[ https://issues.apache.org/jira/browse/MATH-924?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gilles updated MATH-924:
------------------------
Attachment: MATH-924
Keeping the matrix concept, by representing uncorrelated observations with a diagonal weight
matrix (instead of an array), allows to solve this issue with minimal changes to the code:
# The optimizer's API is untouched.
# The possibility to have correlated observations is kept.
The attached patch contains a minimal "DiagonalMatrix" implementation, needing overview (ant
unit tests).
> new multivariate vector optimizers cannot be used with large number of weights
> ------------------------------------------------------------------------------
>
> Key: MATH-924
> URL: https://issues.apache.org/jira/browse/MATH-924
> Project: Commons Math
> Issue Type: Bug
> Reporter: Luc Maisonobe
> Priority: Critical
> Fix For: 3.1.1
>
> Attachments: MATH-924
>
>
> When using the Weigth class to pass a large number of weights to multivariate vector
optimizers, an nxn full matrix is created (and copied) when a n elements vector is used. This
exhausts memory when n is large.
> This happens for example when using curve fitters (even simple curve fitters like polynomial
ones for low degree) with large number of points. I encountered this with curve fitting on
41200 points, which created a matrix with 1.7 billion elements.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
|