Software

lvm4net: an R package for latent variable models for networks (on CRAN and GitHub)

lvm4net provides a range of tools for latent variable models for network data. Most of the models are implemented using a fast variational inference approach. Latent space models for binary networks: the function lsm implements the latent space model (LSM) using a variational inference and squared Euclidian distance; the function lsjm implements latent space joint model (LSJM) for multiplex networks introduced by Gollini and Murphy (2014). These models assume that each node of a network has a latent position in a latent space: the closer two nodes are in the latent space, the more likely they are connected. Functions for binary bipartite networks will be added soon.

tailloss: an R package to estimate the Probability in the Upper Tail of the Aggregate Loss Distribution (on CRAN and GitHub)

Evaluate the probability in the upper tail of the aggregate loss distribution using different methods: Panjer recursion, Monte Carlo simulations, Markov bound, Cantelli bound, Moment bound, and Chernoff bound.

tailloss contains functions to estimate the exceedance probability curve of the aggregated losses. There are two ‘exact’ approaches: Panjer recursion and Monte Carlo simulations, and four approaches producing upper bounds: the Markov bound, the Cantelli bound, the Moment bound, and the Chernoff bound. The upper bounds are useful and effective when the number of events in the catalogue is large, and there is interest in estimating the exceedance probabilities of exceptionally high losses.

  • Gollini, I., and Rougier, J. (2016), ‘Rapidly bounding the exceedance probabilities of high aggregate losses’, Journal of Operational Risk, 11(3), 97-116 (arXiv:1507.01853).

GWmodel: an R package for geographically weighted models (on CRAN and GitHub)

Sometimes spatial data are not described well by some global model, but where there are regions where a suitably localised calibration provides a better description: geographically weighted (GW) models has been designed to tackle these situations.

In the R package GWmodel, we introduce techniques from this particular branch of spatial statistics.
The approach uses a moving window weighting technique, where localised models are found at target locations. Outputs are mapped to provide a useful exploratory tool into the nature of the data spatial heterogeneity. GWmodel includes all the recent methodological advances in this field: GW summary statistics, GW regression, GW principal component analysis, GW generalised linear models, and GW prediction models.