Null value for dataset when using /3/SplitFrame REST API

[KK]

I am trying to submit a REST request to split a frame from a Java client I wrote. All values other than dataset from the request seem to get submitted ok (ratios, destination_frames). I seem to have traced the issue down to the Schema.fillImpl(I impl) on line 451 (H2O v3.2.0.1). The Pojo.copyProperties can't copy the dataset value properly from SplitFrameV3 to SplitFrame. I suspect it's because the copyProperties method assumes a mirror from->to structure and in the source object's dataset property is nested while in the destination object the property is top level. Just a guess.

I also noticed the web UI accomplishes this in a completely different fashion and doesn't even call the SplitFrame API at all but rather uses the very cryptic /99/Rapids API. Are there some known problems here and the SplitFrames API isn't working?

Thanks for any help overcoming this.

KK

[SriSatish]

KK,

Rapids API is the backend for our R and Python interface - low level and not necessarily exposed.

Do you have log files / stack trace and maybe sample data & code that triggers this?

Thanks,
Sri

--
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.

[KK]

The stack trace I get in the job report is:

Got exception 'class java.lang.NullPointerException', with msg 'null' java.lang.NullPointerException at hex.FrameSplitter.compute2(FrameSplitter.java:64) at water.H2O$H2OCountedCompleter.compute(H2O.java:1017) at jsr166y.CountedCompleter.exec(CountedCompleter.java:429) at jsr166y.ForkJoinTask.doExec(ForkJoinTask.java:263) at jsr166y.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:974) at jsr166y.ForkJoinPool.runWorker(ForkJoinPool.java:1477) at jsr166y.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:104)

The project is actually in my GitHub repo:

https://github.com/kkarski/h2o-java-rest-client

You can execute the following unit test assuming you have a h2o instance running under localhost:54321

https://github.com/kkarski/h2o-java-rest-client/blob/master/src/test/java/water/bindings/CreateFrameV3ITCase.java

[Sri Ambati]

Awesome - this is useful!

Thanks, Sri

Sent from my iPhone

[KK]

Any luck finding a culprit?

[rp...@0xdata.com]

On Wednesday, October 28, 2015 at 9:10:51 AM UTC-7, KK wrote:

Any luck finding a culprit?

Sorry for the delay.  QA is working on it as we speak.

 

[rp...@0xdata.com]

On Wednesday, October 28, 2015 at 12:42:46 PM UTC-7, rpeck@... wrote:

On Wednesday, October 28, 2015 at 9:10:51 AM UTC-7, KK wrote:

Any luck finding a culprit?

The Rapids expression syntax has been going through changes recently.

Will it be possible for you to move forward to 3.2.0.9?  If so, we'll get this working for you on 3.2.0.9 and send you the updated code.

Thanks!

 

[KK]

Downloaded 3.2.0.9 and tried to start standalone, got the following error:

java -Xmx1g -jar h2o.jar 

Exception in thread "main" java.lang.RuntimeException: Selected H2O.CLOUD_MULTICAST_IF: name:docker0 (docker0) is not up and running

at water.H2ONode.self(H2ONode.java:189)

at water.init.NetworkInit.initializeNetworkSockets(NetworkInit.java:383)

at water.H2O.startLocalNode(H2O.java:1210)

at water.H2O.main(H2O.java:1611)

at water.H2OStarter.start(H2OStarter.java:18)

at water.H2OApp.main(H2OApp.java:5)

[Krzysztof Karski]

Downgraded to 3.2.0.5 and both the split frame and rapids approach seem to work now.

--
You received this message because you are subscribed to a topic in the Google Groups "H2O Open Source Scalable Machine Learning - h2ostream" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/h2ostream/SWKpceGCWE4/unsubscribe.
To unsubscribe from this group and all its topics, send an email to h2ostream+...@googlegroups.com.