class RGCN(torch.nn.Module):
def __init__(self, in_channels, hidden_channels, out_channels, num_layers,
dropout, num_nodes_dict, x_types, edge_types):
super(RGCN, self).__init__()
node_types = list(num_nodes_dict.keys())
self.embs = ParameterDict({
key: Parameter(torch.Tensor(num_nodes_dict[key], in_channels))
for key in set(node_types).difference(set(x_types))
})
self.convs = ModuleList()
self.convs.append(
RGCNConv(in_channels, hidden_channels, node_types, edge_types))
for _ in range(num_layers - 2):
self.convs.append(
RGCNConv(hidden_channels, hidden_channels, node_types,
edge_types))
self.convs.append(
RGCNConv(hidden_channels, out_channels, node_types, edge_types))
self.dropout = dropout
self.reset_parameters()
def reset_parameters(self):
for emb in self.embs.values():
torch.nn.init.xavier_uniform_(emb)
for conv in self.convs:
conv.reset_parameters()
def forward(self, x_dict, adj_t_dict):
x_dict = copy.copy(x_dict)
for key, emb in self.embs.items():
x_dict[key] = emb
for conv in self.convs[:-1]:
x_dict = conv(x_dict, adj_t_dict)
for key, x in x_dict.items():
x_dict[key] = F.relu(x)
x_dict[key] = F.dropout(x,
p=self.dropout,
training=self.training)
return self.convs[-1](x_dict, adj_t_dict)
class RGCN(torch.nn.Module):
def __init__(
self,
in_channels,
hidden_channels,
out_channels,
num_layers,
dropout,
num_nodes_dict,
x_types,
num_edge_types,
):
super(RGCN, self).__init__()
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.out_channels = out_channels
self.num_layers = num_layers
self.dropout = dropout
node_types = list(num_nodes_dict.keys())
num_node_types = len(node_types)
self.num_node_types = num_node_types
self.num_edge_types = num_edge_types
# Create embeddings for all node types that do not come with features.
self.emb_dict = ParameterDict({
f"{key}": Parameter(torch.Tensor(num_nodes_dict[key], in_channels))
for key in set(node_types).difference(set(x_types))
})
I, H, O = in_channels, hidden_channels, out_channels # noqa
# Create `num_layers` many message passing layers.
self.convs = ModuleList()
self.convs.append(RGCNConv(I, H, num_node_types, num_edge_types))
for _ in range(num_layers - 2):
self.convs.append(RGCNConv(H, H, num_node_types, num_edge_types))
self.convs.append(RGCNConv(H, O, self.num_node_types, num_edge_types))
self.reset_parameters()
def reset_parameters(self):
for emb in self.emb_dict.values():
torch.nn.init.xavier_uniform_(emb)
for conv in self.convs:
conv.reset_parameters()
def group_input(self, x_dict, node_type, local_node_idx):
# Create global node feature matrix.
h = torch.zeros((node_type.size(0), self.in_channels),
device=node_type.device)
for key, x in x_dict.items():
mask = node_type == key
h[mask] = x[local_node_idx[mask]]
for key, emb in self.emb_dict.items():
mask = node_type == int(key)
h[mask] = emb[local_node_idx[mask]]
return h
def forward(self, x_dict, edge_index, edge_type, node_type,
local_node_idx):
x = self.group_input(x_dict, node_type, local_node_idx)
for i, conv in enumerate(self.convs):
x = conv(x, edge_index, edge_type, node_type)
if i != self.num_layers - 1:
x = F.relu(x)
x = F.dropout(x, p=0.5, training=self.training)
return x.log_softmax(dim=-1)
def inference(self, x_dict, edge_index_dict, key2int):
