greta is a new R package for statistical modelling that lets you define
your own statistical model, and fit it to data by MCMC, maximum likelihood, or variational Bayes methods. Unlike existing modelling languages (e.g. WinBUGS, JAGS and Stan), greta models are written in R code, so it's easy to integrate them into your workflow. greta uses Google's TensorFlow library for high-performance numerical computation, so it's really fast, scales well to massive data sets, and can parallelise computation across many CPUs or on GPUs.
I will demonstrate how simple it is to specify complex statistical models
with greta, highlight some features that have been added recently (e.g.
differential equations, mixture models and making predictions after model fitting) and show how you can use or extend greta in other R packages. See the greta website for more details: https://greta-dev.github.io/greta
Nick Golding is a research fellow at the University of Melbourne,
developing software and statistical models for ecology and public health.
He's been an avid R user since 2009, has authored 16 R packages, and was an rOpenSci fellow in 2017.