def forward(self, data): # x has shape [N, in_channels] # edge_index has shape [2, E] # pdb.set_trace() x = data.x edge_index = data.edge_index batch = data.batch edge_attr = data.edge_attr x = F.relu(self.conv1(x, edge_index)) x, edge_index, edge_attr, batch, _, _ = self.pool1(x, edge_index, edge_attr, batch) x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1) x = F.relu(self.conv2(x, edge_index)) x, edge_index, edge_attr, batch, _, _ = self.pool2(x, edge_index, edge_attr, batch) x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1) x = F.relu(self.conv3(x, edge_index)) x, edge_index, edge_attr, batch, _, _ = self.pool3(x, edge_index, edge_attr, batch) x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1) x = x1 + x2 + x3 x = F.relu(self.lin1(x)) x = self.lin2(x) x = x.reshape(-1) return x
def forward(self, x, edge_index, batch, inference = False):
x = self.item_embedding(x)
x = x.squeeze(1)
x = F.leaky_relu(self.conv1(x, edge_index))
x, edge_index, _, batch, _, _= self.pool1(x, edge_index, None, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim = 1)
x = F.leaky_relu(self.conv2(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool2(x, edge_index, None, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim = 1)
x = F.leaky_relu(self.conv3(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim = 1)
x = x1 + x2 + x3
x = self.lin1(x)
x = self.bn1(self.act1(x))
x = self.lin2(x)
x = self.bn2(self.act2(x))
x = F.dropout(x, p = 0.5, training = self.training)
x = self.lin3(x).squeeze(1)
if inference:
x_out = x1 + x2 + x3
return x, x_out
else:
return x
def forward(self, x, edge_index, batch):
# x, edge_index, batch = data.x, data.edge_index, data.batch
x = F.relu(self.conv1(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, None, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool2(x, edge_index, None, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv4(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool4(x, edge_index, None, batch)
x4 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv5(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool5(x, edge_index, None, batch)
x5 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = x1 + x2 + x3 + x4 + x5
x = F.relu(self.lin1(x))
x = F.dropout(x, p=0.5, training=self.training)
x = F.relu(self.lin2(x))
x = F.log_softmax(self.lin3(x), dim=-1)
return x
def forward(self, x, edge_index, batch, edge_attr):
edge_attr = edge_attr.squeeze()
edge_index, edge_attr = self.augment_adj(edge_index, edge_attr,
x.size(0))
x = self.conv1(x, edge_index, edge_attr)
x, edge_index, edge_attr, batch, perm, score1 = self.pool1(
x, edge_index, edge_attr, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
edge_attr = edge_attr.squeeze()
edge_index, edge_attr = self.augment_adj(edge_index, edge_attr,
x.size(0))
x = self.conv2(x, edge_index, edge_attr)
x, edge_index, edge_attr, batch, perm, score2 = self.pool2(
x, edge_index, edge_attr, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = torch.cat([x1, x2], dim=1)
x = F.relu(self.bn4(self.fc1(x)))
x = F.dropout(x, p=0.5, training=self.training)
x = F.relu(self.bn5(self.fc2(x)))
x = F.dropout(x, p=0.5, training=self.training)
x = F.log_softmax(self.fc3(x), dim=-1)
return x, score1, score2
def forward(self, x, edge_index, batch=None):
if len(x.shape
) == 3: #NEEDED FOR EXAI, NOTICE THAT IT MUST BE ONLY ONE MOL
data_list = []
for x_i, edge_index_i in zip(x, edge_index):
data_list.append(Data(x=x_i, edge_index=edge_index_i))
data = Batch.from_data_list(data_list).to(self.device)
x = data.x
batch = data.batch
edge_index = data.edge_index
shape = x.shape
x = x.reshape(-1, shape[-1])
x = self.atom_embedding(x)
x = x.reshape(shape)
x = x.squeeze(1)
x = F.relu(self.conv1(x, edge_index))
x, edge_index, batch, _ = self.pool1(x, edge_index, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index))