# We can perform full-batch inference on GPU.
device = list(x_dict.values())[0].device
x_dict = copy(x_dict)
for key, emb in self.emb_dict.items():
x_dict[int(key)] = emb
adj_t_dict = {}
for key, (row, col) in edge_index_dict.items():
adj_t_dict[key] = SparseTensor(row=col, col=row).to(device)
for i, conv in enumerate(self.convs):
out_dict = {}
for j, x in x_dict.items():
out_dict[j] = conv.root_lins[j](x)
for keys, adj_t in adj_t_dict.items():
src_key, target_key = keys[0], keys[-1]
out = out_dict[key2int[target_key]]
tmp = adj_t.matmul(x_dict[key2int[src_key]], reduce="mean")
out.add_(conv.rel_lins[key2int[keys]](tmp))
if i != self.num_layers - 1:
for j in range(self.num_node_types):
F.relu_(out_dict[j])
x_dict = out_dict
return x_dict
class RGNN(torch.nn.Module):
def __init__(self, in_channels, hidden_channels, out_channels, num_layers,
dropout, num_nodes_dict, x_types, num_edge_types, args):
super(RGNN, self).__init__()
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.out_channels = out_channels
self.num_layers = num_layers
self.dropout = dropout
self.args = args
node_types = list(num_nodes_dict.keys())
num_node_types = len(node_types)
self.num_node_types = num_node_types
self.num_edge_types = num_edge_types
# Create embeddings for all node types that do not come with features.
self.emb_dict = ParameterDict({
f'{key}': Parameter(torch.Tensor(num_nodes_dict[key], in_channels))
for key in set(node_types).difference(set(x_types))
})
I, H, O = in_channels, hidden_channels, out_channels # noqa
if self.args.conv_name == 'rgcn':
self.convs = ModuleList()
self.convs.append(
RGCNConv(I, H, num_node_types, num_edge_types, self.args))
for _ in range(num_layers - 2):
self.convs.append(
RGCNConv(H, H, num_node_types, num_edge_types, self.args))
self.convs.append(
RGCNConv(H, O, self.num_node_types, num_edge_types, self.args))
else:
self.convs = ModuleList()
self.convs.append(
RGSNConv(I, H, num_node_types, num_edge_types, self.args))
for _ in range(num_layers - 2):
self.convs.append(
RGSNConv(H, H, num_node_types, num_edge_types, self.args))
self.convs.append(
RGSNConv(H, O, self.num_node_types, num_edge_types, self.args))
if self.args.Norm4:
self.norm = torch.nn.LayerNorm(I)
self.reset_parameters()
def reset_parameters(self):
root = self.args.feat_dir
Feat_list = [
os.path.join(root, i) for i in [
'./author_FEAT.npy', './field_of_study_FEAT.npy',
'./institution_FEAT.npy'
]
]
for emb, Feat_path in zip(self.emb_dict.values(), Feat_list):
if self.args.FDFT:
emb.data = torch.Tensor(np.load(Feat_path)).to(
self.args.device)
emb.requires_grad = True
else:
torch.nn.init.xavier_uniform_(emb)
for conv in self.convs:
conv.reset_parameters()
if self.args.Norm4:
self.norm.reset_parameters()
def group_input(self,
x_dict,
node_type,
local_node_idx,
n_id=None,
perturb=None):
# Create global node feature matrix.
if n_id is not None:
node_type = node_type[n_id]
local_node_idx = local_node_idx[n_id]
h = torch.zeros((node_type.size(0), self.in_channels),
device=node_type.device)
for key, x in x_dict.items():
mask = node_type == int(key)
h[mask] = x[local_node_idx[mask]] if perturb is None else x[
local_node_idx[mask]] + perturb
for key, emb in self.emb_dict.items():
mask = node_type == int(key)
h[mask] = emb[local_node_idx[mask]]
return h
def forward(self,
n_id,
x_dict,
adjs,
edge_type,
node_type,
local_node_idx,
perturb=None):
x = self.group_input(x_dict, node_type, local_node_idx, n_id, perturb)
if self.args.FDFT:
x = F.dropout(x, p=0.5, training=self.training)
if self.args.Norm4:
x = self.norm(x)
node_type = node_type[n_id]
for i, (edge_index, e_id, size) in enumerate(adjs):
x_target = x[:size[1]] # Target node embeddings.
