-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathSTG-Modification.py
191 lines (158 loc) · 7.15 KB
/
STG-Modification.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
"""
======================= START OF LICENSE NOTICE =======================
Copyright (C) 2022 HONGYI001. All Rights Reserved
NO WARRANTY. THE PRODUCT IS PROVIDED BY DEVELOPER "AS IS" AND ANY
EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL DEVELOPER BE LIABLE FOR
ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE
GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR
OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THE PRODUCT, EVEN
IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
======================== END OF LICENSE NOTICE ========================
Primary Author: HONGYI001
"""
import warnings
import pandas as pd
import ScikitWrapper as skw
from Gini import gini_filter
import torch
import numpy as np
from FeaturePoison import insert_feature_noise
import shap
import seaborn as sns
import matplotlib.pyplot as plt
def print_accuracy(f, X, y):
accuracy = 100*np.sum(f(X) == y)/len(y)
print("Accuracy = {0}%".format(accuracy))
return accuracy
num_normal = np.arange(0, 15, 1)
num_overwhelmed = [3]*len(num_normal)
num_shortcut = [1, 2, 3]
num_shortcut = np.repeat(np.array(num_shortcut), 5)
warnings.simplefilter(action='ignore')
X, y = shap.datasets.iris()
no_stg_results = {
'# features': [],
"accuracy": []
}
gini_half_results = {
'# features': [],
"accuracy": []
}
with_stg_results = {
'# features': [],
"accuracy": []
}
two_step_results = {
'# features': [],
"accuracy": []
}
for t in range(5):
accs = []
for i in range(len(num_normal)):
X_train, X_test, Y_train, Y_test = insert_feature_noise(X, y, num_random_noise=num_normal[i],
num_overwhelemed=num_overwhelmed[i], num_shortcut=num_shortcut[i])
X_train, y_train = X_train.to_numpy().astype(np.float32), Y_train.astype(np.int64)
X_test, y_test = X_test.to_numpy().astype(np.float32), Y_test.astype(np.int64)
clf = skw.ScikitClfWrapper(
input_dim=X_train.shape[1],
number_of_classes=3,
hidden_dims=(16, 16), lam=0, epochs=1000, sigma=0.5,
freeze_till=1000, lr=0.1, verbose=False,
device='cpu')
clf.fit(X_train, Y_train)
acc = print_accuracy(clf.predict, X_test, Y_test)
no_stg_results["accuracy"].append(acc)
total_feature = 2*num_normal[i] + num_overwhelmed[i] + num_shortcut[i]
no_stg_results['# features'].append(total_feature)
print(f"{acc}%, {total_feature+4} features")
for t in range(5):
accs = []
for i in range(len(num_normal)):
X_train, X_test, Y_train, Y_test = insert_feature_noise(X, y, num_random_noise=num_normal[i],
num_overwhelemed=num_overwhelmed[i], num_shortcut=num_shortcut[i])
X_train, X_test, Y_train, Y_test = gini_filter(torch.tensor(X_train.values), torch.tensor(X_test.values),
torch.tensor(Y_train, dtype=torch.int64),
torch.tensor(Y_test, dtype=torch.int64),
left=0.5)
X_train, X_test, Y_train, Y_test = X_train.numpy(), X_test.numpy(), Y_train.numpy(), Y_test.numpy()
X_train, Y_train = X_train.astype(np.float32), Y_train.astype(np.int64)
X_test, Y_test = X_test.astype(np.float32), Y_test.astype(np.int64)
"""
clf = FNNClfWrapper(
input_dim=X_train.shape[1],
number_of_classes=3,
hidden_dims=(16, 16), epochs=500, lr=1e-3, verbose=True,
device='cpu', batch_norm=True, dropout=0.5)
"""
clf = skw.ScikitClfWrapper(
input_dim=X_train.shape[1],
number_of_classes=3,
hidden_dims=(16, 16), lam=0, epochs=1000, sigma=1.0,
freeze_till=1000, lr=0.1, verbose=False,
device='cpu')
# print(X_train.shape)
clf.fit(X_train, Y_train)
acc = print_accuracy(clf.predict, X_test, Y_test)
gini_half_results["accuracy"].append(acc)
total_feature = 2*num_normal[i] + num_overwhelmed[i] + num_shortcut[i]
gini_half_results['# features'].append(total_feature)
print(f"{acc}%, {total_feature} features")
for t in range(5):
accs = []
for i in range(len(num_normal)):
X_train, X_test, Y_train, Y_test = insert_feature_noise(X, y, num_random_noise=num_normal[i],
num_overwhelemed=num_overwhelmed[i], num_shortcut=num_shortcut[i])
X_train, y_train = X_train.to_numpy().astype(np.float32), Y_train.astype(np.int64)
X_test, y_test = X_test.to_numpy().astype(np.float32), Y_test.astype(np.int64)
clf = skw.ScikitClfWrapper(
input_dim=X_train.shape[1],
number_of_classes=3,
hidden_dims=(16, 16), lam=0.1, epochs=1000, sigma=0.5,
freeze_till=0, lr=0.1, verbose=False,
device='cpu')
clf.fit(X_train, Y_train)
acc = print_accuracy(clf.predict, X_test, Y_test)
with_stg_results["accuracy"].append(acc)
total_feature = 2*num_normal[i] + num_overwhelmed[i] + num_shortcut[i]
with_stg_results['# features'].append(total_feature)
_, num_gates = clf.model.get_gates()
print(f"{acc}%, {num_gates}/{total_feature+4} features")
for t in range(5):
accs = []
for i in range(len(num_normal)):
X_train, X_test, Y_train, Y_test = insert_feature_noise(X, y, num_random_noise=num_normal[i],
num_overwhelemed=num_overwhelmed[i], num_shortcut=num_shortcut[i])
X_train, X_test, Y_train, Y_test = gini_filter(torch.tensor(X_train.values), torch.tensor(X_test.values),
torch.tensor(Y_train, dtype=torch.int64),
torch.tensor(Y_test, dtype=torch.int64),
left=0.5)
X_train, X_test, Y_train, Y_test = X_train.numpy(), X_test.numpy(), Y_train.numpy(), Y_test.numpy()
X_train, Y_train = X_train.astype(np.float32), Y_train.astype(np.int64)
X_test, Y_test = X_test.astype(np.float32), Y_test.astype(np.int64)
clf = skw.ScikitClfWrapper(
input_dim=X_train.shape[1],
number_of_classes=3,
hidden_dims=(16, 16), lam=0.1, epochs=1000, sigma=0.5,
freeze_till=0, lr=0.1, verbose=False,
device='cpu')
clf.fit(X_train, Y_train)
acc = print_accuracy(clf.predict, X_test, Y_test)
two_step_results["accuracy"].append(acc)
total_feature = 2*num_normal[i] + num_overwhelmed[i] + num_shortcut[i]
two_step_results['# features'].append(total_feature)
_, num_gates = clf.model.get_gates()
print(f"{acc}%, {num_gates}/{total_feature+4} features")
sns.lineplot(data=no_stg_results, x="# features", y="accuracy",
label='No Feature Selection', marker='o')
sns.lineplot(data=gini_half_results, x="# features", y="accuracy",
label='Gini Half', marker='o')
sns.lineplot(data=with_stg_results, x="# features", y="accuracy",
label='STG', marker='o')
sns.lineplot(data=two_step_results, x="# features", y="accuracy",
label='STG + Gini', marker='o')
plt.show()