Skip to content
New issue

Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.

#

By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? # to your account

Handle GNN model for flexibly using activation functions #7

Merged
merged 1 commit into from
Feb 13, 2023
Merged

Handle GNN model for flexibly using activation functions #7

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion examples/GCN_example.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -64,6 +64,7 @@ def main():
args.num_layers = 3
args.hidden_dim = 32
args.weight_decay = 1e-3
args.device = 'cuda:0'

# GPU / CPU
device = torch.device(args.device if torch.cuda.is_available() else "cpu")
Expand Down Expand Up @@ -94,7 +95,7 @@ def main():
lr=args.lr, weight_decay=args.weight_decay)

print("Device {}".format(device))
if device == torch.device("cuda"):
if device != torch.device("cpu"):
idx_train = idx_train.to(device)
idx_val = idx_val.to(device)
idx_test = idx_test.to(device)
Expand Down
2 changes: 1 addition & 1 deletion examples/GIN_example.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,7 +94,7 @@ def main():
lr=args.lr, weight_decay=args.weight_decay)

print("Device {}".format(device))
if device == torch.device("cuda"):
if device != torch.device("cpu"):
idx_train = idx_train.to(device)
idx_val = idx_val.to(device)
idx_test = idx_test.to(device)
Expand Down
2 changes: 1 addition & 1 deletion examples/GraphSAGE_example.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -92,7 +92,7 @@ def main():
lr=args.lr, weight_decay=args.weight_decay)

print("Device {}".format(device))
if device == torch.device("cuda"):
if device != torch.device("cpu"):
idx_train = idx_train.to(device)
idx_val = idx_val.to(device)
idx_test = idx_test.to(device)
Expand Down
71 changes: 35 additions & 36 deletions graphrl/models/gnn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -34,8 +34,8 @@ def forward(self, input, adj_matrix):
return output

class GCN(BaseModel):
def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0, device="cpu"):
super(GCN, self).__init__(device)
def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0, hidden_layer_activation=nn.LeakyReLU(), output_layer_activation=nn.LogSoftmax()):
super(GCN, self).__init__()
self.num_features = num_features
self.hidden_dim = hidden_dim
self.num_classes = num_classes
Expand All @@ -48,21 +48,23 @@ def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0,
self.conv_layers = nn.ModuleList()
self.linears = nn.ModuleList()

self.conv_layers.append(ConvolutionalLayer(self.num_features))

if self.num_layers < 2:
self.conv_layers.append(ConvolutionalLayer(self.num_features))
self.linears.append(nn.Linear(self.num_features, self.num_classes))
self.activation_modules.append(nn.LogSoftmax())
else:
self.conv_layers.append(ConvolutionalLayer(num_features, self.hidden_dim))
self.linears.append(nn.Linear(self.num_features, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.activation_modules.append(hidden_layer_activation)

for layer in range(self.num_layers - 2):
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim, self.hidden_dim))
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim))
self.linears.append(nn.Linear(self.hidden_dim, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.activation_modules.append(hidden_layer_activation)

self.conv_layers.append(ConvolutionalLayer(self.hidden_dim))
self.linears.append(nn.Linear(self.hidden_dim, self.num_classes))
self.activation_modules.append(nn.LogSoftmax())
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim, self.num_classes))

self.activation_modules.append(output_layer_activation)

def forward(self, x, adj_matrix):
for layer in range(self.num_layers):
Expand All @@ -77,8 +79,8 @@ def forward(self, x, adj_matrix):


class GraphSAGE(BaseModel):
def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0, device="cpu"):
super(GraphSAGE, self).__init__(device)
def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0, hidden_layer_activation=nn.LeakyReLU(), output_layer_activation=nn.LogSoftmax()):
super(GraphSAGE, self).__init__()
self.num_features = num_features
self.hidden_dim = hidden_dim
self.num_classes = num_classes
Expand All @@ -91,21 +93,23 @@ def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0,
self.conv_layers = nn.ModuleList()
self.linears = nn.ModuleList()

self.conv_layers.append(ConvolutionalLayer(self.num_features))

if self.num_layers < 2:
self.conv_layers.append(ConvolutionalLayer(self.num_features))
self.linears.append(nn.Linear(self.num_features * 2, self.num_classes))
self.activation_modules.append(nn.LogSoftmax())
else:
self.conv_layers.append(ConvolutionalLayer(num_features, self.hidden_dim))
self.linears.append(nn.Linear(self.num_features * 2, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.activation_modules.append(hidden_layer_activation)

for layer in range(self.num_layers - 2):
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim, self.hidden_dim))
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim))
self.linears.append(nn.Linear(self.hidden_dim * 2, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.activation_modules.append(hidden_layer_activation)

self.conv_layers.append(ConvolutionalLayer(self.hidden_dim))
self.linears.append(nn.Linear(self.hidden_dim * 2, self.num_classes))
self.activation_modules.append(nn.LogSoftmax())
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim, self.num_classes))

self.activation_modules.append(output_layer_activation)

def forward(self, x, adj_matrix):
for layer in range(self.num_layers):
Expand All @@ -119,8 +123,8 @@ def forward(self, x, adj_matrix):
return output

class GIN(BaseModel):
def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0, device="cpu"):
super(GIN, self).__init__(device)
def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0, hidden_layer_activation=nn.LeakyReLU(), output_layer_activation=nn.LogSoftmax()):
super(GIN, self).__init__()
self.num_features = num_features
self.hidden_dim = hidden_dim
self.num_classes = num_classes
Expand All @@ -133,23 +137,18 @@ def __init__(self, num_features, hidden_dim, num_classes, num_layers, dropout=0,
self.conv_layers = nn.ModuleList()
self.linears = nn.ModuleList()

if self.num_layers < 2:
self.conv_layers.append(ConvolutionalLayer(self.num_features))
self.linears.append(nn.Linear(self.num_features, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.output_activation = nn.LogSoftmax()
self.out_linear = torch.nn.Linear(self.num_features + hidden_dim, self.num_classes)
else:
self.conv_layers.append(ConvolutionalLayer(num_features, self.hidden_dim))
self.linears.append(nn.Linear(self.num_features, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.conv_layers.append(ConvolutionalLayer(self.num_features))
self.linears.append(nn.Linear(self.num_features, self.hidden_dim))
self.activation_modules.append(hidden_layer_activation)

if self.num_layers > 1:
for layer in range(self.num_layers - 1):
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim, self.hidden_dim))
self.conv_layers.append(ConvolutionalLayer(self.hidden_dim))
self.linears.append(nn.Linear(self.hidden_dim, self.hidden_dim))
self.activation_modules.append(nn.LeakyReLU())
self.activation_modules.append(hidden_layer_activation)

self.output_activation = nn.LogSoftmax()
self.out_linear = torch.nn.Linear(self.num_features + hidden_dim*self.num_layers, self.num_classes)
self.out_linear = torch.nn.Linear(self.num_features + hidden_dim*self.num_layers, self.num_classes)
self.output_activation = output_layer_activation

def forward(self, x, adj_matrix):
hidden_states = [x]
Expand Down