Add CDFBijector and QuantileBijector

src/bijectors/cdf_quantile.jl

@@ -0,0 +1,90 @@
+"""
+    CDFBijector(dist::Distributions.ContinuousUnivariateDistribution)
+
+A [`Bijector`](@ref) that transforms the input from the support of the given distribution to
+the unit interval using the cumulative distribution function of the distribution.
+
+The inverse is [`QuantileBijector`](@ref).
+
+# Example
+
+```jldoctest
+julia> using Bijectors: CDFBijector
+
+julia> using Distributions: Normal
+
+julia> b = CDFBijector(Normal());
+
+julia> p = [0.1, 0.5, 0.9];
+
+julia> transform(b, quantile.(Normal(), p)) ≈ p
+true
+```
+"""
+struct CDFBijector{D<:ContinuousUnivariateDistribution} <: Bijector
+    dist::D
+end
+
+Base.:(==)(b1::CDFBijector, b2::CDFBijector) = b1.dist == b2.dist
+
+Functors.@functor CDFBijector
+
+function Base.show(io::IO, b::CDFBijector)
+    print(io, "CDFBijector(")
+    print(io, b.dist)
+    print(io, ")")
+    return nothing
+end
+
+with_logabsdet_jacobian(b::CDFBijector, x) = transform(b, x), logabsdetjac(b, x)
+
+transform(b::CDFBijector, x) = Distributions.cdf.(b.dist, x)
+
+logabsdetjac(b::CDFBijector, x) = Distributions.logpdf.(b.dist, x)
+
+"""
+    QuantileBijector(dist::Distributions.ContinuousUnivariateDistribution)
+
+A [`Bijector`](@ref) that transforms the input from the unit interval to the support of the
+given distribution using the quantile function of the distribution.
+
+The inverse is [`CDFBijector`](@ref).
+
+# Example
+
+```jldoctest
+julia> using Bijectors: QuantileBijector
+
+julia> using Distributions: Gamma
+
+julia> b = QuantileBijector(Gamma());
+
+julia> p = [0.1, 0.5, 0.9];
+
+julia> transform(b, p) ≈ quantile.(Gamma(), p)
+true
+```
+"""
+struct QuantileBijector{D<:ContinuousUnivariateDistribution} <: Bijector
+    dist::D
+end
+
+Base.:(==)(b1::QuantileBijector, b2::QuantileBijector) = b1.dist == b2.dist
+
+Functors.@functor QuantileBijector
+
+function Base.show(io::IO, b::QuantileBijector)
+    print(io, "QuantileBijector(")
+    print(io, b.dist)
+    print(io, ")")
+    return nothing
+end
+
+with_logabsdet_jacobian(b::QuantileBijector, x) = transform(b, x), logabsdetjac(b, x)
+
+transform(b::QuantileBijector, x) = Distributions.quantile.(b.dist, x)
+
+logabsdetjac(b::QuantileBijector, x) = @. -Distributions.logpdf(b.dist, x)
+
+inverse(b::CDFBijector) = QuantileBijector(b.dist)
+inverse(b::QuantileBijector) = CDFBijector(b.dist)

src/interface.jl

@@ -249,6 +249,7 @@ include("bijectors/corr.jl")
 include("bijectors/truncated.jl")
 include("bijectors/named_bijector.jl")
 include("bijectors/ordered.jl")
+include("bijectors/cdf_quantile.jl")
 
 # Normalizing flow related
 include("bijectors/planar_layer.jl")