This failure is in the scala tests. I have verified separately that the
sparse matrix multiplication produces bit for bit identical results, but
the test run saw this:
ALS factorized approximation block:
> {
> 0 => {0:0.3947786998258809,1:-0.08698336365132517,2:-1.057510528973319}
> 1 => {0:0.4076938288168168,1:0.013542226762735074,2:-0.6051700375781449}
> 2 => {0:0.15933739918843964,1:0.0746873472040897,2:-0.3796424489752668}
> }
> norm of residuals 0.029200
> train iteration rmses: List(1.0176360967983437E-6, 1.7052342728874502E-6)
> [31m- als *** FAILED *** [0m
> [31m 0.029199608498124176 was not less than 0.01 (MathSuite.scala:208) [0m
> log4j:WARN No appenders could be found for logger
> (akka.remote.RemoteActorRefProvider$RemotingTerminator).
I don't see this problem running locally.
Is this a non-deterministic test?
On Sat, Jun 7, 2014 at 3:34 PM, Hudson (JIRA) <[email protected]> wrote:
>
> [
> https://issues.apache.org/jira/browse/MAHOUT-1575?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14021002#comment-14021002
> ]
>
> Hudson commented on MAHOUT-1575:
> --------------------------------
>
> FAILURE: Integrated in Mahout-Quality #2640 (See [
> https://builds.apache.org/job/Mahout-Quality/2640/])
> MAHOUT-1575 - Allow 2 extra passes for conjugate gradient (tdunning: rev
> 99fd9ba4f9673b2b2d30c2b403aaa294503cc2a2)
> *
> mrlegacy/src/main/java/org/apache/mahout/math/hadoop/solver/DistributedConjugateGradientSolver.java
> *
> math/src/main/java/org/apache/mahout/math/solver/ConjugateGradientSolver.java
>
>
> > Conjugate gradient assumes best case scenario for convergence
> > -------------------------------------------------------------
> >
> > Key: MAHOUT-1575
> > URL: https://issues.apache.org/jira/browse/MAHOUT-1575
> > Project: Mahout
> > Issue Type: Bug
> > Reporter: Ted Dunning
> >
> > The conjugate gradient solve that we have assumes that it will converge
> within the theoretical n steps where n is the size of the input. In fact,
> convergence may not happen within exactly that number of steps due to
> numerical issues so we should allow for a few extra iterations.
>
>
>
> --
> This message was sent by Atlassian JIRA
> (v6.2#6252)
>
