Visualising weights of a NN

[jaa...@gmail.com]

Hi All,

I'm looking for a way to visualise weights of a Neural Network after training it. Since it seems like H20 is not directly support this yet, I'm trying to load the train model in R and then inspect it's weight using R. But when going through the JSON representation of the Model, unlike the Java counterpart, I couldn't find any variables representing weights. So is there a way to inspect weights of the model using R ?

Amila

[Arno Candel]

Hi Amila,

This is not currently supported, but you can expect this to come to life soon in h2o-dev.

Thanks,

Arno

[jaa...@gmail.com]

Thanks Arno!!

Can you tell me when this would be available.

Amila

[Arno Candel]

Hi Amila,
I would say within the next few weeks.
Thanks,
Arno

> --
> 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/f6P_sytaSuM/unsubscribe.
> To unsubscribe from this group and all its topics, send an email to h2ostream+...@googlegroups.com.
> For more options, visit https://groups.google.com/d/optout.

[Arno Candel]

I have implemented this feature, see

https://0xdata.atlassian.net/browse/PUBDEV-344

Example in https://github.com/h2oai/h2o-dev/blob/master/h2o-r/tests/testdir_algos/deeplearning/runit_deeplearning_weights_and_biases.R

Thanks,

Arno

On Friday, February 27, 2015 at 1:48:06 AM UTC-8, Arno Candel wrote:

Hi Amila,
I would say within the next few weeks.
Thanks,
Arno

> On Feb 26, 2015, at 5:54 PM, jaa...@gmail.com wrote:
>
> Thanks Arno!!
>
> Can you tell me when this would be available.
>
> Amila
>
> --
> 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/f6P_sytaSuM/unsubscribe.

> To unsubscribe from this group and all its topics, send an email to h2ostream+unsubscribe@googlegroups.com.

[jaa...@gmail.com]

Awesome!!

Checking that out..

[jaa...@gmail.com]

On Wednesday, March 4, 2015 at 7:51:47 PM UTC+5:30, jaa...@gmail.com wrote:
> Awesome!!
>
> Checking that out..

Hi Arno,

To get this feature, I tried building h2o-dev, but the build got failed due to the following test failure;

1) testTimeParse2(water.parser.ParseTimeTest)
java.lang.AssertionError: -4.8599999E7 -- 1.0
at org.junit.Assert.fail(Assert.java:88)
at org.junit.Assert.assertTrue(Assert.java:41)
at water.parser.ParserTest.testParsed(ParserTest.java:54)
at water.parser.ParseTimeTest.testTimeParse2(ParseTimeTest.java:63)
at sun.reflect.NativeMethodAccessorImpl.invoke0(Native Method)
at sun.reflect.NativeMethodAccessorImpl.invoke(NativeMethodAccessorImpl.java:57)
at sun.reflect.DelegatingMethodAccessorImpl.invoke(DelegatingMethodAccessorImpl.java:43)
at java.lang.reflect.Method.invoke(Method.java:606)
at org.junit.runners.model.FrameworkMethod$1.runReflectiveCall(FrameworkMethod.java:47)
at org.junit.internal.runners.model.ReflectiveCallable.run(ReflectiveCallable.java:12)
at org.junit.runners.model.FrameworkMethod.invokeExplosively(FrameworkMethod.java:44)
at org.junit.internal.runners.statements.InvokeMethod.evaluate(InvokeMethod.java:17)
at water.TestUtil$2$TimerStatement.evaluate(TestUtil.java:88)
at org.junit.rules.RunRules.evaluate(RunRules.java:20)
at org.junit.runners.ParentRunner.runLeaf(ParentRunner.java:271)
at org.junit.runners.BlockJUnit4ClassRunner.runChld(BlockJUnit4ClassRunner.java:70)
at org.junit.runners.BlockJUnit4ClassRunner.runChild(BlockJUnit4ClassRunner.java:50)
at org.junit.runners.ParentRunner$3.run(ParentRunner.java:238)
at org.junit.runners.ParentRunner$1.schedule(ParentRunner.java:63)
at org.junit.runners.ParentRunner.runChildren(ParentRunner.java:236)
at org.junit.runners.ParentRunner.access$000(ParentRunner.java:53)
at org.junit.runners.ParentRunner$2.evaluate(ParentRunner.java:229)
at org.junit.internal.runners.statements.RunBefores.evaluate(RunBefores.java:26)
at org.junit.internal.runners.statements.RunAfters.evaluate(RunAfters.java:27)
at org.junit.runners.ParentRunner.run(ParentRunner.java:309)
at org.junit.runners.Suite.runChild(Suite.java:127)
at org.junit.runners.Suite.runChild(Suite.java:26)
at org.junit.runners.ParentRunner$3.run(ParentRunner.java:238)
at org.junit.runners.ParentRunner$1.schedule(ParentRunner.java:63)
at org.junit.runners.ParentRunner.runChildren(ParentRunner.java:236)
at org.junit.runners.ParentRunner.access$000(ParentRunner.java:53)
at org.junit.runners.ParentRunner$2.evaluate(ParentRunner.java:229)
at org.junit.runners.ParentRunner.run(ParentRunner.java:309)
at org.junit.runner.JUnitCore.run(JUnitCore.java:160)
at org.junit.runner.JUnitCore.run(JUnitCore.java:138)
at org.junit.runner.JUnitCore.run(JUnitCore.java:117)
at org.junit.runner.JUnitCore.runMain(JUnitCore.java:96)
at org.junit.runner.JUnitCore.runMainAndExit(JUnitCore.java:47)
at org.junit.runner.JUnitCore.main(JUnitCore.java:40)

