nett
is an R package for the analysis of network data
with a focus on community detection and implements
multiple methods for hypothesis testing. It includes model selection and
goodness-of-fit tests of SBM/DCSBM to network data, which are useful in
network statistical analysis. Some of the implemented functionality are
follows: - Spectral clustering with regularization - Conditional
pseudo-likelihood for community detection (Amini, Chen,
Bickel and Levina). - Spectral goodness-of-test for SBM and DCSBM
(inspired by Bickel
and Sarkar, and Lei’s work).
- Likelihood ratio tests and BIC selection for SBM and DCSBM (inspired
by Wang and
Bickel’s work among others.) - Likelihood computations for SBM and
DCSBM. - Network Adjusted Chi-square test for SBM and DCSBM (Zhang and Amini). -
Bethe-Hessian Selection for DCSBM (inspired by Le
and Levina’s work). - …
Most of the computations haven been adapted to Poisson models to make them fast and scalable.
Check out the articles for some examples of how to use the package.
To install, you can use the following command
devtools::install_github("aaamini/nett")
See the related repo linfanz/nac-test, for some experiments comparing goodness-of-fit and model selection approaches.