-
Notifications
You must be signed in to change notification settings - Fork 32
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Implicit gradient failing with matrices #137
Comments
This is outside my area of expertise I'm afraid. |
I think there is a general issue with the adjoint of |
Have you tried wrapping everything in a edit: I'm not sure exactly how the |
You mean this ?
|
Nah, just using KernelFunctions, Flux, LinearAlgebra
let
k = transform(SqExponentialKernel(), 2.0)
ps = Flux.params(k)
X = rand(10, 1); x = vec(X)
A = rand(10, 10)
g = gradient(ps) do
tr(kernelmatrix(k, X, obsdim = 1) * A)
end
g[ps[1]] == nothing
g2 = gradient(k) do k
tr(kernelmatrix(k, X, obsdim = 1) * A)
end
g2[1].transform.s != nothing
g3 = gradient(ps) do
tr(kernelmatrix(k, x) * A)
end
g3[ps[1]] != nothing
end |
Nope same behavior |
I found a fix \o/ ! I think we should avoid relying on |
Here is a MWE:
I think this is related to FluxML/Zygote.jl#692
Any idea on how to solve this @willtebbutt ? It is probably connected to the
ColVecs
structureThe text was updated successfully, but these errors were encountered: