Make pullback error for ColVecs and RowVecs a bit more informative (#523

) * make adjoint error message a bit more informative * Apply suggestions from code review * Update Project.toml * Update src/chainrules.jl
JuliaGaussianProcesses · Sep 27, 2023 · cf937ce · cf937ce · willtebbutt · Sep 27, 2023
1 parent 5127a26
commit cf937ce
Show file tree

Hide file tree

Showing 2 changed files with 9 additions and 5 deletions.
diff --git a/Project.toml b/Project.toml
@@ -1,6 +1,6 @@
 name = "KernelFunctions"
 uuid = "ec8451be-7e33-11e9-00cf-bbf324bd1392"
-version = "0.10.56"
+version = "0.10.57"
 
 [deps]
 ChainRulesCore = "d360d2e6-b24c-11e9-a2a3-2a2ae2dbcce4"

diff --git a/src/chainrules.jl b/src/chainrules.jl
@@ -150,8 +150,11 @@ function ChainRulesCore.rrule(::Type{<:ColVecs}, X::AbstractMatrix)
     function ColVecs_pullback(::AbstractVector{<:AbstractVector{<:Real}})
         return error(
             "Pullback on AbstractVector{<:AbstractVector}.\n" *
-            "This might happen if you try to use gradients on the generic `kernelmatrix` or `kernelmatrix_diag`.\n" *
-            "To solve this issue overload `kernelmatrix(_diag)` for your kernel for `ColVecs`",
+            "This might happen if you try to use gradients on the generic `kernelmatrix` or `kernelmatrix_diag`,\n" *
+            "or because some external computation has acted on `ColVecs` to produce a vector of vectors." *
+            "In the former case, to solve this issue overload `kernelmatrix(_diag)` for your kernel for `ColVecs`." *
+            "In the latter case, one needs to track down the `rrule` whose pullback returns a `Vector{Vector{T}}`," *
+            " rather than a `Tangent`, as the cotangent / gradient for `ColVecs` input, and circumvent it."
         )
     end
     return ColVecs(X), ColVecs_pullback
@@ -162,8 +165,9 @@ function ChainRulesCore.rrule(::Type{<:RowVecs}, X::AbstractMatrix)
     function RowVecs_pullback(::AbstractVector{<:AbstractVector{<:Real}})
         return error(
             "Pullback on AbstractVector{<:AbstractVector}.\n" *
-            "This might happen if you try to use gradients on the generic `kernelmatrix` or `kernelmatrix_diag`.\n" *
-            "To solve this issue overload `kernelmatrix(_diag)` for your kernel for `RowVecs`",
+            "This might happen if you try to use gradients on the generic `kernelmatrix` or `kernelmatrix_diag`,\n" *
+            "or because some external computation has acted on `RowVecs` to produce a vector of vectors." *
+            "If it is the former, to solve this issue overload `kernelmatrix(_diag)` for your kernel for `RowVecs`",
         )
     end
     return RowVecs(X), RowVecs_pullback