[jira] [Commented] (MATH-1390) LUDecomposition slow on larger matrices

["Wilhelm Burger (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/MATH-1390?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15697660#comment-15697660 ] 

Wilhelm Burger commented on MATH-1390:
--------------------------------------

A proposed fix has been placed here: https://github.com/WilhelmBurger/commons-math/tree/wilbur

--Wilhelm

> LUDecomposition slow on larger matrices
> ---------------------------------------
>
>                 Key: MATH-1390
>                 URL: https://issues.apache.org/jira/browse/MATH-1390
>             Project: Commons Math
>          Issue Type: Improvement
>    Affects Versions: 4.0, 3.7
>            Reporter: Wilhelm Burger
>            Priority: Minor
>              Labels: performance
>             Fix For: 4.0, 3.7
>
>   Original Estimate: 6h
>  Remaining Estimate: 6h
>
> Commons-math LUDecomposition runs ca. 10 times slower than the original (referenced) JAMA implementation for a matrix of size 1600 x 1600, slowdown is noticable above size 750 x 750, regardless of matrix type.
> It is suggested to revert to the proven (and fast) JAMA implementation and overhaul the LUDecomposition class (and related classes) in the 'linear' package.



--
This message was sent by Atlassian JIRA
(v6.3.4#6332)