Fits a negative binomial (NB) generalized linear model (GLM) using Adam. Since Adam supports minibatching adaglm can run on very large datasets (millions of datapoints). By default the fitting is done in two steps: 1) Fit a Poisson GLM. 2) Starting with coefficients initialized from the Poisson GLM, fit a NB GLM jointly optimizing the coefficients and concentration.
See vignettes/negbin_glm.Rmd for example usage.