Colin Swaney, Senior Research Software Engineer

Network Hawkes processes (Linderman, 2016) are a class of probabilistic models that combine multivariate Hawkes processes with networks models. In a multivariate Hawkes process, the likelihood of future events depends on the prior history of events, which gives rise to highly nonlinear, endogenously generated dynamics. The network Hawkes process framework permits simultaneous inference on the event generating process and a network model that constrains the pattern of event interactions. NetworkHawkesProcesses is a pure-Julia implementation that provides methods to simulate and estimate this class of processes. It allows researchers to construct models from a flexible set of components and perform inference using a selection of compatible methods, including maximum-likelihood, Markov chain Monte Carlo and variational inference.