src_node_type = node_type
node_type = node_type[:size[1]] # Target node types.
conv = self.convs[i]
x = conv((x, x_target), edge_index, edge_type[e_id], node_type,
src_node_type)
if i != self.num_layers - 1:
x = F.relu(x)
x = F.dropout(x, p=0.5, training=self.training)
return x.log_softmax(dim=-1)
class RGCN(torch.nn.Module):
def __init__(self, in_channels, hidden_channels, out_channels, num_layers,
dropout, num_nodes_dict, x_types, num_edge_types):
super(RGCN, self).__init__()
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.out_channels = out_channels
self.num_layers = num_layers
self.dropout = dropout
node_types = list(num_nodes_dict.keys())
num_node_types = len(node_types)
self.num_node_types = num_node_types
self.num_edge_types = num_edge_types
# Create embeddings for all node types that do not come with features.
self.emb_dict = ParameterDict({
f'{key}': Parameter(torch.Tensor(num_nodes_dict[key], in_channels))
for key in set(node_types).difference(set(x_types))
})
I, H, O = in_channels, hidden_channels, out_channels # noqa
# Create `num_layers` many message passing layers.
self.convs = ModuleList()
self.convs.append(RGCNConv(I, H, num_node_types, num_edge_types))
for _ in range(num_layers - 2):
self.convs.append(RGCNConv(H, H, num_node_types, num_edge_types))
self.convs.append(RGCNConv(H, O, self.num_node_types, num_edge_types))
self.reset_parameters()
def reset_parameters(self):
for emb in self.emb_dict.values():
torch.nn.init.xavier_uniform_(emb)
for conv in self.convs:
conv.reset_parameters()
def group_input(self, x_dict, node_type, local_node_idx, n_id=None):
# Create global node feature matrix.
if n_id is not None:
node_type = node_type[n_id]
local_node_idx = local_node_idx[n_id]
h = torch.zeros((node_type.size(0), self.in_channels),
device=node_type.device)
for key, x in x_dict.items():
mask = node_type == key
h[mask] = x[local_node_idx[mask]]
for key, emb in self.emb_dict.items():
mask = node_type == int(key)
h[mask] = emb[local_node_idx[mask]]
return h
def forward(self, n_id, x_dict, adjs, edge_type, node_type,
local_node_idx):
x = self.group_input(x_dict, node_type, local_node_idx, n_id)
node_type = node_type[n_id]
for i, (edge_index, e_id, size) in enumerate(adjs):
x_target = x[:size[1]] # Target node embeddings.
node_type = node_type[:size[1]] # Target node types.
conv = self.convs[i]
x = conv((x, x_target), edge_index, edge_type[e_id], node_type)
if i != self.num_layers - 1:
x = F.relu(x)
x = F.dropout(x, p=0.5, training=self.training)
return x.log_softmax(dim=-1)
def inference(self, x_dict, edge_index_dict, key2int):
