SleepHGNN/layers.py at main · ComputerVisionFans/SleepHGNN · 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
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.init import ones_, kaiming_uniform_
from torch_geometric.nn import MessagePassing
from torch_geometric.utils import softmax


class HGT(MessagePassing):
    def __init__(self, in_dim, out_dim, n_node_types, n_relation_types, n_heads=8, dropout=0.2):
        super().__init__()
        self.in_dim = in_dim
        self.out_dim = out_dim
        self.n_node_types = n_node_types
        self.n_relation_types = n_relation_types
        self.n_heads = n_heads
        self.d_k = out_dim // self.n_heads
        self.sqrt_dk = math.sqrt(self.d_k)
        self.att = None
        self.initializer = kaiming_uniform_

        self.k_linears = nn.ModuleList()
        self.q_linears = nn.ModuleList()
        self.v_linears = nn.ModuleList()
        self.a_linears = nn.ModuleList()
        self.norms = nn.ModuleList()
        self.mappings = nn.ModuleList()

        for _ in range(self.n_node_types):
            self.k_linears.append(nn.Linear(in_dim, out_dim))
            self.q_linears.append(nn.Linear(in_dim, out_dim))
            self.v_linears.append(nn.Linear(in_dim, out_dim))
            self.a_linears.append(nn.Linear(out_dim, out_dim))
            self.mappings.append(nn.Linear(in_dim, out_dim))
            self.norms.append(nn.LayerNorm(out_dim))

        self.relation_pri = nn.Parameter(torch.ones(self.n_relation_types, self.n_heads))
        self.relation_att = nn.Parameter(torch.Tensor(self.n_relation_types, self.n_heads, self.d_k, self.d_k))
        self.relation_msg = nn.Parameter(torch.Tensor(self.n_relation_types, self.n_heads, self.d_k, self.d_k))
        self.skip = nn.Parameter(torch.ones(self.n_node_types))
        self.drop = nn.Dropout(dropout)

        self.reset_parameters()

    def forward(self, x, edge_index, node_type, edge_type):
        return self.propagate(edge_index, node_feature=x, node_type=node_type, edge_type=edge_type)

    def message(self, edge_index_i, node_feature_i, node_feature_j, node_type_i, node_type_j, edge_type):
        data_size = edge_index_i.size(0)

        res_att = torch.zeros(data_size, self.n_heads).to(node_feature_i.device)
        res_msg = torch.zeros(data_size, self.n_heads, self.d_k).to(node_feature_i.device)

        for source_type in range(self.n_node_types):
            sb = (node_type_j == int(source_type))
            k_linear = self.k_linears[source_type]
            v_linear = self.v_linears[source_type]

            for target_type in range(self.n_node_types):
                tb = (node_type_i == int(target_type)) & sb
                q_linear = self.q_linears[target_type]

                for relation_type in range(self.n_relation_types):
                    idx = (edge_type == int(relation_type)) & tb
                    idx = idx.squeeze()
                    # idx = tb
                    if idx.sum() == 0:
                        continue

                    target_node_vec = node_feature_i[idx]
                    source_node_vec = node_feature_j[idx]

                    '''
                        Step 1: Heterogeneous Mutual Attention
                    '''
                    # (n_edge, out_dim) -> (n_edge, n_heads, d_k)
                    # n_edge = len(target_node_vec)
                    q_mat = q_linear(target_node_vec).view(-1, self.n_heads, self.d_k).to(node_feature_i.device)
                    k_mat = k_linear(source_node_vec).view(-1, self.n_heads, self.d_k).to(node_feature_i.device)
                    '''
                        batch matrix-matrix product: (n, a, b) * (n, b, c) = (n, a, c)

                        (n_edge, n_heads, d_k) -> (n_heads, n_edge, d_k)

                        (n_heads, n_edge, d_k) * (n_heads, d_k, d_k)
                        -> (n_heads, n_edge, d_k)
                        -> (n_edge, n_heads, d_k)
                    '''
                    k_mat = torch.bmm(k_mat.transpose(1, 0), self.relation_att[relation_type]).transpose(1, 0)
                    # (n_edge, n_heads) * (n_heads,) = (n_edge, n_heads)
                    res_att[idx] = (q_mat * k_mat).sum(dim=-1) * self.relation_pri[relation_type] / self.sqrt_dk
                    '''
                        Step 2: Heterogeneous Message Passing
                    '''
                    v_mat = v_linear(source_node_vec).view(-1, self.n_heads, self.d_k)
                    # (n_edge, n_heads, d_k)
                    res_msg[idx] = torch.bmm(v_mat.transpose(1, 0), self.relation_msg[relation_type]).transpose(1, 0)
        '''
            Softmax based on target node's id (edge_index_i). Store attention value in self.att for later visualization.
        '''
        self.att = softmax(res_att, edge_index_i)
        '''
            (data_size, n_heads) -> (data_size, n_heads, 1)

            (data_size, n_heads, d_k) * (data_size, n_heads, 1) = (data_size, n_heads, d_k)
        '''
        res = res_msg * self.att.view(-1, self.n_heads, 1)
        del res_att, res_msg
        return res.view(-1, self.out_dim)  # (data_size, n_heads, d_k) -> (data_size, out_dim)

    def update(self, aggr_out, node_feature, node_type):
        '''
            Step 3: Target-specific Aggregation
            x = W[node_type] * gelu(Agg(x)) + x
        '''
        aggr_out = F.gelu(aggr_out)  # (data_size, out_dim)
        res = torch.zeros(aggr_out.size(0), self.out_dim).to(node_feature.device)
        for target_type in range(self.n_node_types):
            idx = (node_type == int(target_type)).squeeze()
            if idx.sum() == 0:
                continue
            trans_out = self.drop(self.a_linears[target_type](aggr_out[idx]))
            alpha = torch.sigmoid(self.skip[target_type])
            '''
                Add skip connection with learnable weight self.skip[t_id]
            '''
            res[idx] = self.norms[target_type](trans_out * alpha + node_feature[idx] * (1 - alpha))

        return res

    def reset_parameters(self):
        ones_(self.relation_pri)
        ones_(self.skip)
        self.initializer(self.relation_att)
        self.initializer(self.relation_msg)
        for k, q, v, a, norm, mapping in zip(self.k_linears, self.q_linears, self.v_linears, self.a_linears, self.norms,
                                             self.mappings):
            k.reset_parameters()
            q.reset_parameters()
            v.reset_parameters()
            a.reset_parameters()
            norm.reset_parameters()
            mapping.reset_parameters()