turing.jl
probabilistic inference, computation, julia, software library
tutorials
https://turing.ml/dev/tutorials/
tutorial on the beta-bernoulli model
tutorial on linear models
tutorial on variational inference: https://turing.ml/dev/tutorials/9-variationalinference/
- By default, i.e. when calling `vi(m, advi)`, Turing uses a mean-field approximation with a multivariate normal as the base-distribution. Mean-field refers to the fact that we assume all the latent variables to be __independent__. This is the “standard” ADVI approach; see Automatic Differentiation Variational Inference by Dustin Tran, Andrew Gelman, David Blei et al in 2016 for more. In Turing, one can obtain such a mean-field approximation by calling Variational.meanfield(model) for which there exists an internal implementation for update