This package provides simple utilities for constructing zero-inflated distributions.
It is mainly intended to for fitting zero-inflated and hurdle models using Turing.jl.
ZeroInflation.jl exports a single function (zeroinflated) that can be used to construct arbitrary zero-inflated distributions. For example to construct a zero-inflated Poisson distribution with parameter λ = 1.0 and mixing probability θ = 0.2, call
zeroinflated(Poisson(1.0), 0.2)
Sometimes it can be useful (e.g. in regression modelling) to express the mixing distribution not in terms of raw probabilities, but on the logit scale. In this case the distribution with parameter λ = 1.0 and mixing probability θ = logistic(0.0) can be constructed by
zeroinflated(Poisson(1.0), StatsFuns.logistic(0.0))
or by directly providing a logit parameterized Bernoulli distribution,
zeroinflated(Poisson(1.0), Turing.BernoulliLogit(0.0))
.