Scalable Bayesian inference for self-excitatory stochastic processes applied to big American gunfire data