Add type conversion for MLD calls

Calling a MLD with a different type of vector than what it was
defined with caused ForwardDiff tagging errors, due to the DI
preparation mechanism. Added a `covert` to prevent this.

Author: ElOceanografo

src/MarginalLogDensities.jl

@@ -224,7 +224,8 @@ function Base.show(io::IO, mld::MarginalLogDensity)
 end
 
 function (mld::MarginalLogDensity)(v::AbstractVector{T}, data=mld.data; verbose=false) where T
-    return _marginalize(mld, v, data, mld.method, verbose)
+    v1 = convert.(eltype(mld.u), v)
+    return _marginalize(mld, v1, data, mld.method, verbose)
 end
 
 

test/runtests.jl

@@ -149,6 +149,11 @@ end
     @test logpdf_laplace >= mld_laplace.logdensity(x, ())
     @test logpdf_cubature1 >= mld_cubature1.logdensity(x, ())
     @test logpdf_cubature2 >= mld_cubature2.logdensity(x, ())
+
+    # test for correct type promotion
+    v_int = ones(Int, njoint(mld_laplace))
+    v_float = ones(Float64, njoint(mld_laplace))
+    @test mld_laplace(v_int) == mld_laplace(v_float)
 end