| Re: [eigen] Release plan ("bug" 279) |
[ Thread Index |
Date Index
| More lists.tuxfamily.org/eigen Archives
]
Thanks for letting me know. In a month or two,
I might have the time and more importantly the expertise
to try to tackle this myself.
For everyone else, 279 is a feature request for
mixed scalar type operations, not a bug.
I coded around the problem for the first version
of our MCMC application by promoting double-based
matrices to auto-dif variables. That's a small
constant multiple of inefficiency in space and
time (about a factor of 10 for speed and a
factor of 3 for memory).
Eigen on pure auto-dif variable matrices
works great -- I tested everything we're going
to use (multiplication, addition, negation,
inverses, determinants, Cholesky and eigen
decompositions), and they all supply the textbook
answers for gradients.
Our auto-dif's standalone, as is the integration
with Eigen, so I'll post to the list when
we release it.
- Bob
On 8/16/11 2:57 AM, Gael Guennebaud wrote:
On Thu, Aug 11, 2011 at 7:18 PM, Bob Carpenter<carp@xxxxxxxxxxx> wrote:
Is there any plan to tackle Bug 279,
about mixing scalar types in operations
like matrix product? I still need this
for efficient reverse-mode algorithmic
differentiation.
sorry but no progress yet regarding this aspect. My focus is currently
on finalizing the Sparse module.
gael