-
Notifications
You must be signed in to change notification settings - Fork 6
/
Copy pathconv.py
246 lines (187 loc) · 8.94 KB
/
conv.py
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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
import torch
from torch_geometric.nn import MessagePassing
import torch.nn.functional as F
from torch_geometric.nn import global_mean_pool, global_add_pool
from torch_geometric.nn.inits import reset
from ogb.graphproppred.mol_encoder import AtomEncoder,BondEncoder
from torch_geometric.utils import degree
from torch_scatter import scatter_add, scatter_min, scatter_max, scatter_mean
from torch_geometric.utils import softmax
from torch_geometric.nn.norm import GraphNorm
import math
nn_act = torch.nn.ReLU() #ReLU()
F_act = F.relu
class GINConv(MessagePassing):
def __init__(self, emb_dim):
'''
emb_dim (int): node embedding dimensionality
'''
super(GINConv, self).__init__(aggr = "add")
self.mlp = torch.nn.Sequential(torch.nn.Linear(emb_dim, 2*emb_dim), torch.nn.BatchNorm1d(2*emb_dim), nn_act, torch.nn.Linear(2*emb_dim, emb_dim))
self.eps = torch.nn.Parameter(torch.Tensor([0]))
self.bond_encoder = BondEncoder(emb_dim = emb_dim)
def forward(self, x, edge_index, edge_attr):
edge_embedding = self.bond_encoder(edge_attr)
out = self.mlp((1 + self.eps) *x + self.propagate(edge_index, x=x, edge_attr=edge_embedding))
return out
def message(self, x_j, edge_attr):
return F_act(x_j + edge_attr)
def update(self, aggr_out):
return aggr_out
### GCN convolution along the graph structure
class GCNConv(MessagePassing):
def __init__(self, emb_dim):
super(GCNConv, self).__init__(aggr='add')
self.linear = torch.nn.Linear(emb_dim, emb_dim)
self.root_emb = torch.nn.Embedding(1, emb_dim)
self.bond_encoder = BondEncoder(emb_dim = emb_dim)
def forward(self, x, edge_index, edge_attr):
x = self.linear(x)
edge_embedding = self.bond_encoder(edge_attr)
row, col = edge_index
#edge_weight = torch.ones((edge_index.size(1), ), device=edge_index.device)
deg = degree(row, x.size(0), dtype = x.dtype) + 1
deg_inv_sqrt = deg.pow(-0.5)
deg_inv_sqrt[deg_inv_sqrt == float('inf')] = 0
norm = deg_inv_sqrt[row] * deg_inv_sqrt[col]
return self.propagate(edge_index, x=x, edge_attr = edge_embedding, norm=norm) + F_act(x + self.root_emb.weight) * 1./deg.view(-1,1)
def message(self, x_j, edge_attr, norm):
return norm.view(-1, 1) * F_act(x_j + edge_attr)
def update(self, aggr_out):
return aggr_out
### GNN to generate node embedding
class GNN_node(torch.nn.Module):
"""
Output:
node representations
"""
def __init__(self, num_layer, emb_dim, drop_ratio = 0.5, JK = "last", residual = False, gnn_name = 'gin'):
'''
emb_dim (int): node embedding dimensionality
num_layer (int): number of GNN message passing layers
'''
super(GNN_node, self).__init__()
self.num_layer = num_layer
self.drop_ratio = drop_ratio
self.JK = JK
### add residual connection or not
self.residual = residual
if self.num_layer < 2:
raise ValueError("Number of GNN layers must be greater than 1.")