x, edge_index, batch, _ = self.pool2(x, edge_index, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = x1 + x2
x = self.linear(x)
#x = torch.sigmoid(x)
x = x.squeeze(1)
return x
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
edge_attr = None
x = F.relu(self.conv1(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch = self.pool1(x, edge_index, edge_attr, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch = self.pool2(x, edge_index, edge_attr, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index, edge_attr))
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(x1) + F.relu(x2) + F.relu(x3)
before_fc = x
x = F.relu(self.lin1(x))
x = F.dropout(x, p=self.dropout_ratio, training=self.training)
x = self.lin2(x)
x = F.dropout(x, p=self.dropout_ratio, training=self.training)
x_cls = F.relu(self.lin3(before_fc))
x_cls = F.dropout(x_cls, p=self.dropout_ratio, training=self.training)
x_cls = self.lin4(x_cls)
x_cls = F.dropout(x_cls, p=self.dropout_ratio, training=self.training)
x_cls = F.log_softmax(self.lin5(x_cls), dim=-1)
return x, x_cls
def forward(self, x, edge_index, batch, edge_attr):
# edge_attr = edge_attr.squeeze()
# edge_index, edge_attr = self.augment_adj(edge_index, edge_attr, x.size(0))
x = self.conv1(x, edge_index)
if x.norm(p=2, dim=-1).min() == 0:
print('x is zeros')
x, edge_index, edge_attr, batch, perm, score1 = self.pool1(
x, edge_index, edge_attr, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
edge_attr = edge_attr.squeeze()
edge_index, edge_attr = self.augment_adj(edge_index, edge_attr,
x.size(0))
x = self.conv2(x, edge_index)
x, edge_index, edge_attr, batch, perm, score2 = self.pool2(
x, edge_index, edge_attr, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = torch.cat([x1, x2], dim=1) #concate
x = self.bn4(F.relu(self.fc1(x)))
x = F.dropout(x, p=0.5, training=self.training)
x = self.bn5(F.relu(self.fc2(x)))
x = F.dropout(x, p=0.5, training=self.training)
x = F.log_softmax(self.fc3(x), dim=-1)
return x, score1, score2
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
if isinstance(edge_index, tuple):
edge_index = torch.stack(edge_index)
edge_attr = None
x = F.relu(self.conv1(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch = self.pool1(x, edge_index, edge_attr,
batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch = self.pool2(x, edge_index, edge_attr,
batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index, edge_attr))
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(x1) + F.relu(x2) + F.relu(x3)
x = F.relu(self.lin1(x))
x = F.dropout(x, p=self.dropout, training=self.training)
x = F.relu(self.lin2(x))
x = F.dropout(x, p=self.dropout, training=self.training)
pred = self.lin3(x)
return pred
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
edge_attr = None
x = F.relu(self.conv1(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch = self.pool1(x, edge_index, edge_attr,
batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch = self.pool2(x, edge_index, edge_attr,
batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index, edge_attr))
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(x1) + F.relu(x2) + F.relu(x3)
x = F.relu(self.lin1(x))
x = F.dropout(x, p=self.dropout, training=self.training)
x = F.relu(self.lin2(x))
x = F.dropout(x, p=self.dropout, training=self.training)
x = self.lin3(x)
pred = F.log_softmax(x, dim=-1)
if data.y is not None:
loss = F.nll_loss(pred, data.y)
return pred, loss
return pred, None
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x = self.item_embedding(x)
x = x.squeeze(1)
x = F.relu(self.conv1(x, edge_index))
# x, edge_index, _, batch, _, _ = self.pool1(x=x,edge_index=edge_index,batch=batch)
x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, None, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index))
