src.algorithms.deep package

Subpackages

• src.algorithms.deep.models package
  • Submodules
  • src.algorithms.deep.models.FCNet module
    • FCNet
      • FCNet.forward()
  • src.algorithms.deep.models.GCNEncoder module
    • GCNEncoder
      • GCNEncoder.forward()
  • src.algorithms.deep.models.MVGRLModel module
    • Discriminator
      • Discriminator.forward()
    • GCN
      • GCN.forward()
    • MVGRLModel
      • MVGRLModel.encode()
      • MVGRLModel.forward()
    • Projection
      • Projection.forward()
    • Readout
      • Readout.forward()
  • Module contents

Submodules

src.algorithms.deep.ARGA module

class src.algorithms.deep.ARGA.ARGA(graph: Graph, num_clusters: int, lr: float = 0.001, latent_dim: int = 16, epochs: int = 100, k: int = 3, use_pretrained: bool = True, save_model: bool = False)

Bases: DeepAlgorithm

Adversarially Regularized Graph Autoencoder algorithm

Parameters:

• graph (Graph) – Graph object

• lr (float) – Learning rate

• latent_dim (int) – Latent dimension

• epochs (int) – Number of epochs to run

• k (int) – Number of iterations to train the discriminator

• use_pretrained (bool) – Boolean flag to indicate if pretrained model should be used

• save_model (bool) – Boolean flag to indicate if the model should be saved after training

src.algorithms.deep.DeepAlgorithm module

class src.algorithms.deep.DeepAlgorithm.DeepAlgorithm(graph: Graph, num_clusters: int, lr: float = 0.001, latent_dim: int = 16, epochs: int = 100, use_pretrained: bool = True, save_model: bool = False)

Bases: Algorithm

Base class for Deep Graph Clustering algorithms

Parameters:

• graph (Graph) – Graph object

• lr (float) – Learning rate

• latent_dim (int) – Latent dimension

• epochs (int) – Number of epochs to run

• use_pretrained (bool) – Boolean flag to indicate if pretrained model should be used

• save_model (bool) – Boolean flag to indicate if the model should be saved after training

run() → None

Trains the model and runs k-means clustering on the node embeddings.

src.algorithms.deep.GAE module

class src.algorithms.deep.GAE.GAE(graph: Graph, num_clusters: int, lr: float = 0.001, latent_dim: int = 16, epochs: int = 100, use_pretrained: bool = True, save_model: bool = False)

Bases: DeepAlgorithm

Graph Autoencoder algorithm

Parameters:

• graph (Graph) – Graph object

• lr (float) – Learning rate

• latent_dim (int) – Latent dimension

• epochs (int) – Number of epochs to run

• use_pretrained (bool) – Boolean flag to indicate if pretrained model should be used

• save_model (bool) – Boolean flag to indicate if the model should be saved after training

src.algorithms.deep.MVGRL module

class src.algorithms.deep.MVGRL.MVGRL(graph: Graph, num_clusters: int, lr: float = 0.001, latent_dim: int = 16, epochs: int = 100, use_pretrained: bool = True, save_model: bool = False)

Bases: DeepAlgorithm

Multi-View Graph Representation Learning algorithm

Parameters:

• graph (Graph) – Graph object

• lr (float) – Learning rate

• latent_dim (int) – Latent dimension

• epochs (int) – Number of epochs to run

• use_pretrained (bool) – Boolean flag to indicate if pretrained model should be used

• save_model (bool) – Boolean flag to indicate if the model should be saved after training

src.algorithms.deep.utils module

src.algorithms.deep.utils.compute_diffusion_matrix(adj: ndarray, alpha: float = 0.2) → ndarray

Computes the diffusion matrix for MVGRL using PageRank

Parameters:

• adj (np.ndarray) – Adjacency matrix

• alpha (float) – Teleport probability

Returns:

Diffusion matrix

Return type:

np.ndarray

src.algorithms.deep.utils.get_clusters(z: ndarray, n_clusters: int, method: str = 'kmeans') → ndarray

Cluster the encoded data using the specified method

Parameters:

• z (np.ndarray) – Latent space

• n_clusters (int) – Number of clusters

• method (str) – Clustering method

Returns:

Cluster labels

Return type:

np.ndarray

Module contents