GNN3.3 GAT (PaperReading&Implementation)

GNN学习笔记

GNN从入门到精通课程笔记

3.3 GAT (ICLR '18)

• Graph Attention Network (ICLR '18)

• Paper Reading
  

  • the importance of node $j$’s features to node $i$: $e_{i,j}=a(W{h}_i,W{h}_j)$
  • masked attention: only compute $e_{ij}$ for nodes $j\in {\mathcal{N}}_i$, where ${\mathcal{N}}_i$ is some neighborhood of node $i$ in the graph.
  • $a$: a single-layer feedforward neural network
  • ${\alpha }_{ij}={\text{Softmax}}_j(e_{ij})$
  • the final output features for every node: $h_i^{{}^{\prime }}=\sigma ({\Sigma }_{j\in {\mathcal{N}}_i}{\alpha }_{ij}W{h}_j)$
  • Final layer: averaging multi-head attention
    • $h_i^{{}^{\prime }}=\sigma (\frac{1}{K}{\Sigma }_{k=1}^K{\Sigma }_{j\in {\mathcal{N}}_i}{\alpha }_{ij}^k{W}^k{h}_j)$

• Implementation

• [1] https://github.com/Diego999/pyGAT.git

class GraphAttentionLayer(nn.Module):
    """
    Simple GAT layer, similar to https://arxiv.org/abs/1710.10903
    """
    def __init__(self, in_features, out_features, dropout, alpha, concat=True):
        super(GraphAttentionLayer, self).__init__()
        self.dropout = dropout
        self.in_features = in_features
        self.out_features = out_features
        self.alpha = alpha
        self.concat = concat

        self.W = nn.Parameter(torch.empty(size=(in_features, out_features)))
        nn.init.xavier_uniform_(self.W.data, gain=1.414)
        self.a = nn.Parameter(torch.empty(size=(2*out_features, 1)))
        nn.init.xavier_uniform_(self.a.data, gain=1.414)

        self.leakyrelu = nn.LeakyReLU(self.alpha)

    def forward(self, h, adj):
        Wh = torch.mm(h, self.W) # h.shape: (N, in_features), Wh.shape: (N, out_features)
        e = self._prepare_attentional_mechanism_input(Wh)

        zero_vec = -9e15*torch.ones_like(e)
        attention = torch.where(adj > 0, e, zero_vec) # mask attention
        attention = F.softmax(attention, dim=1)
        attention = F.dropout(attention, self.dropout, training=self.training)
        h_prime = torch.matmul(attention, Wh)

        if self.concat:
            return F.elu(h_prime)
        else:
            return h_prime

    def _prepare_attentional_mechanism_input(self, Wh):
        N = Wh.size()[0] # number of nodes

        # Below, two matrices are created that contain embeddings in their rows in different orders.
        # (e stands for embedding)
        # These are the rows of the first matrix (Wh_repeated_in_chunks): 
        # e1, e1, ..., e1,            e2, e2, ..., e2,            ..., eN, eN, ..., eN
        # '-------------' -> N times  '-------------' -> N times       '-------------' -> N times
        # 
        # These are the rows of the second matrix (Wh_repeated_alternating): 
        # e1, e2, ..., eN, e1, e2, ..., eN, ..., e1, e2, ..., eN 
        # '----------------------------------------------------' -> N times
        # 
        
        Wh_repeated_in_chunks = Wh.repeat_interleave(N, dim=0)  # 复制
        Wh_repeated_alternating = Wh.repeat(N, 1)
        # Wh_repeated_in_chunks.shape == Wh_repeated_alternating.shape == (N * N, out_features)

        # The all_combination_matrix, created below, will look like this (|| denotes concatenation):
        # e1 || e1
        # e1 || e2
        # e1 || e3
        # ...
        # e1 || eN
        # e2 || e1
        # e2 || e2
        # e2 || e3
        # ...
        # e2 || eN
        # ...
        # eN || e1
        # eN || e2
        # eN || e3
        # ...
        # eN || eN

        all_combinations_matrix = torch.cat([Wh_repeated_in_chunks, Wh_repeated_alternating], dim=1)
        # all_combinations_matrix.shape == (N * N, 2 * out_features)

        return all_combinations_matrix.view(N, N, 2 * self.out_features)
        
class GAT(nn.Module):
    def __init__(self, nfeat, nhid, nclass, dropout, alpha, nheads):
        """Dense version of GAT."""
        super(GAT, self).__init__()
        self.dropout = dropout

        self.attentions = [GraphAttentionLayer(nfeat, nhid, dropout=dropout, alpha=alpha, concat=True) for _ in range(nheads)]
        for i, attention in enumerate(self.attentions):
            self.add_module('attention_{}'.format(i), attention)

        self.out_att = GraphAttentionLayer(nhid * nheads, nclass, dropout=dropout, alpha=alpha, concat=False)

    def forward(self, x, adj):
        x = F.dropout(x, self.dropout, training=self.training)
        x = torch.cat([att(x, adj) for att in self.attentions], dim=1)
        x = F.dropout(x, self.dropout, training=self.training)
        x = F.elu(self.out_att(x, adj))
        return F.log_softmax(x, dim=1)