[
https://issues.apache.org/jira/browse/MAHOUT-1508?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13967465#comment-13967465
]
Sebastian Schelter commented on MAHOUT-1508:
--------------------------------------------
Dmitriy, I agree with your comment in general. I'd like to propose a few minor
immediate changes to help me do efficient summation of outer product matrices
in MAHOUT-1464 :
* _matrixLike()_ in our sparse vectors should return a _SparseMatrix_ not
return a _SparseRowMatrix_ as its mainly used for outer products which produce
many zero rows
* _MatrixWritable_ should preserve the type of sparse matrix (currently it
always deserializes as _SparseRowMatrix_)
* provide an optimized assign for addition in _SparseMatrix_
> Performance problems with sparse matrices
> -----------------------------------------
>
> Key: MAHOUT-1508
> URL: https://issues.apache.org/jira/browse/MAHOUT-1508
> Project: Mahout
> Issue Type: Bug
> Components: Math
> Reporter: Sebastian Schelter
> Priority: Blocker
> Fix For: 1.0
>
>
> I'm currently working with the new Scala DSL and running into a problem with
> SparseMatrix and SparseRowMatrix.
> The problem is that they don't implement a specialized assign(Matrix other,
> DoubleDoubleFunction f) function, but use the implementation from
> AbstractMatrix which loops through all (!) possible entries.
> We have to fix this asap.
--
This message was sent by Atlassian JIRA
(v6.2#6252)
