Announcement: GraSPy 0.0.2

We're happy to announce the release of GraSPy 0.0.2! GraSPy is a Python package for understanding the properties of random graphs that arise from modern datasets, such as social networks and brain networks.

For more information, please visit our website and our tutorials.

Highlights

This release is the result of 3 months of work with over 16 pull requests by 5 contributors. Highlights include:

  • Nonparametric hypothesis testing method for testing two non-vertex matched graphs.
  • Plotting updates to pairplot, gridplot and heatmaps.
  • Sampling degree-correlcted stochatic block models (DC-SBM).
  • import_edgelist function for importing single or multiple edgelists.
  • Enforcing Black formatting for the package.

Improvements

  • Embedding methods are now fully sklearn-compliant. This is tested via check_estimator function in sklearn.
  • gridplot and heatmap can now plot hierchical labels.
  • New Laplacian computing method ('R-DAD') by adding a constant to the diagonal degree matrix.
  • Semiparametric testing only checks for largest connected component (LCC) in the intial embeddings.
  • Various bug fixes.
  • Various tutorial latex fixes.
  • Various documentation clarifications.
  • More consistent documentation.

API Changes

  • check_lcc argument in AdjacencySpectralEmbed, LaplacianSpectralEmbed, and OmnibusEmbed classes, which checks if input graph(s) are fully connected when check_lcc is True.
  • gridplot and heatmap now have a inner_hier_labels and outer_hier_labels, which are used for hierarchical labeling of nodes.
  • to_laplace function now has regularizer arg for when form is 'R-DAD'.
  • sbm function now has dc and dc_kws arguments for sampling SBM with degree-correction.

Deprecations

None.