-
Notifications
You must be signed in to change notification settings - Fork 34
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
Autodiff broken in latest release #67
Comments
Maybe you forgot to add a chain rule for that constructor? |
Ah, I see what the problem is. I have a rule ChainRulesCore.rrule(::typeof(ComponentArray), data, axes) = ComponentArray(data, axes), Δ->(getdata(Δ), getaxes(Δ)) because that is how you'd normally define a rule for a function. But since Unfortunately I'm a little busy tonight to look into how I'm supposed to handle this tonight, so the fix won't come until probably tomorrow night. Hopefully it's something easy, but looking through the ChainRules docs, I don't see how to handle it just yet. Maybe it doesn't even need that rule. |
Nevermind, I think I fixed it. I just deleted that rule. Seems to work fine now, I think. At least it's completing without an error. If you want to get the fix now without waiting for it to be registered, just do: julia> m = @which ComponentArrays.ChainRulesCore.rrule(ComponentArray, identity, [1.0]);
julia> Base.delete_method(m); |
It doesn't return a |
I don't quite understand what you mean. The type of the gradient isn't a |
The |
If you mean |
Turned out to be a pretty dumb mistake. I should have done |
v0.8.20 seems to have broken autodiff code that worked fine in v0.8.19.
MWE (excuse the clutter)
Error message:
The text was updated successfully, but these errors were encountered: