Polling Error

[michael....@gmail.com]

I keep getting polling errors, if i increase nfolds above 2.
I get this for GBM, RF and DL.

Its for the Kaggle malware contest using multiclass classification

Any idea?


Using version 1722
Java 1.8


> cvmodel <- h2o.gbm(data=train.h2o,
+ nfolds = 3,
+ x=c(2:ncol(train.h2o) -1),
+ y=ncol(train.h2o),
+ n.trees = 400,
+ interaction.depth = 20,
+ shrinkage = 0.01,
+ grid.parallelism = 4)
| | 0%
Polling fails:
<simpleError in .h2o.__poll(client, job_key): Got exception 'class java.lang.AssertionError', with msg 'null'
java.lang.AssertionError
at hex.gbm.DRealHistogram.scoreMSE(DRealHistogram.java:106)
at hex.gbm.GBM$GBMDecidedNode.bestCol(GBM.java:637)
at hex.gbm.DTree$DecidedNode.<init>(DTree.java:387)
at hex.gbm.GBM$GBMDecidedNode.<init>(GBM.java:625)
at hex.gbm.GBM.makeDecided(GBM.java:617)
at hex.gbm.SharedTreeModelBuilder$ScoreBuildOneTree.onCompletion(SharedTreeModelBuilder.java:691)
at jsr166y.CountedCompleter.tryComplete(CountedCompleter.java:386)
at water.MRTask2.compute2(MRTask2.java:426)
at water.H2O$H2OCountedCompleter.compute(H2O.java:656)
at jsr166y.CountedCompleter.exec(CountedCompleter.java:429)
at jsr166y.ForkJoinTask.doExec(ForkJoinTask.java:263)
at jsr166y.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:974)
at jsr166y.ForkJoinPool.runWorker(ForkJoinPool.java:1477)
at jsr166y.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:104)
>

[Tom Kraljevic]

Hi Michael,


This definitely looks like a bug.
Since engineering focus is now on h2o-dev, that’s probably where it will get fixed.


Thanks,
Tom

> --
> You received this message because you are subscribed to the Google Groups "H2O & Open Source Scalable Machine Learning - h2ostream" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to h2ostream+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[ccl...@gmail.com]

This is an edge-case assert, for dealing with errors less than 1 FP ULP.
You can safely turn off this assert, by launching a JVM without the "-ea" flag.
As Tom has mentioned, this issue was seen & fixed in h2o-dev already.

Cliff

[michael....@gmail.com]

Thanks,

Any idea when this will be pushed the R library h2o?
Went down the build process but have java 1.8 not 1.7 and get build failures for testMultiNode

Execution failed for task ':h2o-core:testMultiNode'.

Really dont want to spend a lot of time getting java 1.7 back on the machine for this and then figuring out how everything connects