Edit on GitHub

Univariate ADVI example

But the real utility of TransformedDistribution becomes more apparent when using transformed(dist, b) for any bijector b. To get the transformed distribution corresponding to the Beta(2, 2), we called transformed(dist) before. This is simply an alias for transformed(dist, bijector(dist)). Remember bijector(dist) returns the constrained-to-constrained bijector for that particular Distribution. But we can of course construct a TransformedDistribution using different bijectors with the same dist. This is particularly useful in something called Automatic Differentiation Variational Inference (ADVI).[2] An important part of ADVI is to approximate a constrained distribution, e.g. Beta, as follows:

  1. Sample x from a Normal with parameters μ and σ, i.e. x ~ Normal(μ, σ).
  2. Transform x to y s.t. y ∈ support(Beta), with the transform being a differentiable bijection with a differentiable inverse (a "bijector")

This then defines a probability density with same support as Beta! Of course, it's unlikely that it will be the same density, but it's an approximation. Creating such a distribution becomes trivial with Bijector and TransformedDistribution:

julia> using StableRNGs: StableRNG
julia> rng = StableRNG(42);
julia> dist = Beta(2, 2)Beta{Float64}(α=2.0, β=2.0)
julia> b = bijector(dist)              # (0, 1) → ℝBijectors.Logit{Float64, Float64}(0.0, 1.0)
julia> b⁻¹ = inverse(b)                # ℝ → (0, 1)Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0))
julia> td = transformed(Normal(), b⁻¹) # x ∼ 𝓝(0, 1) then b(x) ∈ (0, 1)UnivariateTransformed{Normal{Float64}, Inverse{Bijectors.Logit{Float64, Float64}}}(
dist: Normal{Float64}(μ=0.0, σ=1.0)
transform: Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0))
)
julia> x = rand(rng, td)                   # ∈ (0, 1)0.3384404850130036

It's worth noting that support(Beta) is the closed interval [0, 1], while the constrained-to-unconstrained bijection, Logit in this case, is only well-defined as a map (0, 1) → ℝ for the open interval (0, 1). This is of course not an implementation detail. ℝ is itself open, thus no continuous bijection exists from a closed interval to ℝ. But since the boundaries of a closed interval has what's known as measure zero, this doesn't end up affecting the resulting density with support on the entire real line. In practice, this means that

julia> td = transformed(Beta())UnivariateTransformed{Beta{Float64}, Bijectors.Logit{Float64, Float64}}(
dist: Beta{Float64}(α=1.0, β=1.0)
transform: Bijectors.Logit{Float64, Float64}(0.0, 1.0)
)
julia> inverse(td.transform)(rand(rng, td))0.8130302707446476

will never result in 0 or 1 though any sample arbitrarily close to either 0 or 1 is possible. Disclaimer: numerical accuracy is limited, so you might still see 0 and 1 if you're lucky.

Multivariate ADVI example

We can also do multivariate ADVI using the Stacked bijector. Stacked gives us a way to combine univariate and/or multivariate bijectors into a singe multivariate bijector. Say you have a vector x of length 2 and you want to transform the first entry using Exp and the second entry using Log. Stacked gives you an easy and efficient way of representing such a bijector.

julia> using Bijectors: SimplexBijector
       
       # Original distributions
julia> dists = (Beta(), InverseGamma(), Dirichlet(2, 3));
       
       # Construct the corresponding ranges
julia> ranges = [];
julia> idx = 1;
julia> for i in 1:length(dists)
           d = dists[i]
           push!(ranges, idx:(idx + length(d) - 1))
       
           global idx
           idx += length(d)
       end;
julia> ranges
       
       # Base distribution; mean-field normal3-element Vector{Any}:
 1:1
 2:2
 3:4
julia> num_params = ranges[end][end]4
julia> d = MvNormal(zeros(num_params), ones(num_params));
       
       # Construct the transform
julia> bs = bijector.(dists);     # constrained-to-unconstrained bijectors for dists
julia> ibs = inverse.(bs);            # invert, so we get unconstrained-to-constrained
julia> sb = Stacked(ibs, ranges) # => Stacked <: Bijector
       
       # Mean-field normal with unconstrained-to-constrained stacked bijectorStacked(Any[Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0)), Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp), Inverse{Bijectors.SimplexBijector}(Bijectors.SimplexBijector())], Any[1:1, 2:2, 3:4], Any[1:1, 2:2, 3:5])
julia> td = transformed(d, sb);
julia> y = rand(td)5-element Vector{Float64}:
 0.17841505686495443
 1.5336323424298905
 0.7156704776573194
 0.04085757365673251
 0.24347194868594813
julia> 0.0 ≤ y[1] ≤ 1.0true
julia> 0.0 < y[2]true
julia> sum(y[3:4]) ≈ 1.0false

