[jira] [Commented] (MAHOUT-1907) Correctly implement VCL vector %*% matrix multiplication.

["Andrew Palumbo (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MAHOUT-1907?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15884927#comment-15884927 ] 

Andrew Palumbo commented on MAHOUT-1907:
----------------------------------------

Seeing some small Numerical differences here when operations are on GPU vs OMP/CPU or JVM/CPU.
"fixed" by relaxing the epsilon to 1e-10 from 1e-16;  though this may affect highly iterative algorithms.:

{code}
+ ViennaCL/cpu/OpenMP Sparse multiplication time: 1577 ms.
- VCL Dense Matrix %*% Dense vector *** FAILED ***
 1.5059797815742444E-16 was not less than 1.0E-16 (ViennaCLSuiteVCL.scala:349)
{code} 

> Correctly implement VCL vector %*% matrix multiplication.
> ---------------------------------------------------------
>
>                 Key: MAHOUT-1907
>                 URL: https://issues.apache.org/jira/browse/MAHOUT-1907
>             Project: Mahout
>          Issue Type: Task
>    Affects Versions: 0.12.2
>            Reporter: Andrew Palumbo
>            Assignee: Andrew Palumbo
>            Priority: Blocker
>             Fix For: 0.13.0
>
>
> Vector/Matrix Multiplication is currently commented out in the VCL PR.  after much discussion have found the Bug, will implement it after VCL is pushed.



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)