[
https://issues.apache.org/jira/browse/SPARK-17824?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15555448#comment-15555448
]
Yanbo Liang edited comment on SPARK-17824 at 10/7/16 3:53 PM:
--------------------------------------------------------------
[~sethah]
I saw your proposal at SPARK-17748:
{code}
class CholeskySolver extends NormalEquationsSolver
class QuasiNewtonSolver extends NormalEquationSolver
{code}
I think this JIRA is used to add another solver:
{code}
class QRSolver extends NormalEquationsSolver
{code}
The difference is that both {{CholeskySolver}} and {{QRSolver}} are used to
solve normal equation with L2 regularization by LAPACK route; however,
{{QuasiNewtowSolver}} is used to solve normal equation with L2 and
L1/elasticnet penalties by local LBFGS and OWLQN respectively. I think the
purpose of these solvers may overlap, but they may have different performance
and feasibility. Please correct me if I have misunderstand. Thanks!
was (Author: yanboliang):
[~sethah]
I saw your proposal at SPARK-17748:
{code}
class CholeskySolver extends NormalEquationsSolver
class QuasiNewtonSolver extends NormalEquationSolver
{code}
I think this JIRA is used to add another solver:
{{class QRSolver extends NormalEquationsSolver}}
The difference is that both {{CholeskySolver}} and {{QRSolver}} are used to
solve normal equation with L2 regularization by LAPACK route; however,
{{QuasiNewtowSolver}} is used to solve normal equation with L2 and
L1/elasticnet penalties by local LBFGS and OWLQN respectively. I think the
purpose of these solvers may overlap, but they may have different performance
and feasibility. Please correct me if I have misunderstand. Thanks!
> QR solver for WeightedLeastSquares
> ----------------------------------
>
> Key: SPARK-17824
> URL: https://issues.apache.org/jira/browse/SPARK-17824
> Project: Spark
> Issue Type: New Feature
> Components: ML
> Reporter: Yanbo Liang
> Assignee: Yanbo Liang
>
> Cholesky decomposition is unstable (for near-singular and rank deficient
> matrices) and only works on positive definite matrices which can not be
> guaranteed in all cases, it was often used when matrix A is very large and
> sparse due to faster calculation. QR decomposition has better numerical
> properties than Cholesky and can works on matrices which are not positive
> definite. Spark MLlib {{WeightedLeastSquares}} use Cholesky decomposition to
> solve normal equation currently, we should also support or move to QR solver
> for better stability. I'm preparing to send a PR.
> cc [~dbtsai] [~sethah]
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
