sparsebm 1.6.2

An implementation of Stochastic Bloc model and Latent Block model efficient with sparse matrices

Getting started with SparseBM

SparseBM is a python module for handling sparse graphs with Block Models. The module is an implementation of the variational inference algorithm for the Stochastic Block Model (SBM) and the Latent Block Model (LBM) for sparse graphs, which leverages the sparsity of edges to scale to very large numbers of nodes. The module can use Cupy to take advantage of the hardware acceleration provided by graphics processing units (GPU).

Installing

The SparseBM module is distributed through the PyPI repository and the documentation is available here.

With GPU acceleration (recommended if GPUs are available)

This option is recommended if GPUs are available to speedup computation.

With the package installer pip:

pip3 install sparsebm[gpu]

The Cupy module will be installed as a dependency.

Alternatively Cupy can be installed separately, and will be used by sparsebm if available.

pip3 install sparsebm
pip3 install cupy

Without GPU acceleration

Without GPU acceleration, only CPUs are used. The infererence process still uses sparsity, but no GPU linear algebra operations.

pip3 install sparsebm

For users who do not have GPU, we recommend the free serverless Jupyter notebook environment provided by Google Colab where the Cupy module is already installed and ready to be used with a GPU.

Example with the Stochastic Block Model

• Generate a synthetic graph for analysis with SBM:

  from sparsebm import generate_SBM_dataset
  
  dataset = generate_SBM_dataset(symmetric=True)
  

• Infer with the Bernoulli Stochastic Bloc Model:

  from sparsebm import SBM
  
  # A number of classes must be specified. Otherwise see model selection.
  model = SBM(dataset.clusters)
  model.fit(dataset.data, symmetric=True)
  print("Labels:", model.labels)
  

• Compute performance:

  from sparsebm.utils import ARI
  ari = ARI(dataset.labels, model.labels)
  print("Adjusted Rand index is {:.2f}".format(ari))
  

Example with the Latent Block Model

• Generate a synthetic graph for analysis with LBM:

  from sparsebm import generate_LBM_dataset
  
  dataset = generate_LBM_dataset()
  

• Use the Bernoulli Latent Bloc Model:

  from sparsebm import LBM
  
  # A number of classes must be specified. Otherwise see model selection.
  model = LBM(
      dataset.row_clusters,
      dataset.column_clusters,
      n_init_total_run=1,
  )
  model.fit(dataset.data)
  print("Row Labels:", model.row_labels)
  print("Column Labels:", model.column_labels)
  

• Compute performance:

  from sparsebm.utils import CARI
  cari = CARI(
      dataset.row_labels,
      dataset.column_labels,
      model.row_labels,
      model.column_labels,
  )
  print("Co-Adjusted Rand index is {:.2f}".format(cari))