# x,edge_index, _, batch, _, _ = self.pool2(x=x,edge_index=edge_index,batch=batch)
x, edge_index, _, batch, _, _ = self.pool2(x, edge_index, None, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index))
# x,edge_index, _, batch, _, _ = self.pool3(x=x,edge_index=edge_index,batch=batch)
x, edge_index, _, batch, _, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = x1 + x2 + x3
x = self.lin1(x)
x = self.act1(x)
x = self.lin2(x)
x = self.act2(x)
x = F.dropout(x, p=0.5, training=self.training)
x = torch.sigmoid(self.lin3(x)).squeeze(1)
return x
def forward(self, data):
x, edge_index_prior, batch, edge_weight_prior = data.x, data.edge_index, \
data.batch, data.edge_attr
edge_weight_learn = self.edge_weight_learn
edge_index_learn = self.edge_index_learn
batch_size = x.shape[0] // self.num_channels
_edge_weight_learn = edge_weight_learn
for i in range(batch_size - 1):
edge_weight_learn = torch.cat(
(edge_weight_learn, _edge_weight_learn), dim=0)
edge_index_learn = edge_index_learn.repeat(1, batch_size).to(device)
# calculate the prior graph representation
x_prior = F.relu(
self.conv_prior_1(x, edge_index_prior, edge_weight_prior))
x_prior = F.relu(
self.conv_prior_2(x_prior, edge_index_prior, edge_weight_prior))
x_prior = gap(x_prior, batch)
# calculate the learned graph representation
x_learn = F.relu(
self.conv_learn_1(x, edge_index_learn, edge_weight_learn))
x_learn = F.relu(
self.conv_learn_2(x_learn, edge_index_learn, edge_weight_learn))
x_learn = gap(x_learn, batch)
x = x_prior + x_learn
# x = torch.cat([x_prior, x_learn], dim=1)
x = self.fc(x)
return x
def forward(self, batched_data):
x, edge_index, edge_attr, node_depth, batch = batched_data.x, batched_data.edge_index, batched_data.edge_attr, batched_data.node_depth, batched_data.batch
x = self.node_encoder(x, node_depth.view(-1,))
xs = []
x = F.relu(self.conv1(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch, _, _ = self.pool1(x, edge_index, edge_attr, batch)
xs += [torch.cat([gmp(x, batch), gap(x, batch)], dim=1)]
for i in range(self.num_layers-1):
x = F.relu(self.convs[i](x, edge_index, edge_attr))
x, edge_index, edge_attr, batch, _, _ = self.pools[i](x, edge_index, edge_attr, batch)
xs += [torch.cat([gmp(x, batch), gap(x, batch)], dim=1)]
x = xs[0]
for i in range(1, len(xs)):
x += xs[i]
x = F.relu(self.lin1(x))
x = F.dropout(x, p=self.dropout_ratio, training=self.training)
x = F.relu(self.lin2(x))
pred_list = []
for i in range(self.max_seq_len):
pred_list.append(self.graph_pred_linear_list[i](x))
return pred_list
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x1 = F.relu(self.conv1(x, edge_index))
x2 = F.relu(self.conv2(x1, edge_index))
x2 = x1 + x2
x3 = F.relu(self.conv3(x2, edge_index))
x3 = x2 + x3
x = F.relu(self.conv4(x3, edge_index))
x, edge_index, _, batch, _ = self.pool1(x, edge_index, None, batch)
x4 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
#x4 = x3 + x4
x5 = F.relu(self.conv5(x, edge_index))
x5 = x + x5
x6 = F.relu(self.conv6(x5, edge_index))
x6 = x5 + x6
x7 = F.relu(self.conv7(x6, edge_index))
x7 = x6 + x7
x8 = F.relu(self.conv8(x7, edge_index))
x8 = x7 + x8
x = F.relu(self.conv9(x8, edge_index))
x, edge_index, _, batch, _ = self.pool2(x, edge_index, None, batch)
x9 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
#x9 = x8 + x9
x10 = F.relu(self.conv10(x, edge_index))
x10 = x + x10
x11 = F.relu(self.conv11(x10, edge_index))
x11 = x10 + x11
x12 = F.relu(self.conv12(x11, edge_index))
x12 = x11 + x12
x13 = F.relu(self.conv13(x12, edge_index))
x13 = x12 + x13
x = F.relu(self.conv14(x13, edge_index))
x, edge_index, _, batch, _ = self.pool3(x, edge_index, None, batch)
x14 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
#x14 = x13 + x14
x = x4 + x9 + x14
x = F.relu(self.lin1(x))
x = F.dropout(x, p=self.dropout_ratio, training=self.training)
x = F.relu(self.lin2(x))
x = F.log_softmax(self.lin3(x), dim=-1)
return x
def forward(self, data):
x, edge_index, batch = data
edge_attr = None
edge_index = edge_index.transpose(0, 1)
