Inferring Latent Diffusion Networks

This is an R implementation of the netinf algorithm (Gomez Rodriguez, Leskovec, and Krause, 2010). Given a set of events that spread between a set of nodes the algorithm infers the most likely stable diffusion network that is underlying the diffusion process.


NetworkInference 1.2.3

New Features

Bug Fixes

  • Fixed potential memory error from iterating over the beginning/end of a std::map

NetworkInference 1.2.2

New Features

Bug Fixes

  • netinf() with log-normal model didn't run because of an index error in the argument check for params
  • Fixed memory allocation error caused by uninitialized comparison

NetworkInference 1.2.1

Bug Fixes

  • netinf_ used ceiling on integer which caused error on Solaris

NetworkInference 1.2.0

New Features

Changes to netinf()

  • netinf() got another speed-up. After the first edge, the computation time for each edge is reduced by the factor number of nodes in the network
  • Number of edges can now be chosen using a Vuong style test. If this procedure should be used, a p-value is chosen at which the inference of new edges stops. This value is specified via the new p_value_cutoff argument to netinf().
  • This lead to the netinf output having a fourth column now, containing the p-value for each edge. The p-value is also available if a fixed number of edges is chosen.
  • If no starting values are provided via the params argument parameters are initialized by choosing the midpoint between the maximum possible parameter value and the minimum possible value. These values are derived using the closed form MLE of the respective parameter, derived from either the minimum possible diffusion times (assuming a diffusion 'chain', i.e. a -> b -> c -> ...) or the maximum possible diffusion times (assuming a diffusion 'fan', i.e. a -> b, a -> c, a -> d,...).
  • n_edges can now specify either an absolute number of edges, or a p-value cutoff in the interval (0, 1) for the Vuong test
  • The log normal distribution is now available as a diffusion model. With this comes a change in the arguments for netinf. Instead of lambda, parameters are now specified with a vector (or scalar depending on distribution) params. For exponential and rayleigh distributions params is just the rate / alpha parameter. For the log-normal distribution params specifies mean and variance (in that order). See the netinf() documentation for details on specificaiton and parametrization (?netinf).
  • The output from netinf() now contains information on the model, parameters and iterations as attributes. See the documentation for details.
  • The policies dataset has been updated with over 600 new policies from the SPID database (access via data(policies)).
  • Inferred cascade trees can now be returned by setting trees = TRUE.

New functions

  • New function drop_nodes() now allows to drop nodes from all cascades in a cascade object.

Changes to simulate_cascades()

  • simulate_cascades() now supports passing of additional (isolated in the diffusion network) nodes via the nodes argument.
  • simulate_cascades() now also supports the log-normal distribution.

Bug Fixes

  • Inference of very uninformative edges (large number of edges) could lead for the software to break. Fixed now
  • In simulate_cascades() with partial cascades provided, it was possible that nodes experienced an event earlier than the last event in the partial cascade. Now, the earliest event time is the last observed event time in the partial cascade.

Other changes

  • C++ code is now modularized and headers are properly documented

NetworkInference 1.1.2

New Features

  • We made changes to the internal data structures of the netinf function, so it is much faster and memory efficient now.
  • netinf() now has a shiny progress bar!
  • as.cascade is now completely removed (see release note on version 1.1.0).
  • New convenience function to subset cascades by time (subset_cascade_time) and by cascade id (subset_cascade).

Bug Fixes

  • Long running functions (that call compiled code) can now be interrupted without crashing the R session.
  • as_cascade_long() and as_cascade_wide() handle date input correctly now.
  • as_cascade_wide() couldn't handle data input of class data.table.

NetworkInference 1.1.1

Bug Fixes

  • Use of igraph now conditional compliant with Writing R Extensions
  • Fixed version number displayed in startup message

NetworkInference 1.1.0

New Features

  • Data format (long or wide) of as.cascade is not bound to the class of the data object anymore. In 1.0.0 wide format had to be a matrix and long format had to be a dataframe. This did not make much sense. as.cascade is now deprecated and replaced by two new functions as_cascade_long and as_cascade_wide.

Bug Fixes

  • x and y axis labels in plot.cascade with option label_nodes=FALSE were reversed

NetworkInference 1.0.0

First release

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


1.2.3 by Fridolin Linder, 4 months ago

Report a bug at

Browse source code at

Authors: Fridolin Linder [aut, cre] , Bruce Desmarais [ctb]

Documentation:   PDF Manual  

MIT + file LICENSE license

Imports Rcpp, assertthat, checkmate, ggplot2, ggrepel, stats

Suggests testthat, knitr, rmarkdown, pander, igraph, utils, dplyr

Linking to Rcpp, RcppProgress

System requirements: C++11

See at CRAN