It is just a reformulation and doesn't need any Convex.jl internals, but it is not super obvious how to do it (a direct scalar loopy approach would be terribly slow), and we support 1D convolutions so it seems like it is within scope.
I still need to write tests. Also it's not clear if this should just be a method to conv or a separate function. Adding a method definitely feels more Julian, but since we have the "treat everything like a matrix" matlab thing, I'd worry a little there could be weird interactions there.
closes #376
It is just a reformulation and doesn't need any Convex.jl internals, but it is not super obvious how to do it (a direct scalar loopy approach would be terribly slow), and we support 1D convolutions so it seems like it is within scope.
I still need to write tests. Also it's not clear if this should just be a method to
conv
or a separate function. Adding a method definitely feels more Julian, but since we have the "treat everything like a matrix" matlab thing, I'd worry a little there could be weird interactions there.