x = self.relu(self.conv1(x, edge_index, edge_attr), negative_slope=0.1)
x, edge_index, edge_attr, batch, _, _ = self.pool1(
x, edge_index, None, batch)
# x, edge_index, edge_attr, batch, _ = self.pool1(x, edge_index, edge_attr, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = self.relu(self.conv2(x, edge_index, edge_attr), negative_slope=0.1)
x, edge_index, edge_attr, batch, _, _ = self.pool2(
x, edge_index, None, batch)
# x, edge_index, edge_attr, batch, _ = self.pool2(x, edge_index, edge_attr, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
#
x = self.relu(self.conv3(x, edge_index, edge_attr), negative_slope=0.1)
x, edge_index, edge_attr, batch, _, _ = self.pool3(
x, edge_index, None, batch)
# x, edge_index, edge_attr, batch, _ = self.pool2(x, edge_index, edge_attr, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
# x = self.relu(self.conv4(x, edge_index, edge_attr), negative_slope=0.1)
# x, edge_index, edge_attr, batch, _, _ = self.pool4(x, edge_index, None, batch)
# x, edge_index, edge_attr, batch, _ = self.pool2(x, edge_index, edge_attr, batch)
# x4 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
# x = self.relu(self.conv5(x, edge_index, edge_attr),negative_slope=0.1)
# x, edge_index, edge_attr, batch, _, _ = self.pool5(x, edge_index, None, batch)
# x, edge_index, edge_attr, batch, _ = self.pool3(x, edge_index, edge_attr, batch)
x_information_score = self.calc_information_score(x, edge_index)
score = torch.sum(torch.abs(x_information_score), dim=1)
# x5 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = self.relu(x1, negative_slope=0.1) + self.relu(
x2, negative_slope=0.1) + self.relu(x3, negative_slope=0.1)
# x = self.relu(x1,negative_slope=0.1) + self.relu(x2,negative_slope=0.1)
# + self.relu(x3,negative_slope=0.1)+self.relu(x4,negative_slope=0.1)+self.relu(x5,negative_slope=0.1)
# x = F.relu(x1)
graph_emb = x
# x = self.lin1(x)
x = self.relu(self.lin1(x), negative_slope=0.1)
# x=self.bn1(x)
# x = F.dropout(x, p=self.dropout_ratio, training=self.training)
x = self.relu(self.lin2(x), negative_slope=0.1)
# x=self.bn2(x)
# x = F.dropout(x, p=self.dropout_ratio, training=self.training)
x = self.lin3(x)
# x = F.log_softmax(x, dim=-1)
return x, score.mean(), graph_emb
def forward(self, data, conv_train=False):
x = data.x
edge_index = data.edge_index
x1 = self.norm1(self.act1(self.conv1(x, edge_index)))
x = self.dropout(x1)
x2 = self.norm2(self.act2(self.conv2(x, edge_index)))
x = self.dropout(x2)
x3 = self.norm3(self.act3(self.conv3(x, edge_index)))
h_conv = torch.cat([x1, x2, x3], dim=1)
#compute GNN only output
conv_batch_avg = gap(h_conv, data.batch)
conv_batch_add = gadd(h_conv, data.batch)
conv_batch_max = gmp(h_conv, data.batch)
h_GNN = torch.cat([conv_batch_avg, conv_batch_add, conv_batch_max],
dim=1)
gnn_out = self.out_fun(self.lin_GNN(h_GNN))
if conv_train:
return None, None, gnn_out
#SOM
_, _, som_out_1 = self.som1(x1)
_, _, som_out_2 = self.som2(x2)
_, _, som_out_3 = self.som3(x3)
#READOUT
h1 = self.out_norm1(self.act1(self.out_conv1(som_out_1, edge_index)))
h2 = self.out_norm2(self.act2(self.out_conv2(som_out_2, edge_index)))
h3 = self.out_norm3(self.act3(self.out_conv3(som_out_3, edge_index)))
som_out_conv = torch.cat([h1, h2, h3], dim=1)
som_batch_avg = gap(som_out_conv, data.batch)
som_batch_add = gadd(som_out_conv, data.batch)
som_batch_max = gmp(som_out_conv, data.batch)
h = torch.cat([som_batch_avg, som_batch_add, som_batch_max], dim=1)
h = self.out_norm4(h)
h = self.out_act(self.lin_out1(h))
h = self.dropout(h)
h = self.out_act(self.lin_out2(h))
h = self.dropout(h)
h = self.out_fun(self.lin_out3(h))
return h, h_conv, gnn_out
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
# print('x =', x)
# print('====================')
item_id = x[:, :, 0]
category = x[:, :, 1]
# print("item_id: ", item_id)
# print("category: \n", category)
# print('====================')
emb_item = self.item_embedding(item_id).squeeze(1)
emb_category = self.category_embedding(category).squeeze(1)
# emb_item = emb_item.squeeze(1)
# emb_cat
x = torch.cat([emb_item, emb_category], dim=1)
# print(x.shape)
x = F.relu(self.conv1(x, edge_index))
# print(x.shape)
x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, None, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool2(x, edge_index, None, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = x1 + x2 + x3
x = self.lin1(x)
x = self.act1(x)
x = self.lin2(x)
x = F.dropout(x, p=0.5, training=self.training)
x = self.act2(x)
outputs = []
for i in range(x.size(0)):
output = torch.matmul(emb_item[data.batch == i], x[i, :])
outputs.append(output)
x = torch.cat(outputs, dim=0)
x = torch.sigmoid(x)
# save
# savePath = "./output/test_model.pth"
# torch.save(model.state_dict(), savePath)
return x
def forward(self, data):
# I put the edge weight at the position for edge attribute such that the weight can also be masked.
modular_data, ddi_edge_index, ddi_edge_attr = data
modular_output = []
ids = list(modular_data.keys())
for modular_id in ids:
x, edge_index, edge_weight, batch = modular_data[modular_id]
x = x.to(self.args.device)
edge_index = edge_index.to(self.args.device)
edge_weight = edge_weight.to(self.args.device)
batch = batch.to(self.args.device)
x = F.relu(self.conv1(x, edge_index, edge_weight))
batch = batch.long()
x, edge_index, edge_weight, batch, _, _ = self.pool1(
x, edge_index, edge_weight, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index, edge_weight))
x, edge_index, edge_weight, batch, _, _ = self.pool2(
x, edge_index, edge_weight, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index, edge_weight))
x, edge_index, edge_weight, batch, _, _ = self.pool3(
x, edge_index, edge_weight, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
out_x = torch.cat((x1, x2, x3), dim=1)
modular_output.append(out_x)
modular_feature = torch.cat(tuple(modular_output))
modular_feature = nn.Dropout(self.args.dropout_ratio)(modular_feature)
# x = F.relu(self.conv4(modular_feature, ddi_edge_index, ddi_edge_attr))
x = F.relu(self.conv_noattn(modular_feature, ddi_edge_index))
pos_source, pos_target, neg_source, neg_target = self.feature_split(
x, ddi_edge_index)
pos_feat_x = F.sigmoid(self.lin1(pos_source))
pos_feat_y = F.sigmoid(self.lin2(pos_target))
neg_feat_x = F.sigmoid(self.lin1(neg_source))
neg_feat_y = F.sigmoid(self.lin2(neg_target))
pos_attr = F.sigmoid(self.lin3(ddi_edge_attr[0:self.args.batch_size]))
neg_attr = F.sigmoid(self.lin3(ddi_edge_attr[self.args.batch_size:]))
loss_pos_vec = pos_feat_x + pos_attr - pos_feat_y
loss_neg_vec = neg_feat_x + neg_attr - neg_feat_y
norm_pos = torch.norm(loss_pos_vec, p=2, dim=1)
norm_neg = torch.norm(loss_neg_vec, p=2, dim=1)
loss = 2 * self.ddi_nhid - torch.norm(
loss_pos_vec, p=2,
dim=1) + self.args.neg_decay * torch.norm(loss_neg_vec, p=2, dim=1)
return loss, norm_pos, norm_neg, pos_feat_x
def forward(self, x, batch, c_size):
if self.cfg == 'ATT':
x_sum = gap(x, batch)
c = torch.tanh(torch.matmul(x_sum, self.W)) # context vector
c = torch.cat(
[p.repeat(c_size[idx], 1) for idx, p in enumerate(c)], dim=0)
alpha = torch.sigmoid((x * c).sum(dim=1).view(-1, 1))
return global_add_pool(x * alpha, batch)
elif self.cfg == 'AVG':
return gap(x, batch)
def forward(self, x, edge_index, batch):
# x, edge_index, batch = data.x, data.edge_index, data.batch
x = F.relu(self.conv1(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, None, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool2(x, edge_index, None, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv4(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool4(x, edge_index, None, batch)
x4 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv5(x, edge_index))
x, edge_index, _, batch, _, _ = self.pool5(x, edge_index, None, batch)