When I trained a DL using the latest stable release (Nunes (2.8.6.2)), still I couldn't see any model parameter named weights or biases. Is this available only in h2o-dev?

[Arno Candel]

Sorry about that. You can try the latest stable (or nightly) h2o-dev release download from http://h2o.ai/download

Cheers,
Arno

> --
> 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/f6P_sytaSuM/unsubscribe.

> To unsubscribe from this group and all its topics, send an email to h2ostream+...@googlegroups.com.

[ccl...@gmail.com]

Looks like a Java TimeZone failure. What timezone was this run in?

Cliff

[jaa...@gmail.com]

The machine is in IST.

I tried with the nightly build from h2o-dev. Initially when starting up the server, it simply failed giving following error;
03-23 01:22:17.089 192.168.52.5:54321 14374 FJ-126-7 FATAL: Attempting to join /192.168.52.9:54321 with an H2O version mismatch (md5 differs). (Is H2O already running?) Exiting.

However, after specifying -mkd5skip option I was able to get it running.
Now, when I go to the home page, it directs me to H2O flow page instead of showing the selection menu.

[cli...@0xdata.com]

Ok, the test failure is ignorable and probably relates to looking at a dataset in IST vs PST.  We'll work on the test.  Meanwhile you can comment out the test & see if the build finishes out.

The other error: "attempting to join" usually means you have a rogue or dead older H2O instance launched & running on the same cloud-name, IP & port.  You might want to kill off any extra instances you find.  Letting them cluster up (skipping the MD5 check) will lead to a weirdly broken cluster.

Cliff

[jaa...@gmail.com]

Hi Cliff,
Thanks for pointing that out. An instance running on a different machine has been attempting to join into the cluster and that's been causing the problem. After killing the rogue node I was able to start the instance without skipping MD5 check. However it still redirects me to flow page when I try to navigate to Home page.

[ccl...@gmail.com]

Hopefully the IST TZ test is fixed now (in h2o-dev bleeding edge).
The normal landing page for h2o-dev is Flow, so this is expected.... were you expecting the H2O-1 landing page?
The main GUI is very different from H2O-1, although all the same functionality is there; we're based off of the iPython Notebook style although the language is javaScript instead of Python. You can still click your way through all the same data-load, model-build & -view steps as in h2o-1.

Cliff

[jaa...@gmail.com]

Hi Cliff, Arno,
Thanks for helping out. I was able to see the trained weights after using the latest nightly build.