def forward(self, data, pos_edge_index, neg_edge_index):
x, edge_index = data.x, data.train_edge_index
# 0 layer
# s1 = self.score1(x, masked_node)
# masked_node = x[data.cold_mask_node]
# 1 layer
x = F.relu(self.conv1(x, edge_index))
# masked_node = F.relu(self.conv1(masked_node, torch.zeros([2,1], dtype=edge_index.dtype, device= edge_index.device)))
# s2 = self.score2(x, masked_node)
# 2 layer
x = self.conv2(x, edge_index)
# masked_node = self.conv2(masked_node, torch.zeros([2,1], dtype=edge_index.dtype, device= edge_index.device))
# s3 = self.score2(x, masked_node)
# x[data.cold_mask_node] = masked_node
total_edge_index = torch.cat([pos_edge_index, neg_edge_index], dim=-1)
x_j = torch.index_select(x, 0, total_edge_index[0])
x_i = torch.index_select(x, 0, total_edge_index[1])
return torch.einsum("ef,ef->e", x_i, x_j)
# return F.softmax((s1+s2+s3), 1)
dataset = get_planetoid_dataset(args.dataset, args.normalize_features)
dataset = dataset.shuffle()
permute_masks = random_planetoid_splits if args.random_splits else None
run(dataset, Net(dataset), args.runs, args.epochs, args.lr, args.weight_decay,
args.early_stopping, permute_masks)
x = F.relu(self.conv1(x, edge_index))
# masked_node = F.relu(self.conv1(masked_node, torch.zeros([2,1], dtype=edge_index.dtype, device= edge_index.device)))
# s1 = self.score(x, masked_node)
# x_j = torch.index_select(x, 0, total_edge_index[0])
# x_i = torch.index_select(x, 0, total_edge_index[1])
# s1 = x_i-x_j
# 2 layer5
x = self.conv2(x, edge_index)
x_j = torch.index_select(x, 0, total_edge_index[0])
x_i = torch.index_select(x, 0, total_edge_index[1])
# dist = x_j-x_i
# out = F.relu(self.score(dist).squeeze())
# score_loss = torch.matmul(dist, self.score.weight.squeeze()).mean()
return torch.einsum("ef,ef->e", x_i,
x_j), torch.zeros(total_edge_index.size())
# return out, score_loss
# return F.log_softmax(x, dim=1)
dataset = get_planetoid_dataset(args.dataset, args.normalize_features)
run(dataset, Net(dataset), args.runs, args.epochs, args.lr, args.weight_decay,
args.early_stopping)
x = self.conv2_ssl(x, added_index) # LAYER 2
return F.log_softmax(x, dim=1), z,r.size(-1)
# ego-Facebook com-Amazon ego-gplus ego-twitter
# name = "ego-Facebook"
# path = osp.join(osp.dirname(osp.realpath(__file__)), '..', 'data', name)
# dataset = SNAPDataset(path, name)
# # pdb.set_trace()
# run(dataset, DoubleNet(dataset, dataset[0].num_feature, int(dataset[0].num_class)), args.runs, args.epochs, args.lr, args.weight_decay,
# args.early_stopping)
motif_size =3
max_value = 1000
gen_interval= n_epochs_gen = 3
dis_interval= n_epochs_dis = 3
batch_size_gen = 64
batch_size_dis = 64
# "CiteSeer", "PubMed"
names = ["CiteSeer"]
for name in names:
dataset = get_planetoid_dataset(name, args.normalize_features)
dataset = dataset.shuffle()
# pdb.set_trace()
run(dataset, DoubleNet(dataset, dataset.num_features, dataset.num_classes), args.runs, args.epochs, args.lr, args.weight_decay,
args.early_stopping)