On Wed, Dec 3, 2014 at 10:26 AM, Alexandre Gramfort <
[email protected]> wrote:
> > A compromise would be to just implement the Cython routine in a separate
> > file, while sharing the same file for the pure Python side.
>
> sounds reasonable.
>
> > That said, using a separate class for A
> A compromise would be to just implement the Cython routine in a separate
> file, while sharing the same file for the pure Python side.
sounds reasonable.
> That said, using a separate class for Adagrad would allow to get rid of
> irrelevant hyper-parameters.
+1
> Some code from the SGD module
A compromise would be to just implement the Cython routine in a separate
file, while sharing the same file for the pure Python side.
That said, using a separate class for Adagrad would allow to get rid of
irrelevant hyper-parameters. Some code from the SGD module can probably be
factorized and OVR
FYI Mathieu and myself converged on this with Danny, believing it's
the easiest move.
It could later be improved and extended with schemes that also use feature
specific learning rates.
any thought?
Alex
On Tue, Dec 2, 2014 at 2:48 PM, Daniel Sullivan wrote:
> Hey All,
>
> I've been looking a
Hey All,
I've been looking at adding Adagrad to SGD for a while now. Alex, Mathieu
and I were discussing the possibility of having a separate class for
Adagrad entirely. The benefits of this would be that the implementation of
SGD would not get muddled up and Adagrad could be implemented in a much
