src.algorithms.deep.models package

Submodules

src.algorithms.deep.models.FCNet module

class src.algorithms.deep.models.FCNet.FCNet(in_channels: int, out_channels: int)[source]

Bases: Module

Fully connected three-layer neural network

Parameters:

  • in_channels (int) – Number of input channels

  • out_channels (int) – Number of output channels

forward(x: tensor) → tensor[source]

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

src.algorithms.deep.models.GCNEncoder module

class src.algorithms.deep.models.GCNEncoder.GCNEncoder(in_channels: int, latent_dim: int, activation: ~torch.nn.modules.module.Module = <class 'torch.nn.modules.activation.ReLU'>)[source]

Bases: Module

Graph Convolutional Network Encoder for Graph Autoencoder

Parameters:

  • in_channels (int) – Number of input channels

  • latent_dim (int) – Latent dimension

  • activation (nn.Module) – Activation function

forward(x: tensor, edge_index: tensor) → tensor[source]

Forward pass

Parameters:

  • x (torch.tensor) – Node features

  • edge_index (torch.tensor) – Edge index

Returns:

Node embeddings

Return type:

torch.tensor

src.algorithms.deep.models.MVGRLModel module

Adapted from https://github.com/kavehhassani/mvgrl/blob/master/node/train.py

class src.algorithms.deep.models.MVGRLModel.Discriminator(in_channels: int)[source]

Bases: Module

Discriminator module

Parameters:

in_channels (int) – Number of features in the hidden GCN layers

forward(ha: tensor, hb: tensor, Ha: tensor, Hb: tensor, Ha_corrupted: tensor, Hb_corrupted: tensor) → tensor[source]

Forward pass of the discriminator computer the MI between the two representations of the views

Parameters:

  • ha (torch.tensor) – Graph embedding of the original view

  • hb (torch.tensor) – Graph embedding of the diffused view

  • Ha (torch.tensor) – Node embedding of the original view

  • Hb (torch.tensor) – Node embedding of the diffused view

  • Ha_corrupted (torch.tensor) – Node embedding of the corrupted original view

  • Hb_corrupted (torch.tensor) – Node embedding of the corrupted diffused view

Returns:

Discriminator output

Return type:

torch.tensor

class src.algorithms.deep.models.MVGRLModel.GCN(in_channels: int, out_channels: int)[source]

Bases: Module

Graph Convolutional Network (GCN) with as single layer

Parameters:

  • in_channels (int) – Number of input features

  • out_channels (int) – Number of output features

forward(x: tensor, edge_index: tensor, edge_weight: tensor = None) → tensor[source]

Forward pass

Parameters:

  • x (torch.tensor) – Input features

  • edge_index (torch.tensor) – Edge index tensor

  • edge_weight (torch.tensor) – Edge weight tensor (if any)

Returns:

Embeddings of the nodes at each GCN layer

Return type:

torch.tensor

class src.algorithms.deep.models.MVGRLModel.MVGRLModel(in_channels: int, latent_dim: int)[source]

Bases: Module

Multi-View Graph Representation Learning (MVGRL) model

Parameters:

  • in_channels (int) – Number of input features

  • latent_dim (int) – Dimension of the latent space

encode(x: tensor, edge_index: tensor, diff_edge_index: tensor, diff_edge_weight: tensor) → tensor[source]

Embedding function

Parameters:

  • x (torch.tensor) – Input features

  • edge_index (torch.tensor) – Edge index tensor

  • diff_edge_index (torch.tensor) – Diffused edge index tensor

  • diff_edge_weight (torch.tensor) – Diffused edge weight tensor

Returns:

Node embeddings

Return type:

torch.tensor

forward(x: tensor, edge_index: tensor, diff_edge_index: tensor, diff_edge_weight: tensor, corrupted_idx: tensor = None)[source]

Forward pass, a=alpha (original view), b=beta (diffused view)

Parameters:

  • x (torch.tensor) – Input features

  • edge_index (torch.tensor) – Edge index tensor

  • diff_edge_index (torch.tensor) – Diffused edge index tensor

  • diff_edge_weight (torch.tensor) – Diffused edge weight tensor

  • corrupted_idx (torch.tensor) – Corrupted index tensor

class src.algorithms.deep.models.MVGRLModel.Projection(latent_dim: int)[source]

Bases: Module

Projection layer

Parameters:

latent_dim (int) – Dimension of the latent space

forward(h: tensor) → tensor[source]

Forward pass

Parameters:

h (torch.tensor) – Node embeddings

Returns:

Projected embeddings

Return type:

torch.tensor

class src.algorithms.deep.models.MVGRLModel.Readout(latent_dim: int)[source]

Bases: Module

Readout function for a one-layer GCN model

forward(h1: tensor, h2: tensor) → tensor[source]

Pooling layer

Parameters:

h (torch.tensor) – Node embeddings at the first GCN layer

Returns:

Pooled embeddings

Return type:

torch.tensor

Module contents