Release of eventnet 1.2

[juergen lerner]

Dear all,

We are happy to announce the release of eventnet version 1.2, which comes with two major additions: new hyperedge statistics ("geometrically-weighted subset repetition") and new hyperedge observation generators ("hyperedge perturbation observation") along with a bug-fix related with the CATDIFF aggregation function.

Geometrically-weighted subset repetition provides similar effects as subset repetition but scales the overlap of future hyperevents with prior hyperevents in a smooth way, which is largely inspired by geometrically-weighted statistics for ERGM; see for example Hunter and Handcock, 2006 (http://dx.doi.org/10.1198/106186006X133069). Preliminary experimental evidence suggests that GW subset repetition is especially beneficial in large and sparse event networks. See the reference guide on RHEM effects in the eventnet wiki for additional explanations. 

The hyperedge perturbation observation generators provide new ways to sample non-events. They may be especially beneficial in event networks in which a randomly sampled subset of nodes has, loosely speaking, a "snowballs chance in hell" to ever jointly participate in an event. In such a situation many hyperedge statistics will have a close-to infinite effect - or from another point of view, model estimation will not converge. The hyperedge perturbation observation generators sample non-events that have a large overlap with the actually observed events by randomly adding, removing, or swapping just a few of the participating nodes. Thus, by design, the sampled non-events are similar to the observed events, while the perturbation creates some variation in hyperedge statistics, potentially allowing to estimate their effect on event rates. Currently there is still limited experience with hyperedge perturbation in RHEM and any feedback would be highly appreciated. 

Best wishes,

  Juergen