Tensorflow works around this limitation by "reinterpreting" memory buffers as Eigen Tensors of different rank at runtime: https://github.com/tensorflow/tensorflow/blob/master/tensorflow/core/framework/tensor.h#L414-L442
On Tue, Oct 29, 2019 at 10:33 AM Rasmus Munk Larsen <[email protected]> wrote: > This was an early design decision. Benoit Steiner did have some plans to > add support for dynamic tensors, but it was never implemented. There are no > current plans to add this feature, but you are of course welcome to > contribute it. > > Rasmus > > On Tue, Oct 29, 2019 at 10:23 AM William Tambellini <[email protected]> > wrote: > >> Hello Tom >> I would also favor ranking to be a runtime thing : that s also the choice >> made by xtensor and ArrayFire : >> https://xtensor.readthedocs.io/en/latest/index.html >> http://arrayfire.org/docs/gettingstarted.htm >> Now, are eigen::tensor imposing ranking to be template/compil-time for >> performance reasons ? >> Regards >> W. >> >> <http://bit.ly/2LB1qRN> >> *Share yourfeedback with us* <https://www.surveymonkey.com/r/PYF190816> >> >> ------------------------------ >> *From:* Peterka, Thomas <[email protected]> >> *Sent:* Friday, October 25, 2019 9:12 AM >> *To:* [email protected] <[email protected]> >> *Subject:* [eigen] Eigen tensor module >> >> Hello, >> >> Is there any way to give the rank of an Eigen tensor at run time, or are >> there any plans to make this a dynamic run-time parameter? The tensor >> module would be extremely useful for me if not for the limitation that the >> rank needs to be specified as compile-time template parameter. I am unable >> to use Eigen tensors for a generic n-dimensional application where I don’t >> know the tensor rank in advance. >> >> Please let me know if I’m missing something or if there are any future >> plans to change this? The boost tensor module, for example, has dynamic >> ranks and sizes: >> >> https://www.boost.org/doc/libs/develop/libs/numeric/ublas/doc/tensor.html >> >> >> Thanks, >> Tom >> ----------------- >> Tom Peterka >> Mathematics and Computer Science Division >> Argonne National Laboratory >> 9700 S. Cass Ave. >> Lemont, IL 60439 >> Bldg 240, Rm 2148 >> e-mail: [email protected] <[email protected]> >> web: www.mcs.anl.gov/~tpeterka >> phone: 630-252-7198 <(630)%20252-7198> >> >> >> >> Click here >> <https://www.mailcontrol.com/sr/5cD8Oxo38DjGX2PQPOmvUsrLibhXE7-SS2I_aEdfuQBxshAU2gYXIumlly3wfsiivaKfZpb_jEqCTIJSz5pNWQ==> >> to report this email as spam. >> >