x5 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = x1 + x2 + x3 + x4 + x5
x = 1/5 * x
sx = x
x = x.unsqueeze(dim=2)
# attention层
x = F.relu(self.convAtt1(x))
x = self.poolAtt1(x)
x = F.relu(self.convAtt2(x))
x = self.poolAtt2(x)
x = F.relu(self.convAtt3(x))
x = self.poolAtt3(x)
x = F.relu(self.convAtt4(x))
x = self.poolAtt4(x)
x = F.relu(self.convAtt5(x))
x = self.poolAtt5(x)
x = F.relu(self.convAtt6(x))
x = self.poolAtt6(x)
x = x.squeeze()
x = (x + 1) * sx
x = F.relu(self.lin1(x))
x = F.dropout(x, p=0.5, training=self.training)
x = F.relu(self.lin2(x))
x = F.log_softmax(self.lin3(x), dim=-1)
return x
def forward(self, data):
# I put the edge weight at the position for edge attribute such that the weight can also be masked.
modular_data, ddi_edge_index, neg_edge_index, ddi_edge_attr, neg_edge_attr = data
modular_output = []
ids = list(modular_data.keys())
for modular_id in ids:
x, edge_index, edge_weight, batch = modular_data[modular_id]
x = x.to(self.args.device)
edge_index = edge_index.to(self.args.device)
edge_weight = edge_weight.to(self.args.device)
batch = batch.to(self.args.device)
x = F.relu(self.conv1(x, edge_index, edge_weight))
batch = batch.long()
x, edge_index, edge_weight, batch, _, _ = self.pool1(
x, edge_index, edge_weight, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index, edge_weight))
x, edge_index, edge_weight, batch, _, _ = self.pool2(
x, edge_index, edge_weight, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index, edge_weight))
x, edge_index, edge_weight, batch, _, _ = self.pool3(
x, edge_index, edge_weight, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
out_x = torch.cat((x1, x2, x3), dim=1)
modular_output.append(out_x)
modular_feature = torch.cat(tuple(modular_output))
modular_feature = nn.Dropout(self.args.dropout_ratio)(modular_feature)
x = F.relu(self.conv4(modular_feature, ddi_edge_index, ddi_edge_attr))
pos_source, pos_target, neg_source, neg_target = self.feature_split(
x, ddi_edge_index, neg_edge_index)
# sigmoid or softmax or nothing, add relu
pos_feat_x = self.lin1(pos_source)
pos_feat_y = self.lin2(pos_target)
neg_feat_x = self.lin1(neg_source)
neg_feat_y = self.lin2(neg_target)
pos_attr = self.lin3(ddi_edge_attr)
neg_attr = self.lin3(neg_edge_attr)
norm_pos, norm_neg = self.xent_loss(pos_feat_x, pos_feat_y, neg_feat_x,
neg_feat_y)
pos_tgt = torch.ones_like(norm_pos)
neg_tgt = torch.zeros_like(norm_neg)
#
loss = nn.BCEWithLogitsLoss()(
norm_pos, pos_tgt) + nn.BCEWithLogitsLoss()(norm_neg, neg_tgt)
return loss, norm_pos, norm_neg, pos_feat_x
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x = F.relu(self.input_GCL(x, edge_index))
x, edge_index, _, batch, _, _ = self.input_GPL(x, edge_index, None,
batch)
# (batch size, hidden)
out = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
for i in range(self.layer_num - 1):
x = F.relu(getattr(self, f"hidden_GCL{i}")(x, edge_index))
x, edge_index, _, batch, _, _ = getattr(self, f"hidden_GPL{i}")(
x, edge_index, None, batch)
out += torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
return out
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
if(self.training == True):
mask, torchmask = random_drop_node(self.node_per_graph, (int)(batch.size()[0] / self.node_per_graph), 0.5, 0.5)
x = x[mask]
batch = batch[mask]
edge_index, _ = subgraph(torchmask, edge_index, relabel_nodes = True)
x0 = self.linprev(x, edge_index)
x1 = F.relu(self.conv1(x0, edge_index))
x1 = self.bn1(x1)
x2 = F.relu(self.conv2(x1, edge_index))
x2 = self.bn2(x2)
x3 = F.relu(self.conv3(x2, edge_index))
x3 = self.bn3(x3)
x4 = F.relu(self.conv4(x3, edge_index))
x4 = self.bn4(x4)
out = torch.cat([x0,x1,x2,x3,x4], dim=1)
out = torch.cat([gmp(out, batch), gap(out, batch)], dim = 1)
out = self.mlp(out)
out = F.log_softmax(out, dim=-1)
return out
def forward(self, data):
x, batch = data.x, data.batch
edge_index = knn_graph(x, 100, batch) #?
