ResMPS/OtherNN.py at master · YemingMeng/ResMPS · 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
import torch as th
import torch.nn as nn
from Tools import trid


class EM(th.nn.Module):

    def __init__(self, D: int, n_res: int, n_class: int,
                 relu=False,
                 dropout_prob=0,
                 device=None,
                 order=None,
                 std=1e-3):
        super( ).__init__( )
        self.D, self.n_res, self.n_class = D, n_res, n_class
        self.relu  = relu
        if order is None:
            order = 1
        self.order = order
        if device is None:
            device = 'cuda' if th.cuda.is_available( ) else 'cpu'
        self.device      = device
        self.lrten       = nn.Parameter(
            th.randn(n_res, D, D, device=device) * std)
        self.vir_feature = th.ones(1, D, device=device) / D
        self.bias        = nn.Parameter(
            th.zeros(n_res, D, device=device))
        self.fc      = nn.Linear(D, n_class).to(device)
        if dropout_prob > 0:
            self.dropout = th.nn.Dropout(p=dropout_prob)
        else:
            self.dropout = None
        if order == 2:
            self.M  = trid(n_res, dim=2, device=device)
        else:
            self.M1 = trid(n_res, dim=2, device=device)
            self.M2 = trid(n_res, dim=3, device=device)

    def forward(self, data: th.tensor):
        vf   = th.squeeze(self.vir_feature)
        data = data.view(*data.shape[:-2], -1).squeeze(1)
        if self.order == 1:
            if not self.relu:
                temp = th.einsum('nlr, bn -> blr', self.lrten, data)
                temp = th.einsum('l, blr -> br', vf, temp)
            else:
                temp = th.einsum('nlr, bn -> bnlr', self.lrten, data)
                temp = th.einsum('l, bnlr -> bnr', vf, temp)
                temp = temp + self.bias
                temp = nn.functional.relu(temp)
                if self.dropout:
                    temp = self.dropout(temp)
                temp = th.sum(temp, axis=1)
        if self.order == 2:
            temp1 = th.einsum('nlr, bn -> blr', self.lrten, data)
            temp1 = th.einsum('l, blr -> br', vf, temp1)

            W    = th.einsum('nlr, bn -> bnlr', self.lrten, data)
            lW   = th.einsum('l, bnlr -> bnr', vf, W)
            lWM  = th.einsum('bnr, nm -> bmr', lW, self.M)
            temp2 = th.einsum('bmr, bmro -> bo', lWM, W)

            temp = temp1 + temp2
        if self.order == 3:
            temp1 = th.einsum('nlr, bn -> blr', self.lrten, data)
            temp1 = th.einsum('l, blr -> br', vf, temp1)

            W     = th.einsum('nlr, bn -> bnlr', self.lrten, data)
            lW    = th.einsum('l, bnlr -> bnr', vf, W)
            lWM   = th.einsum('bnr, nm -> bmr', lW, self.M1)
            temp2 = th.einsum('bmr, bmro -> bo', lWM, W)

            lW0W1 = th.einsum('bix, bjxy -> bijy', lW, W)
            lW0W1 = th.einsum('bijy, ijk -> bky', lW0W1, self.M2)
            temp3 = th.einsum('bky, bkyo -> bo', lW0W1, W)

            temp = temp1 + temp2 + temp3
        vf   = temp + vf
        rst  = self.fc(vf)
        return rst

    def project_grad(self):
        pass

    def project_tensor(self):
        pass


class FashionCNN(nn.Module):

    def __init__(self):
        super(FashionCNN, self).__init__( )

        self.layer1 = nn.Sequential(
            nn.Conv2d(in_channels=1, out_channels=32,
                      kernel_size=3, padding=1),
            nn.BatchNorm2d(32),
            nn.ReLU( ),
            nn.MaxPool2d(kernel_size=2, stride=2)
        )

        self.layer2 = nn.Sequential(
            nn.Conv2d(in_channels=32, out_channels=64, kernel_size=3),
            nn.BatchNorm2d(64),
            nn.ReLU( ),
            nn.MaxPool2d(2)
        )

        self.fc1 = nn.Linear(in_features=64*6*6, out_features=600)
        self.drop = nn.Dropout2d(0.25)
        self.fc2 = nn.Linear(in_features=600, out_features=120)
        self.fc3 = nn.Linear(in_features=120, out_features=10)

    def forward(self, x):
        out = self.layer1(x)
        out = self.layer2(out)
        out = out.view(out.size(0), -1)
        out = self.fc1(out)
        out = self.drop(out)
        out = self.fc2(out)
        out = self.fc3(out)

        return out


class FullLinear(th.nn.Module):

    def __init__(self, n_res: int, n_class: int,
                 device=None, depth=1):
        super( ).__init__( )
        self.n_res, self.n_class = n_res, n_class
        if device is None:
            device = 'cuda' if th.cuda.is_available( ) else 'cpu'
        self.device = device
        self.depth = depth
        if depth == 1:
            self.Linear = th.nn.Linear(n_res, n_class).to(device)
        elif depth == 2:
            self.Linear1 = th.nn.Linear(n_res, n_res).to(device)
            self.Linear2 = th.nn.Linear(n_res, n_class).to(device)
        elif depth == 3:
            self.Linear1 = th.nn.Linear(n_res, n_res).to(device)
            self.Linear2 = th.nn.Linear(n_res, n_res).to(device)
            self.Linear3 = th.nn.Linear(n_res, n_class).to(device)

    def forward(self, data: th.tensor):
        data = data.view(*data.shape[:-2], -1).squeeze(1)
        if self.depth == 1:
            rst  = self.Linear(data)
        elif self.depth == 2:
            data = self.Linear1(data)
            data = nn.functional.relu(data)
            rst  = self.Linear2(data)
        elif self.depth == 3:
            data = self.Linear1(data)
            data = nn.functional.relu(data)
            data = self.Linear2(data)
            data = nn.functional.relu(data)
            rst  = self.Linear3(data)
        return rst

    def project_grad(self):
        pass

    def project_tensor(self):
        pass


def main( ):
    cf = FullLinear(784, 10)
    data = th.rand(1000, 1, 28, 28, device=cf.device)
    print(cf(data).shape)


if __name__ == "__main__":
    main( )