# We can perform full-batch inference on GPU.
device = list(x_dict.values())[0].device
x_dict = copy(x_dict)
for key, emb in self.emb_dict.items():
x_dict[int(key)] = emb
adj_t_dict = {}
for key, (row, col) in edge_index_dict.items():
adj_t_dict[key] = SparseTensor(row=col, col=row).to(device)
for i, conv in enumerate(self.convs):
out_dict = {}
for j, x in x_dict.items():
out_dict[j] = conv.root_lins[j](x)
for keys, adj_t in adj_t_dict.items():
src_key, target_key = keys[0], keys[-1]
out = out_dict[key2int[target_key]]
tmp = adj_t.matmul(x_dict[key2int[src_key]], reduce='mean')
out.add_(conv.rel_lins[key2int[keys]](tmp))
if i != self.num_layers - 1:
for j in range(self.num_node_types):
F.relu_(out_dict[j])
x_dict = out_dict
return x_dict
def fullBatch_inference(self, x_dict, edge_index, edge_type, node_type,
local_node_idx):
x = self.group_input(x_dict, node_type, local_node_idx)
node_type = node_type[edge_index[1].unique()]
for i, conv in enumerate(self.convs):
x = conv(x, edge_index, edge_type, node_type)
if i != self.num_layers - 1:
x = F.relu(x)
x = F.dropout(x, p=self.dropout, training=self.training)
return x
def group_inference(self, x_dict, edge_index_dict, key2int):
x_dict = copy(x_dict)
for k, emb in self.emb_dict.items():
x_dict[int(k)] = emb
for i, conv in enumerate(self.convs):
out_dict = dict()
for n, x_target in x_dict.items():
edge_index_n = []
edge_type_n = []
x_nbr = []
for k, e_i in edge_index_dict.items():
if key2int[k[-1]] == n:
edge_index_n.append(e_i)
edge_type_n.append(
e_i.new_full((e_i.size(1), ), key2int[k]))
x_nbr.append(x_dict[key2int[k[0]]])
edge_index_n = torch.cat(edge_index_n, dim=1)
edge_type_n = torch.cat(edge_type_n, dim=0)
x = torch.cat([x_target] + x_nbr, dim=0)
node_type_n = edge_index_n.new_full((x_target.size(0), ), n)
x_out = conv((x, x_target), edge_index_n, edge_type_n,
node_type_n)
if i != self.num_layers - 1:
x_out = F.relu(x_out)
x_out = F.dropout(x_out,
p=self.dropout,
training=self.training)
out_dict[n] = x_out
# if i != self.num_layers - 1:
# for j in range(self.num_node_types):
# F.relu_(out_dict[j])
x_dict = out_dict
return x_dict
class BRGCN(torch.nn.Module):
def __init__(self,
in_channels,
hidden_channels,
out_channels,
num_layers,
dropout,
neg_slope,
heads,
num_relations,
num_nodes_dict,
x_types,
attention='Bilevel'):
super(BRGCN, self).__init__()
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.out_channels = out_channels
self.num_layers = num_layers
self.dropout = dropout
self.heads = heads
self.num_relations = num_relations
node_types = list(num_nodes_dict.keys())
num_node_types = len(node_types)
self.emb_dict = ParameterDict({
f'{key}': Parameter(torch.Tensor(num_nodes_dict[key], in_channels))
for key in set(node_types).difference(set(x_types))
})
self.convs = ModuleList()
if attention == 'Bilevel':
self.convs.append(
BRGCNConv(self.in_channels, self.hidden_channels, neg_slope,
self.num_relations, num_node_types, self.heads,
self.dropout))
for _ in range(num_layers - 2):
self.convs.append(
BRGCNConv(self.hidden_channels, self.hidden_channels,
neg_slope, self.num_relations, num_node_types,
self.heads, self.dropout))
self.convs.append(
BRGCNConv(self.heads * self.hidden_channels, self.out_channels,
neg_slope, self.num_relations, num_node_types, 1,
self.dropout))
elif attention == 'Node':
self.convs = ModuleList()
self.convs.append(
BRGCNConvNode(self.in_channels, self.hidden_channels,
neg_slope, self.num_relations, num_node_types,
self.heads, self.dropout))
for _ in range(num_layers - 2):
self.convs.append(
BRGCNConvNode(self.hidden_channels, self.hidden_channels,
self.neg_slope, self.num_relations,
num_node_types, self.heads, self.dropout))
self.convs.append(
BRGCNConvNode(self.heads * self.hidden_channels,
self.out_channels, self.neg_slope,
self.num_relations, num_node_types, 1,
self.dropout))
elif attention == 'Relation':
self.convs = ModuleList()
self.convs.append(
BRGCNConvRel(self.in_channels, self.hidden_channels, neg_slope,
self.num_relations, num_node_types, self.heads,
self.dropout))
for _ in range(num_layers - 2):
self.convs.append(
BRGCNConvRel(self.hidden_channels, self.hidden_channels,
self.neg_slope, self.num_relations,
self.heads, num_node_types, self.dropout))
self.convs.append(
BRGCNConvRel(self.heads * self.hidden_channels,
self.out_channels, self.neg_slope,
self.num_relations, num_node_types, 1,
self.dropout))
else:
raise NotImplementedError
self.reset_parameters()
def reset_parameters(self):
for embedding in self.emb_dict.values():
init.xavier_uniform_(embedding)
for conv in self.convs:
conv.reset_parameters()
def group_input(self, x_dict, node_type, local_node_idx, n_id=None):
