File size: 8,076 Bytes
ad9fbbf | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 | import numpy as np
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.nn.parameter import Parameter
import math
# The following gather functions
def gather_edges(edges, neighbor_idx):
# Features [B,N,N,C] at Neighbor indices [B,N,K] => Neighbor features [B,N,K,C]
neighbors = neighbor_idx.unsqueeze(-1).expand(-1, -1, -1, edges.size(-1))
edge_features = torch.gather(edges, 2, neighbors)
return edge_features
def gather_nodes(nodes, neighbor_idx):
# Features [B,N,C] at Neighbor indices [B,N,K] => [B,N,K,C]
# Flatten and expand indices per batch [B,N,K] => [B,NK] => [B,NK,C]
neighbors_flat = neighbor_idx.view((neighbor_idx.shape[0], -1))
neighbors_flat = neighbors_flat.unsqueeze(-1).expand(-1, -1, nodes.size(2))
# Gather and re-pack
neighbor_features = torch.gather(nodes, 1, neighbors_flat)
neighbor_features = neighbor_features.view(list(neighbor_idx.shape)[:3] + [-1])
return neighbor_features
def cat_neighbors_nodes(h_nodes, h_neighbors, E_idx):
h_nodes = gather_nodes(h_nodes, E_idx)
h_nn = torch.cat([h_neighbors, h_nodes], -1)
return h_nn
class Normalize(nn.Module):
def __init__(self, features, epsilon=1e-6):
super(Normalize, self).__init__()
self.gain = nn.Parameter(torch.ones(features))
self.bias = nn.Parameter(torch.zeros(features))
self.epsilon = epsilon
def forward(self, x, dim=-1):
mu = x.mean(dim, keepdim=True)
sigma = torch.sqrt(x.var(dim, keepdim=True) + self.epsilon)
gain = self.gain
bias = self.bias
# Reshape
if dim != -1:
shape = [1] * len(mu.size())
shape[dim] = self.gain.size()[0]
gain = gain.view(shape)
bias = bias.view(shape)
return gain * (x - mu) / (sigma + self.epsilon) + bias
class TransformerLayer(nn.Module):
def __init__(self, num_hidden, num_in, num_heads=4, dropout=0.2):
super(TransformerLayer, self).__init__()
self.dropout = nn.Dropout(dropout)
self.relu = nn.LeakyReLU()
self.attn_norm = nn.LayerNorm(num_hidden, eps=1e-6)
self.norm = nn.ModuleList([Normalize(num_hidden) for _ in range(3)])
self.ffn_norm = nn.LayerNorm(num_hidden, eps=1e-6)
self.attention = NeighborAttention(num_hidden, num_in, num_heads)
self.dense1 = PositionWiseFeedForward(num_hidden, num_hidden * 4)
self.dense2 = PositionWiseFeedForward(num_hidden, num_hidden * 4)
self.dense3 = PositionWiseFeedForward(num_hidden, num_hidden * 4)
self.weight1 = Parameter(torch.FloatTensor(128 * 2, 128))
self.weight2 = Parameter(torch.FloatTensor(128 * 2, 128))
self.gcn1 = GCN(128)
self.gcn2 = GCN(128)
self.reset_parameters()
def reset_parameters(self):
stdv = 1. / math.sqrt(128)
self.weight1.data.uniform_(-stdv, stdv)
self.weight2.data.uniform_(-stdv, stdv)
def forward(self, h_V, h_E,adj,h0,l,mask_V=None, mask_attend=None): # mask_attend [B, L, K]
# Concatenate h_V_i to h_E_ij
""" Parallel computation of full transformer layer """
# Self-attention
# h_V = self.attn_norm(h_V)
# h_V = self.relu(h_V)
first_dimension = h_V.shape[0]
theta = min(1, math.log(1.5 / l + 1))
dh = self.attention(h_V, h_E, mask_attend)
h_V = self.norm[0](h_V + self.dropout(dh))
h_V = self.relu(h_V)
# # theta = min(1, math.log(1.5 / l + 1))
dh = self.gcn1(h_V, adj, mask_V)
support = torch.cat([dh, h0], -1)
# support = support.view(-1, 64)
r = (1 - 0.7) * dh + 0.7 * h0
weight_expanded = self.weight1.unsqueeze(0).expand(first_dimension, 256, 128)
output = theta * torch.bmm(support, weight_expanded) + (1 - theta) * r
