AuthECG/Authentication.py at master · PriceWang/AuthECG · GitHub

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
import argparse
import keras
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib import ticker
from tqdm import tqdm

parser = argparse.ArgumentParser("Authentication", add_help=False)
parser.add_argument(
    "--model_path",
    default="model.h5",
    type=str,
    help="model path",
)
parser.add_argument(
    "--data_path",
    default="PTB_dataset.csv",
    type=str,
    help="dataset path",
)
args = parser.parse_args()


def weightTransform(W, mode=1, n=1):
    if mode == 1:
        W_new = W
    elif mode == 2:
        W_new = np.sign(W)
    else:
        W_new = np.round(W * (np.power(2, n)))
    return W_new


def rebuildModel(model_path):
    model = keras.models.load_model(model_path)
    count = 0
    layer_outputs = []
    for layer in model.layers:
        if layer.name.startswith("dropout"):
            break
        if layer.name.startswith("conv1d"):
            weights = layer.get_weights()
            for i in range(len(weights)):
                w = weightTransform(weights[i])
                count = count + np.size(w)
                weights[i] = w
            layer.set_weights(weights)
        output = layer.output
        layer_outputs.append(output)
    print("Original Weight\nMultiplication Number: ", count)
    model_template = keras.Model(inputs=model.input, outputs=layer_outputs)
    return model_template


def rebuildModelBNN(model_path):
    model = keras.models.load_model(model_path)
    count_inv = 0
    layer_outputs = []
    for layer in model.layers:
        if layer.name.startswith("dropout"):
            break
        if layer.name.startswith("conv1d"):
            weights = layer.get_weights()
            for i in range(len(weights)):
                w = weightTransform(weights[i], 2)
                count_inv = count_inv + np.sum(w < 0)
                weights[i] = w
            layer.set_weights(weights)
        output = layer.output
        layer_outputs.append(output)
    print("Binary Weight\nInversion Number: ", count_inv)
    model_template = keras.Model(inputs=model.input, outputs=layer_outputs)
    return model_template


def rebuildModelENN(model_path, n):
    model = keras.models.load_model(model_path)
    count = 0
    count_inv = 0
    count_shift = 0
    layer_outputs = []
    for layer in model.layers:
        if layer.name.startswith("dropout"):
            break
        if layer.name.startswith("conv1d"):
            weights = layer.get_weights()
            for i in range(len(weights)):
                w = weightTransform(weights[i], 3, n)
                if i == 0:
                    count = count + np.size(w)
                    count_shift = count_shift + np.size(w)
                    count_inv = count_inv + np.sum(w < 0)
                if i == 1:
                    count = count + np.size(w)
                weights[i] = w / (np.power(2, n))
            layer.set_weights(weights)
        output = layer.output
        layer_outputs.append(output)
    print(
        "Exponent Weight\nAddition Number: ",
        count,
        "Inversion Number: ",
        count_inv,
        "Bit-shift Number: ",
        count_shift,
    )
    model_template = keras.Model(inputs=model.input, outputs=layer_outputs)
    return model_template


def dataProcessing(dataset_path):
    dataset = pd.read_csv(dataset_path)
    patients = pd.unique(dataset["label"])
    users = np.random.choice(patients, int(np.floor(len(patients) / 2)), replace=False)
    test_user = dataset.loc[dataset["label"].isin(users)]
    user_database = test_user.groupby("record").head(1)
    test_user = test_user.sample(n=1000, replace=False)
    test_intruder = (dataset.loc[~dataset["label"].isin(users)]).sample(
        n=1000, replace=False
    )
    return user_database, test_user, test_intruder


def databaseGeneration(model, user_database):
    col = [
        column for column in user_database.columns if column not in ["label", "record"]
    ]
    # template, use the outputs from the last layer
    template = model.predict(user_database[col].values, verbose=0)[-1]
    return template


def authentication(model, database, login, threshold):
    login_data = model.predict(login, verbose=0)[-1]
    for login_part in login_data:
        for database_part in database:
            # if np.linalg.norm(login_part - database_part) < threshold:
            if np.corrcoef(login_part, database_part, rowvar=0)[0][1] > threshold:
                return True
    return False


def login(model, database, test_user, test_intruder, threshold):
    user_number = len(test_user["record"].unique())
    intruder_number = len(test_intruder["record"].unique())
    test_number = user_number + intruder_number
    user_score = 0
    for user in tqdm(test_user.groupby("record"), desc="Verifying Users"):
        login = user[1].drop(columns=["label", "record"]).values
        if authentication(model, database, login, threshold):
            user_score = user_score + 1
    print("User Accuracy: {:.2%}".format(user_score / user_number))
    intruder_score = 0
    for user in tqdm(test_intruder.groupby("record"), desc="Verifying Intruders"):
        login = user[1].drop(columns=["label", "record"]).values
        if not authentication(model, database, login, threshold):
            intruder_score = intruder_score + 1
    print("Intruder Accuracy: {:.2%}".format(intruder_score / intruder_number))
    accuracy = (user_score + intruder_score) / test_number
    print("Average Accuracy : {:.2%}".format(accuracy))
    return accuracy


if __name__ == "__main__":
    model_path = args.model_path
    data_path = args.data_path
    user_database, test_user, test_intruder = dataProcessing(data_path)
    accuracy = []
    model = rebuildModel(model_path)
    database = databaseGeneration(model, user_database)
    accuracy.append(login(model, database, test_user, test_intruder, 0.9))
    model = rebuildModelBNN(model_path)
    database = databaseGeneration(model, user_database)
    accuracy.append(login(model, database, test_user, test_intruder, 0.984))
    model = rebuildModelENN(model_path, 1)
    database = databaseGeneration(model, user_database)
    accuracy.append(login(model, database, test_user, test_intruder, 0.963))
    model = rebuildModelENN(model_path, 2)
    database = databaseGeneration(model, user_database)
    accuracy.append(login(model, database, test_user, test_intruder, 0.935))
    model = rebuildModelENN(model_path, 3)
    database = databaseGeneration(model, user_database)
    accuracy.append(login(model, database, test_user, test_intruder, 0.935))
    net = ["original", "binary", "cut_off", "exponent_n1", "exponent_n2", "exponent_n3"]
    fig, ax = plt.subplots()
    plt.plot(net, accuracy, color="b")
    plt.scatter(net, accuracy, color="r", marker="v")
    plt.ylabel("Accuracy")
    ax.yaxis.set_major_formatter(ticker.PercentFormatter(xmax=1, decimals=0))
    for a, b in zip(net, accuracy):
        plt.text(a, b + 0.001, "{:.2%}".format(b), ha="center", va="bottom", fontsize=9)
    plt.savefig(
        "performance.pdf",
        dpi=1200,
        bbox_inches="tight",
    )