Laplace Approximation intermediates

[willtebbutt]

As noted in the code, this line makes quite strong assumptions about x

ApproximateGPs.jl/src/laplace.jl

Line 60 in bae0a1e

f = similar(xs, length(xs)) # will be mutated in-place to "warm-start" the Newton steps

.

When I went to use the laplace approximation code I initially stumbled around with it for a while, before realising I could just call build_laplace_objective!, and build f myself.

I'm not entirely sure what a proper fix that gives the correct initialisation all of the time would look like, but I think we can tell when it's definitely going to produce an error downstream, because I'm pretty sure we'll get an error if the element type of xs isn't a subtype of Real? Should we perhaps check this case, and provide a useful error message if it's hit?

@st-- ?

[st--]

Could you add a bit more context/details on what made it error ?

[willtebbutt]

The particular thing I was doing involved some stuff in Stheno.jl, but the issue would have occured if xs isa ColVecs. I'll produce an MWE.

[willtebbutt]

using AbstractGPs
using ApproximateGPs
using KernelFunctions

build_latent_gp() = LatentGP(GP(SEKernel()), BernoulliLikelihood(), 1e-8)

x = ColVecs(randn(2, 5))
_, y = rand(build_latent_gp()(x))

build_laplace_objective(build_latent_gp, x, y)()

yields

ERROR: MethodError: Cannot `convert` an object of type Float64 to an object of type SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}
Closest candidates are:
  convert(::Type{V}, ::CoordinateTransformations.Polar) where V<:(AbstractVector) at ~/.julia/packages/CoordinateTransformations/9po3r/src/coordinatesystems.jl:67
  convert(::Type{V}, ::CoordinateTransformations.Cylindrical) where V<:(AbstractVector) at ~/.julia/packages/CoordinateTransformations/9po3r/src/coordinatesystems.jl:270
  convert(::Type{V}, ::CoordinateTransformations.Spherical) where V<:(AbstractVector) at ~/.julia/packages/CoordinateTransformations/9po3r/src/coordinatesystems.jl:269
  ...
Stacktrace:
  [1] setindex!(A::Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, x::Float64, i1::Int64)
    @ Base ./array.jl:903
  [2] _unsafe_copyto!(dest::Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, doffs::Int64, src::Vector{Float64}, soffs::Int64, n::Int64)
    @ Base ./array.jl:253
  [3] unsafe_copyto!
    @ ./array.jl:307 [inlined]
  [4] _copyto_impl!
    @ ./array.jl:331 [inlined]
  [5] copyto!
    @ ./array.jl:317 [inlined]
  [6] copyto!
    @ ./array.jl:343 [inlined]
  [7] copyto!
    @ ./broadcast.jl:954 [inlined]
  [8] copyto!
    @ ./broadcast.jl:913 [inlined]
  [9] materialize!
    @ ./broadcast.jl:871 [inlined]
 [10] materialize!
    @ ./broadcast.jl:868 [inlined]
 [11] (::ApproximateGPs.var"#13#16"{Tuple{}, AbstractGPs.LatentFiniteGP{AbstractGPs.FiniteGP{GP{AbstractGPs.ZeroMean{Float64}, SqExponentialKernel{Distances.Euclidean}}, ColVecs{Float64, Matrix{Float64}, SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, Diagonal{Float64, FillArrays.Fill{Float64, 1, Tuple{Base.OneTo{Int64}}}}}, BernoulliLikelihood{LogisticLink}}, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}})()
    @ ApproximateGPs ~/.julia/packages/ApproximateGPs/XL25n/src/laplace.jl:83
 [12] ignore_derivatives
    @ ~/.julia/packages/ChainRulesCore/uxrij/src/ignore_derivatives.jl:26 [inlined]
 [13] (::ApproximateGPs.var"#objective#15"{Bool, Nothing, Int64, Vector{SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, typeof(build_latent_gp), ColVecs{Float64, Matrix{Float64}, SubArray{Float64, 1, Matrix{Float64}, Tuple{Base.Slice{Base.OneTo{Int64}}, Int64}, true}}, Vector{Bool}})()
    @ ApproximateGPs ~/.julia/packages/ApproximateGPs/XL25n/src/laplace.jl:78
 [14] top-level scope
    @ REPL[80]:1