Normalizing flows

A very interesting application is that of normalizing flows.[1] Usually this is done by sampling from a multivariate normal distribution, and then transforming this to a target distribution using invertible neural networks. Currently there are two such transforms available in Bijectors.jl: PlanarLayer and RadialLayer. Let's create a flow with a single PlanarLayer:

julia> d = MvNormal(zeros(2), ones(2));
julia> b = PlanarLayer(2)PlanarLayer(w = [-1.0408637312049929, 2.235251162004739], u = [-0.24164166198564066, 1.652917908177465], b = [0.12060496907886155])
julia> flow = transformed(d, b)MultivariateTransformed{DiagNormal, PlanarLayer{Vector{Float64}, Vector{Float64}}}(
dist: DiagNormal(
dim: 2
μ: [0.0, 0.0]
Σ: [1.0 0.0; 0.0 1.0]
)

transform: PlanarLayer(w = [-1.0408637312049929, 2.235251162004739], u = [-0.24164166198564066, 1.652917908177465], b = [0.12060496907886155])
)
julia> flow isa MultivariateDistributiontrue

That's it. Now we can sample from it using rand and compute the logpdf, like any other Distribution.

julia> y = rand(rng, flow)2-element Vector{Float64}:
 -0.7401802806074489
  1.6730062227158746
julia> logpdf(flow, y)         # uses inverse of `b`-2.4224845733786786

Similarily to the multivariate ADVI example, we could use Stacked to get a bounded flow:

julia> d = MvNormal(zeros(2), ones(2));
julia> ibs = inverse.(bijector.((InverseGamma(2, 3), Beta())));
julia> sb = Stacked(ibs) # == Stacked(ibs, [i:i for i = 1:length(ibs)]Stacked((Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp), Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0))), (1:1, 2:2), (1:1, 2:2))
julia> b = sb ∘ PlanarLayer(2)Stacked((Base.Fix1{typeof(broadcast), typeof(exp)}(broadcast, exp), Inverse{Bijectors.Logit{Float64, Float64}}(Bijectors.Logit{Float64, Float64}(0.0, 1.0))), (1:1, 2:2), (1:1, 2:2)) ∘ PlanarLayer(w = [0.2220050576914493, -0.1436477237262901], u = [-1.4152050850988749, 0.17109500087483392], b = [-0.714064037930183])
julia> td = transformed(d, b);
julia> y = rand(rng, td)2-element Vector{Float64}:
 10.38293034211274
  0.7469793180100731
julia> 0 < y[1]true
julia> 0 ≤ y[2] ≤ 1true

Want to fit the flow?

julia> using Zygote
       
       # Construct the flow.
julia> b = PlanarLayer(2)
       
       # Convenient for extracting parameters and reconstructing the flow.PlanarLayer(w = [1.7754907696739541, -0.5938276464383798], u = [1.3385218771986869, 2.185666507949883], b = [0.23493273069341675])
julia> using Functors
julia> θs, reconstruct = Functors.functor(b);
       
       # Make the objective a `struct` to avoid capturing global variables.
julia> struct NLLObjective{R,D,T}
           reconstruct::R
           basedist::D
           data::T
       end
julia> function (obj::NLLObjective)(θs)
           transformed_dist = transformed(obj.basedist, obj.reconstruct(θs))
           return -sum(Base.Fix1(logpdf, transformed_dist), eachcol(obj.data))
       end
       
       # Some random data to estimate the density of.
julia> xs = randn(2, 1000);
       
       # Construct the objective.
julia> f = NLLObjective(reconstruct, MvNormal(2, 1), xs);
       
       # Initial loss.
julia> @info "Initial loss: $(f(θs))"
       
       # Train using gradient descent.[ Info: Initial loss: 4518.38908626241
julia> ε = 1e-3;
julia> for i in 1:100
           (∇s,) = Zygote.gradient(f, θs)
           θs = fmap(θs, ∇s) do θ, ∇
               θ - ε .* ∇
           end
       end
       
       # Final loss
julia> @info "Final loss: $(f(θs))"
       
       # Very simple check to see if we learned something useful.[ Info: Final loss: 2862.6025267572877
julia> samples = rand(transformed(f.basedist, f.reconstruct(θs)), 1000);
julia> mean(eachcol(samples)) # ≈ [0, 0]2-element Vector{Float64}:
 -0.08048446136305033
  0.03643599036394901
julia> cov(samples; dims=2)   # ≈ I2×2 Matrix{Float64}:
 0.991613   0.0192021
 0.0192021  1.02701

We can easily create more complex flows by simply doing PlanarLayer(10) ∘ PlanarLayer(10) ∘ RadialLayer(10) and so on.