|Summary:||Add view to nonZeros of SparseMatrix|
|Severity:||Feature Request||CC:||chtz, darcy, gael.guennebaud|
|Bug Depends on:|
Description Darcy 2016-08-05 03:48:23 UTC
DESIRED FEATURE: The ability to call a method that zeros the coefficients in a sparse matrix. Currently the setZero() method zeros all allocated memory but it would be nice to set the Aij coefficients to zero, while keeping the row and column arrays allocated and populated.
Comment 1 Gael Guennebaud 2016-08-22 09:04:20 UTC
Exposing a dense "view" on the non-zeros would solve this issue, e.g.: mat.nonZerosView().setZero(); while enabling many more efficient manipulations on the non zero coefficients. I've suggested this solution several times, e.g. in bug 1104. Implementing it for compressed matrices is just a matter of returning a 1D dense Eigen::Map. For un-compressed ones, this is much more tricky because we would need to return a list of dense Map for which random access would be extremely inefficient.
Comment 2 Christoph Hertzberg 2016-08-23 12:29:34 UTC
I like the idea of providing a nonZerosView(). If it is to complicated for un-compressed matrices, I think simply restricting this feature to compressed matrices would be ok. Minor question: Should the return value be a vector or an array? I'm tending towards array, since nonZerosView() would behave more like independent scalars.
Comment 3 Gael Guennebaud 2016-08-23 14:24:49 UTC
I agree with the two suggestions (compressed-only + array).
Comment 4 Gael Guennebaud 2016-08-29 10:08:38 UTC
Done for compressed mode: https://bitbucket.org/eigen/eigen/commits/614aeef Supporting non compressed mode would require a more general expression to concatenate an arbitrary number of vectors/matrices.
Comment 5 Darcy 2016-08-29 10:14:21 UTC
Fantastic, suits my purposes. Many thanks!
Comment 6 Nobody 2019-12-04 16:05:05 UTC
-- GitLab Migration Automatic Message -- This bug has been migrated to gitlab.com's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.com/libeigen/eigen/issues/1271.