random.randn(n_inliers // 4, 2)
X1[:, 0] = X1[:, 0]-3
X1[:, 1] = X1[:, 1]+3
X2 = 0.6 * np.random.randn(n_inliers // 4, 2)
X3 = 0.6 * np.random.randn(n_inliers // 4, 2)
X3[:, 0] = X3[:, 0]+2.5
X3[:, 1] = X3[:, 1]+3
X4 = 0.6 * np.random.randn(n_inliers // 4, 2)
X4[:, 0] = X4[:, 0]+3.5
X4[:, 1] = X4[:, 1]-3.5
X = np.r_[X1, X2, X3, X4]
# Add outliers
# X (200, 2)
# X = np.r_[X, np.random.uniform(low=-6, high=6, size=(n_outliers, 2))]
# X5 = 0.5 * np.random.randn(n_outliers, 2)
# X5[:, 0] = X5[:, 0]-4.5
# X5[:, 1] = X5[:, 1]-4.5
# X = np.r_[X, X5]
X5 = 0.8 * np.random.randn(n_outliers // 4, 2)
X5[:, 0] = X5[:, 0]-3.5
X5[:, 1] = X5[:, 1]+2.5
X6 = 0.8 * np.random.randn(n_outliers // 4, 2)
X7 = 0.8 * np.random.randn(n_outliers // 4, 2)
X7[:, 0] = X7[:, 0]+3.5
X7[:, 1] = X7[:, 1]+3
X8 = 0.8 * np.random.randn(n_outliers // 4, 2)
X8[:, 0] = X8[:, 0]+3.5
X8[:, 1] = X8[:, 1]-3.5
X = np.r_[X, X5, X6, X7, X8]
pseudo_labels = []
boundary_scores = []
# fit the data and tag outliers
teacher = IForest()
teacher.fit(X)
score = teacher.decision_function(X).reshape(-1,1)
b_score = teacher.decision_function(np.c_[xx.ravel(), yy.ravel()]).reshape(-1,1)
scaler = MinMaxScaler().fit(np.concatenate([score, b_score], axis=0))
score = scaler.transform(score).reshape(-1,1)
b_score = scaler.transform(b_score).reshape(-1,1)
pseudo_labels.append(score)
boundary_scores.append(b_score)
student = Model()
student = student.cuda()
student.train()
for rep in range(40):
dataset = TensorDataset(torch.from_numpy(X).float(), torch.from_numpy(pseudo_labels[-1]).float())
dataloader = DataLoader(dataset, batch_size=128, shuffle=True)
for epoch in range(20):
running_loss = 0.0
for iter, (train_x, train_y) in enumerate(dataloader):
train_x, train_y = train_x.cuda(), train_y.cuda()
loss = student.train_step(train_x, train_y)
running_loss += loss.item()
print(f'Epoch {epoch} loss: {running_loss / len(dataloader)}')
score = student.decision_function(torch.from_numpy(X).float().cuda(), torch.from_numpy(np.concatenate(pseudo_labels, axis=-1)).float())
score = score.data.numpy().reshape(-1,1)
b_score = student.decision_function(torch.from_numpy(np.c_[xx.ravel(), yy.ravel()]).float().cuda(), torch.from_numpy(np.concatenate(boundary_scores, axis=-1)).float())
b_score = b_score.data.numpy().reshape(-1,1)
scaler = MinMaxScaler().fit(np.concatenate([score, b_score], axis=0))
score = scaler.transform(score).reshape(-1,1)
b_score = scaler.transform(b_score).reshape(-1,1)
pseudo_labels.append(score)
boundary_scores.append(b_score)
type_ = 'iforest_local'
np.save(f'./results_sythentic/{type_}/X', X)
np.save(f'./results_sythentic/{type_}/xx', xx)
np.save(f'./results_sythentic/{type_}/yy', yy)
np.save(f'./results_sythentic/{type_}/pseudo_labels', pseudo_labels)
np.save(f'./results_sythentic/{type_}/boundary_scores', boundary_scores)
np.save(f'./results_sythentic/{type_}/n_samples', n_samples)
np.save(f'./results_sythentic/{type_}/outliers_fraction', outliers_fraction)