# Create global node feature matrix.
if n_id is not None:
node_type = node_type[n_id]
local_node_idx = local_node_idx[n_id]
h = torch.zeros((node_type.size(0), self.in_channels),
device=node_type.device)
for key, x in x_dict.items():
mask = node_type == key
h[mask] = x[local_node_idx[mask]]
for key, emb in self.emb_dict.items():
mask = node_type == int(key)
h[mask] = emb[local_node_idx[mask]]
return h
def forward(self, n_id, x_dict, adjs, edge_type, node_type,
local_node_idx):
'''
:param n_id: node index for the source nodes
:param x_dict: Node embedding dictionary with node type as key
:param adjs: source to node structure, (edge_index, e_id, size)
:param edge_type: the edge type for all edges
:param node_type: the node type for all nodes
:param local_node_idx: transform the global node id to a local one order for each type of node
:return:
'''
x = self.group_input(x_dict, node_type, local_node_idx, n_id)
node_type = node_type[n_id]
for i, (edge_index, e_id, size) in enumerate(adjs):
x_target = x[:size[1]] # Target node embeddings.
node_type = node_type[:size[1]] # Target node types.
conv = self.convs[i]
x = conv((x, x_target), edge_index, edge_type[e_id], node_type)
if i != self.num_layers - 1:
x = F.relu(x)
x = F.dropout(x, p=self.dropout, training=self.training)
return x.log_softmax(dim=-1)
def fullBatch_inference(self, x_dict, edge_index, edge_type, node_type,
local_node_idx):
x = self.group_input(x_dict, node_type, local_node_idx)
node_type = node_type[edge_index[1].unique()]
for i, conv in enumerate(self.convs):
x = conv(x, edge_index, edge_type, node_type)
if i != self.num_layers - 1:
x = F.relu(x)
x = F.dropout(x, p=self.dropout, training=self.training)
return x
def group_inference(self, x_dict, edge_index_dict, key2int):
x_dict = copy(x_dict)
for k, emb in self.emb_dict.items():
x_dict[int(k)] = emb
for i, conv in enumerate(self.convs):
out_dict = dict()
for n, x_target in x_dict.items():
edge_index_n = []
edge_type_n = []
node_type_nbr = []
x_nbr = []
for k, e_i in edge_index_dict.items():
if key2int[k[-1]] == n:
edge_index_n.append(e_i)
edge_type_n.append(
e_i.new_full((e_i.size(1), ), key2int[k]))
node_type_nbr.append(
e_i.new_full((x_dict[key2int[k[0]]].size(0), ),
key2int[k[0]]))
x_nbr.append(x_dict[key2int[k[0]]])
edge_index_n = torch.cat(edge_index_n, dim=1)
edge_type_n = torch.cat(edge_type_n, dim=0)
x = torch.cat([x_target] + x_nbr, dim=0)
node_type_target = edge_index_n.new_full((x_target.size(0), ),
n)
node_type_n = torch.cat([node_type_target] + node_type_nbr,
dim=0)
x_out = conv((x, x_target), edge_index_n, edge_type_n,
node_type_n)
if i != self.num_layers - 1:
x_out = F.relu(x_out)
x_out = F.dropout(x_out,
p=self.dropout,
training=self.training)
out_dict[n] = x_out
# if i != self.num_layers - 1:
# for j in range(self.num_node_types):
# F.relu_(out_dict[j])
x_dict = out_dict
return x_dict