h_V = self.norm[1](h_V + self.dropout(self.relu(output)))
dh = self.dense1(h_V)
h_V = self.norm[2](h_V + self.dropout(dh))
if mask_V is not None:
mask_V = mask_V.unsqueeze(-1)
h_V = mask_V * h_V
return h_V
class PositionWiseFeedForward(nn.Module):
def __init__(self, num_hidden, num_ff):
super(PositionWiseFeedForward, self).__init__()
self.W_in = nn.Linear(num_hidden, num_ff, bias=True)
self.W_out = nn.Linear(num_ff, num_hidden, bias=True)
def forward(self, h_V):
h = F.relu(self.W_in(h_V))
h = self.W_out(h)
return h
class NeighborAttention(nn.Module):
def __init__(self, num_hidden, num_in, num_heads=4):
super(NeighborAttention, self).__init__()
self.num_heads = num_heads
self.num_hidden = num_hidden
# Self-attention layers: {queries, keys, values, output}
self.W_Q = nn.Linear(num_hidden, num_hidden, bias=False)
self.W_K = nn.Linear(num_in, num_hidden, bias=False)
self.W_V = nn.Linear(num_in, num_hidden, bias=False)
self.W_O = nn.Linear(num_hidden, num_hidden, bias=False)
def _masked_softmax(self, attend_logits, mask_attend, dim=-1):
""" Numerically stable masked softmax """
negative_inf = np.finfo(np.float32).min
attend_logits = torch.where(
mask_attend > 0,
attend_logits,
attend_logits.new_tensor(negative_inf),
)
attend = F.softmax(attend_logits, dim)
attend = mask_attend * attend
return attend
def forward(self, h_V, h_E, mask_attend=None):
""" Self-attention, graph-structured O(Nk)
Args:
h_V: Node features [N_batch, N_nodes, N_hidden]
h_E: Neighbor features [N_batch, N_nodes, K, N_hidden]
mask_attend: Mask for attention [N_batch, N_nodes, K] mask_attend [B, L, K]
Returns:
h_V: Node update
"""
# Queries, Keys, Values
n_batch, n_nodes, n_neighbors = h_E.shape[:3]
n_heads = self.num_heads
d = int(self.num_hidden / n_heads)
Q = self.W_Q(h_V).view([n_batch, n_nodes, 1, n_heads, 1, d])
K = self.W_K(h_E).view([n_batch, n_nodes, n_neighbors, n_heads, d, 1])
V = self.W_V(h_E).view([n_batch, n_nodes, n_neighbors, n_heads, d])
# Attention with scaled inner product
attend_logits = torch.matmul(Q, K).view([n_batch, n_nodes, n_neighbors, n_heads]).transpose(-2,-1)
attend_logits = attend_logits / np.sqrt(d)
if mask_attend is not None:
# Masked softmax
mask = mask_attend.unsqueeze(2).expand(-1,-1,n_heads,-1)
attend = self._masked_softmax(attend_logits, mask) # [B, L, heads, K]
else:
attend = F.softmax(attend_logits, -1)
# Attentive reduction
h_V_update = torch.matmul(attend.unsqueeze(-2), V.transpose(2,3)) # [B, L, heads, 1, K] × [B, L, heads, K, d]
h_V_update = h_V_update.view([n_batch, n_nodes, self.num_hidden])
h_V_update = self.W_O(h_V_update)
return h_V_update
class GCN(nn.Module):
def __init__(self, hidden_size):
super(GCN, self).__init__()
self.hidden_size = hidden_size
self.weight = Parameter(torch.FloatTensor(hidden_size, hidden_size))
self.reset_parameters()
def reset_parameters(self):
stdv = 1. / math.sqrt(self.hidden_size)
self.weight.data.uniform_(-stdv, stdv)
def forward(self, x, adj, mask):
# mask shape: (B, N), where 1 indicates valid nodes and 0 indicates padded nodes
B, N, D = x.size()
# Expand weight for batch multiplication
expanded_weight = self.weight.unsqueeze(0).expand(B, -1, -1)
# Apply mask to x and adj
x = x * mask.unsqueeze(-1)
adj = adj * mask.unsqueeze(-1) * mask.unsqueeze(1)
# Perform graph convolution
y = torch.bmm(adj, x)
y = torch.bmm(y, expanded_weight)
# Apply mask to output
y = y * mask.unsqueeze(-1)
return y
|