Skip to content

Conversation

@mlubin
Copy link
Collaborator

@mlubin mlubin commented Sep 2, 2016

@giordano, these ended up breaking JuMP because we symbolically differentiate the list of derivatives to get 2nd order derivatives, and Calculus doesn't know how to differentiate oftype. We use pi in a number of other places without oftype, so I'm going to back out this change until there's a more satisfying way to handle this case.

@giordano
Copy link
Member

giordano commented Sep 3, 2016

I'm sorry for having broken a downstream project, but all these derivatives are essentially untested, it's easy to break something without noticing it. In any case, I'm fine with this temporary solution.

As a definitive solution, how about setting the gradient of oftype(x, y) to (zero(x), one(x))?

@mlubin
Copy link
Collaborator Author

mlubin commented Sep 3, 2016

@giordano, that would work except for the missing infrastructure to differentiate two-argument functions

@giordano
Copy link
Member

giordano commented Sep 3, 2016

except for the missing infrastructure to differentiate two-argument functions

Yes, that's unfortunate.

@mlubin mlubin merged commit ae95d7f into master Sep 3, 2016
@mlubin mlubin deleted the ml/nooftype branch September 3, 2016 00:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants