I want to share a package that I have developed over the last months, for Approximate Bayesian Computation (ABC) / inference problems with an intractable likelihood. In ABC, models only need to be simulated (to circumvent the specification of a likelihood function), allowing for a broad application of Bayesian inference.
Reason for this implementation was the current lack (at least from what I have found) in ABC packages to compute model evidences as the basis for model selection/comparison (via model probabilities or Bayes Factors). ABCdeZ.jl comes with an algorithm for that, next to other features (please check the docs :)). Also be reminded of the usual caveats of ABC for model selection (see docs or wiki).
This required some modification of a previous SMC algorithm; as such ABCdeZ.jl adapts some code from KissABC.jl. Big thanks for the developers of that package; I added a shared license file in ABCdeZ.jl.
I’m very happy to get feedback! The package is registered, if interested check it out here