self.atom_encoder = AtomEncoder(emb_dim)
###List of GNNs
self.convs = torch.nn.ModuleList()
self.batch_norms = torch.nn.ModuleList()
for layer in range(num_layer):
if gnn_name == 'gin':
self.convs.append(GINConv(emb_dim))
elif gnn_name == 'gcn':
self.convs.append(GCNConv(emb_dim))
else:
raise ValueError('Undefined GNN type called {}'.format(gnn_name))
self.batch_norms.append(torch.nn.BatchNorm1d(emb_dim))
# self.batch_norms.append(GraphNorm(emb_dim))
def forward(self, batched_data):
x, edge_index, edge_attr, batch = batched_data.x, batched_data.edge_index, batched_data.edge_attr, batched_data.batch
### computing input node embedding
h_list = [self.atom_encoder(x)]
for layer in range(self.num_layer):
h = self.convs[layer](h_list[layer], edge_index, edge_attr)
h = self.batch_norms[layer](h)
if layer == self.num_layer - 1:
#remove relu for the last layer
h = F.dropout(h, self.drop_ratio, training = self.training)
else:
h = F.dropout(F_act(h), self.drop_ratio, training = self.training)
if self.residual:
h += h_list[layer]
h_list.append(h)
### Different implementations of Jk-concat
if self.JK == "last":
node_representation = h_list[-1]
elif self.JK == "sum":
node_representation = 0
for layer in range(self.num_layer + 1):
node_representation += h_list[layer]
return node_representation
### Virtual GNN to generate node embedding
class GNN_node_Virtualnode(torch.nn.Module):
"""
Output:
node representations
"""
def __init__(self, num_layer, emb_dim, drop_ratio = 0.5, JK = "last", residual = False, gnn_name = 'gin', atom_encode=True):
'''
emb_dim (int): node embedding dimensionality
'''
super(GNN_node_Virtualnode, self).__init__()
self.num_layer = num_layer
self.drop_ratio = drop_ratio
self.JK = JK
### add residual connection or not
self.residual = residual
if self.num_layer < 2:
raise ValueError("Number of GNN layers must be greater than 1.")
self.atom_encode = atom_encode
if self.atom_encode:
self.atom_encoder = AtomEncoder(emb_dim)
### set the initial virtual node embedding to 0.
self.virtualnode_embedding = torch.nn.Embedding(1, emb_dim)
torch.nn.init.constant_(self.virtualnode_embedding.weight.data, 0)
### List of GNNs
self.convs = torch.nn.ModuleList()
### batch norms applied to node embeddings
self.batch_norms = torch.nn.ModuleList()
### List of MLPs to transform virtual node at every layer
self.mlp_virtualnode_list = torch.nn.ModuleList()
for layer in range(num_layer):
if gnn_name == 'gin':
self.convs.append(GINConv(emb_dim))
elif gnn_name == 'gcn':
self.convs.append(GCNConv(emb_dim))
else:
raise ValueError('Undefined GNN type called {}'.format(gnn_name))
# self.batch_norms.append(GraphNorm(emb_dim))
self.batch_norms.append(torch.nn.BatchNorm1d(emb_dim))
for layer in range(num_layer - 1):
self.mlp_virtualnode_list.append(torch.nn.Sequential(torch.nn.Linear(emb_dim, 2*emb_dim), torch.nn.BatchNorm1d(2*emb_dim), nn_act, \
torch.nn.Linear(2*emb_dim, emb_dim), torch.nn.BatchNorm1d(emb_dim), nn_act))
def forward(self, batched_data):
x, edge_index, edge_attr, batch = batched_data.x, batched_data.edge_index, batched_data.edge_attr, batched_data.batch
### virtual node embeddings for graphs
virtualnode_embedding = self.virtualnode_embedding(torch.zeros(batch[-1].item() + 1).to(edge_index.dtype).to(edge_index.device))
if self.atom_encode:
h_list = [self.atom_encoder(x)]
else:
h_list = [x]
for layer in range(self.num_layer):
### add message from virtual nodes to graph nodes
h_list[layer] = h_list[layer] + virtualnode_embedding[batch]
### Message passing among graph nodes
h = self.convs[layer](h_list[layer], edge_index, edge_attr)
h = self.batch_norms[layer](h)
if layer == self.num_layer - 1:
#remove relu for the last layer
h = F.dropout(h, self.drop_ratio, training = self.training)
else:
h = F.dropout(F_act(h), self.drop_ratio, training = self.training)
if self.residual:
h = h + h_list[layer]
h_list.append(h)
### update the virtual nodes
if layer < self.num_layer - 1:
### add message from graph nodes to virtual nodes
virtualnode_embedding_temp = global_add_pool(h_list[layer], batch) + virtualnode_embedding
### transform virtual nodes using MLP
if self.residual:
virtualnode_embedding = virtualnode_embedding + F.dropout(self.mlp_virtualnode_list[layer](virtualnode_embedding_temp), self.drop_ratio, training = self.training)
else:
virtualnode_embedding = F.dropout(self.mlp_virtualnode_list[layer](virtualnode_embedding_temp), self.drop_ratio, training = self.training)
### Different implementations of Jk-concat
if self.JK == "last":
node_representation = h_list[-1]
elif self.JK == "sum":
node_representation = 0
for layer in range(self.num_layer + 1):
node_representation += h_list[layer]
return node_representation