soft-assertion-fuzzer/run_models.py at main · ISU-PAAL/soft-assertion-fuzzer · 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
import torch
import torch.nn as nn
import numpy as np
import os

current_file_dir = os.path.dirname(os.path.abspath(__file__))


class YourModel(nn.Module):
    def __init__(self, input_size):
        super(YourModel, self).__init__()
        self.layers = nn.Sequential(
            nn.Linear(input_size, 256),
            nn.BatchNorm1d(256),
            nn.ReLU(),
            nn.Dropout(0.3),
            nn.Linear(256, 128),
            nn.BatchNorm1d(128),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(128, 64),
            nn.BatchNorm1d(64),
            nn.ReLU(),
            nn.Linear(64, 32),
            nn.BatchNorm1d(32),
            nn.ReLU(),
            nn.Linear(32, 3),
            nn.Softmax(dim=1)
        )

    def forward(self, x):
        return self.layers(x)


model_cache = {}


def load_model(model_path, input_size):
    if model_path not in model_cache:
        model = YourModel(input_size)
        state_dict = torch.load(model_path, map_location=torch.device('cpu'))
        model.load_state_dict(state_dict, strict=False)
        model.eval()
        model_cache[model_path] = model
    return model_cache[model_path]


def process_input(input_x):
    if isinstance(input_x, torch.Tensor):
        input_x = input_x.detach().numpy()
    input_flat = np.reshape(input_x, (1, -1))
    return torch.from_numpy(input_flat).float()


def predict(model_type, input_x, file_name):
    input_tensor = process_input(input_x)
    model_path = os.path.join(current_file_dir, f'resources/models/{model_type}/{file_name}')
    model = load_model(model_path, input_tensor.shape[1])
    with torch.no_grad():
        predictions = model(input_tensor)
        predicted_indices = torch.argmax(predictions, dim=1)

        index_to_label = {0: -1, 1: 0, 2: 1}
        predicted_labels = index_to_label[predicted_indices.item()]

    return predicted_labels


def get_direction_for_softmax(input_x, model_type):
    return predict(model_type, input_x, 'Softmax_model_v100.pth')


def get_direction_for_log(input_x, model_type):
    return predict(model_type, input_x, 'log_dataset_model_v002.pth')


def get_direction_for_sigmoid(input_x, model_type):
    return predict(model_type, input_x, 'sigmoid_dataset_model_v001.pth')


def get_direction_for_exp(input_x, model_type):
    # return predict(model_type, input_x, 'exp_dataset_model_v100.pth')

    return predict(model_type, input_x, 'exp_dataset_model_v101.pth')

def get_direction_for_tanh(input_x, model_type):
    return predict(model_type, input_x, 'tanh_dataset_model_v001.pth')


def get_direction_for_sqrt(input_x, model_type):
    return predict(model_type, input_x, 'sqrt_dataset_model_v001.pth')


def get_direction_for_rsqrt(input_x, model_type):
    return predict(model_type, input_x, 'rsqrt_dataset_model_v001.pth')


def get_direction_for_relu(input_x, model_type):
    return predict(model_type, input_x, 'relu_model_v100.pth')


def get_direction_for_conv2d(input_x, model_type):
    return predict(model_type, input_x, 'conv2d_dataset_model_v001.pth')


def predict_for_binary_parameter(input_x, input_y, model_type, file_name):
    # Flatten and concatenate input_x and input_y
    if isinstance(input_x, torch.Tensor):
        input_x = input_x.detach().numpy()
    if isinstance(input_y, torch.Tensor):
        input_y = input_y.detach().numpy()

    input_x_flat = np.reshape(input_x, (1, -1))
    input_y_flat = np.reshape(input_y, (1, -1))

    combined_input = np.concatenate((input_x_flat, input_y_flat), axis=1)

    # Convert to PyTorch tensor
    input_tensor = torch.from_numpy(combined_input).float()

    # Load the model
    # model_path = os.path.join(current_file_dir, f'resources/models/{model_type}/matmul_dataset_model_v001.pth')
    model_path = os.path.join(current_file_dir, f'resources/models/{model_type}/{file_name}')
    model = load_model(model_path, input_tensor.shape[1])

    # Make prediction
    with torch.no_grad():
        predictions = model(input_tensor)
        predicted_labels = torch.argmax(predictions, dim=1)

    # Return the prediction adjusted by -1, as in your previous functions
    return predicted_labels.item() - 1


def get_direction_for_matmul(input_x, input_y, model_type):
    return predict_for_binary_parameter(input_x, input_y, model_type, 'matmul_dataset_model_v001.pth')


def get_direction_for_div(input_x, input_y, model_type):
    return predict_for_binary_parameter(input_x, input_y, model_type, 'matmul_div_model_v001.pth')


def get_direction_for_cos_sim(input_x, input_y, model_type):
    return predict_for_binary_parameter(input_x, input_y, model_type, 'cosine_similarity_model_v002.pth')