edge_index, _ = dropout_adj(edge_index, p=0.3) #?
batch = data.batch
y=data.x
y=self.point1(y, edge_index) #dim=n_intermediate
pointlist=[y]
for f in range(self.point_depth-1):
y=self.pointfkt[f](y, edge_index)
pointlist.append(y)
y=torch.cat(pointlist, dim=1) #dim=n_intermediate*point_depth
y = torch.cat([gap(y, batch), gmp(y, batch)], dim=1)
x = self.batchnorm1(y)
for g in range(self.lin_depth):
x=F.leaky_relu(self.linearfkt[g](x))
if (g-1)%3==0 and self.lin_depth-1>g: #g=1,4,7,... u. noch mind. zwei weitere Layers
x = self.drop[g](x)
x = self.out(x)
if self.classification:
x = torch.sigmoid(x)
x = x.view(-1)
return x
def forward_single(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
target = data.target
# print('x shape = ', x.shape)
x = self.conv1(x, edge_index)
x = self.relu(x)
x = self.conv2(x, edge_index)
x = self.relu(x)
# apply global max pooling (gmp) and global mean pooling (gap)
x = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = self.relu(self.fc_g1(x))
x = self.dropout(x)
x = self.fc_g2(x)
embedded_xt = self.embedding_xt(target)
conv_xt = self.conv_xt_1(embedded_xt)
# flatten
xt = conv_xt.view(-1, 32 * 121)
xt = self.fc1_xt(xt)
# concat
xc = torch.cat((x, xt), 1)
# add some dense layers
xc = self.fc1(xc)
xc = self.relu(xc)
xc = self.dropout(xc)
xc = self.fc2(xc)
xc = self.relu(xc)
xc = self.dropout(xc)
out = self.out(xc)
return out
def forward(self, data, hidden_layer_aggregator=None):
X = data.x
k = self.max_k
#compute Laplacian
L_edge_index, L_values = get_laplacian(data.edge_index,
normalization="sym")
L = torch.sparse.FloatTensor(L_edge_index, L_values,
torch.Size([X.shape[0],
X.shape[0]])).to_dense()
H = [X]
for i in range(k - 1):
xhi_layer_i = torch.mm(torch.matrix_power(L, i + 1), X)
H.append(xhi_layer_i)
H = self.lin(torch.cat(H, dim=1), self.xhi_layer_mask)
H = self.reservoir_act_fun(H)
H = self.bn_hidden_rec(H)
H_avg = gap(H, data.batch)
H_add = gadd(H, data.batch)
H_max = gmp(H, data.batch)
H = torch.cat([H_avg, H_add, H_max], dim=1)
if self.output == "funnel" or self.output is None:
return self.funnel_output(H)
elif self.output == "one_layer":
return self.one_layer_out(H)
elif self.output == "restricted_funnel":
return self.restricted_funnel_output(H)
else:
assert False, "error in output stage"
def forward(self, data):
"""tbd."""
