ggnn/model.py at master · ASzot/ggnn · 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
import torch
import torch.nn as nn

class ListModule(object):
    """
    In PyTorch it does not currently work to have an array of variables in a
    module. These parameters will not be included in the overall module's
    parameter list. To get around this you need manually add them to the
    module. This is then a helper to access the particular sub-module from the
    "list"

    See https://discuss.pytorch.org/t/list-of-nn-module-in-a-nn-module/219/2
    for more disucssion.
    """
    def __init__(self, module, prefix):
        self.module = module
        self.prefix = prefix

    def __getitem__(self, i):
        return getattr(self.module, self.prefix + str(i))


class PropModel(nn.Module):
    def __init__(self, state_dim, n_nodes, n_edge_types):
        super().__init__()

        self.n_nodes = n_nodes
        self.n_edge_types = n_edge_types
        self.reset_gate = nn.Sequential(
                nn.Linear(state_dim * 3, state_dim),
                nn.Sigmoid()
            )

        self.update_gate = nn.Sequential(
                nn.Linear(state_dim * 3, state_dim),
                nn.Sigmoid()
            )

        self.transform = nn.Sequential(
                nn.Linear(state_dim * 3, state_dim),
                nn.Tanh()
            )

    def forward(self, state_in, state_out, state_cur, adj_matrix):
        A_in = adj_matrix[:, :, :self.n_nodes * self.n_edge_types]
        A_out = adj_matrix[:, :, self.n_nodes * self.n_edge_types:]

        # bmm is a Batch matrix multiplication
        a_in = torch.bmm(A_in, state_in)
        a_out = torch.bmm(A_out, state_out)

        # Equation 2 of section 3.2 in the paper
        # Which nodes can pass information to other nodes between edges.
        a = torch.cat((a_in, a_out, state_cur), 2)

        # Equation 3 of section 3.2
        z = self.update_gate(a)

        # Equation 4 of section 3.2
        r = self.reset_gate(a)

        # Equation 5  of section 3.2
        h_hat = self.transform(torch.cat((a_in, a_out, r * state_cur), 2))

        # Equation 6 of section 3.2
        output = (1 - z) * state_cur + z * h_hat

        return output


class GGNN(nn.Module):
    def __init__(self, state_dim, annotation_dim, n_edge_types, n_nodes,
            n_steps):
        super().__init__()

        self.state_dim = state_dim
        self.annotation_dim = annotation_dim
        self.n_edge_types = n_edge_types
        self.n_nodes = n_nodes
        self.n_steps = n_steps

        for i in range(self.n_edge_types):
            in_fc = nn.Linear(self.state_dim, self.state_dim)
            out_fc = nn.Linear(self.state_dim, self.state_dim)

            self.add_module('in_%i' % i, in_fc)
            self.add_module('out_%i' % i, out_fc)

        self.prop_model = PropModel(self.state_dim, self.n_nodes,
                self.n_edge_types)

        self.out = nn.Sequential(
                nn.Linear(self.state_dim + self.annotation_dim,
                    self.state_dim),
                nn.Tanh(),
                nn.Linear(self.state_dim, 1)
            )

        # Initialize nodes
        for m in self.modules():
            if isinstance(m, nn.Linear):
                m.weight.data.normal_(0.0, 0.02)
                m.bias.data.fill_(0)

        self.in_fcs = AttrProxy(self, 'in_')
        self.out_fcs = AttrProxy(self, 'out_')


    def forward(self, init_hidden_state, annotation, adj_matrix):
        hidden_state = init_hidden_state

        for i_step in range(self.n_steps):
            in_states = []
            out_states = []
            for i in range(self.n_edge_types):
                in_fc = self.in_fcs[i]
                out_fc = self.out_fcs[i]

                # Apply equation 8 from section 4 to infer the intermediate
                # annotations.
                in_states.append(in_fc(hidden_state))
                out_states.append(out_fc(hidden_state))

            in_states = torch.stack(in_states).transpose(0, 1).contiguous()
            in_states = in_states.view(-1, self.n_nodes * self.n_edge_types,
                    self.state_dim)

            out_states = torch.stack(out_states).transpose(0, 1).contiguous()
            out_states = out_states.view(-1, self.n_nodes * self.n_edge_types,
                    self.state_dim)

            hidden_state = self.prop_model(in_states, out_states, hidden_state, adj_matrix)

        # Equation 7 of section 3.3
        output = self.out(torch.cat((hidden_state, annotation), 2))
        output = output.sum(2)

        return output