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