# Get graph input
x, edge_index, batch = data.x, data.edge_index, data.batch
# Get protein input
target = data.target
x = self.conv1(x, edge_index)
x = self.relu(x)
x = self.conv2(x, edge_index)
x = self.relu(x)
# Apply global max pooling (gmp) and global mean pooling (gap)
x = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = self.relu(self.fc_g1(x))
x = self.dropout(x)
x = self.fc_g2(x)
embedded_xt = self.embedding_xt(target)
conv_xt = self.conv_xt_1(embedded_xt)
# Flatten
xt = conv_xt.view(-1, 32 * 121)
xt = self.fc1_xt(xt)
# Concat
xc = torch.cat((x, xt), 1)
# Add some dense layers
xc = self.fc1(xc)
xc = self.relu(xc)
xc = self.dropout(xc)
xc = self.fc2(xc)
xc = self.relu(xc)
xc = self.dropout(xc)
out = self.out(xc)
return out
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
item_id = x[:, :, 0]
category = x[:, :, 1]
emb_item = self.item_embedding(item_id).squeeze(1)
emb_category = self.category_embedding(category).squeeze(1)
# emb_item = emb_item.squeeze(1)
# emb_cat
x = torch.cat([emb_item, emb_category], dim=1)
# print(x.shape)
x = F.relu(self.conv1(x, edge_index))
# print(x.shape)
x, edge_index, _, batch, _ = self.pool1(x, edge_index, None, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv2(x, edge_index))
x, edge_index, _, batch, _ = self.pool2(x, edge_index, None, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = F.relu(self.conv3(x, edge_index))
x, edge_index, _, batch, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = x1 + x2 + x3
x = self.lin1(x)
x = self.act1(x)
x = self.lin2(x)
x = F.dropout(x, p=0.5, training=self.training)
x = self.act2(x)
outputs = []
for i in range(x.size(0)):
output = torch.matmul(emb_item[data.batch == i], x[i, :])
outputs.append(output)
x = torch.cat(outputs, dim=0)
x = torch.sigmoid(x)
return x
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x = self.bn1(x)
x = F.relu(self.bn2(self.conv1(x, edge_index)))
x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, None, batch)
x = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
x = self.lin1(x)
return x
def forward(self, data):
#print('net 2 forward method called with parameters ' +str(data))
#again, all of it from https://github.com/rusty1s/pytorch_geometric/blob/master/examples/enzymes_topk_pool.py
x, edge_index, edge_attr, batch = data.x, data.edge_index, data.edge_attr, data.batch
x = F.relu(self.conv1(x, edge_index, edge_attr))
#x=F.relu(self.conv1(x, edge_index))
x, edge_index, edge_attr, batch, _, _ = self.pool1(
x, edge_index, edge_attr, batch)
#x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, batch)
#x, edge_index, edge_attr, batch, _, _ = self.pool1(x, edge_index, edge_attr, batch)
#x, edge_index, _, batch, _, _ = self.pool1(x, edge_index, batch)
x1 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
#print('batch size x1 :' + str(batch.shape))
#print('x1 shape : ' + str(x1.shape))
x = F.relu(self.conv2(x, edge_index, edge_attr))
x, edge_index, edge_attr, batch, _, _ = self.pool2(
x, edge_index, edge_attr, batch)
#x, edge_index, _, batch, _, _ = self.pool2(x, edge_index, batch)
x2 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
"""
#print('batch size x2 :' + str(batch.shape))
#print('x2 shape : ' + str(x2.shape))
x = F.relu(self.conv3(x, edge_index))
x, edge_index, _, batch, _ = self.pool3(x, edge_index, None, batch)
x3 = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)""" """
#print('batch size x3 :' + str(batch.shape))
#print('x3 shape : ' + str(x3.shape))
x = x1 + x2 + x3"""
x = x1 + x2
x = F.relu(self.lin1(x))
#print('x after lin1 : ' + str(x.shape))
x = F.dropout(x, p=0.5, training=self.training)
x = F.relu(self.lin2(x))
#print('x after lin2 : ' + str(x.shape))
x = F.log_softmax(self.lin3(x), dim=-1)
#print('x after softmax : ' + str(x.shape))
return x
def forward(self, data):
x, edge_index, batch = data.x, data.edge_index, data.batch
x = self.conv1(x, edge_index)
x = self.conv2(x, edge_index)
x = torch.cat([gmp(x, batch), gap(x, batch)], dim=1)
# x = torch.sigmoid(self.fc(x)).squeeze(1)
x = self.fc(x)
return x
