java SM failing on larger datasets

[Jonathan Huggins]

Hi all,

I'm encountering the following error when I run the sequence memoizer
on a number of larger (250,000+ token) datasets:

??? Java exception occurred:
java.lang.IllegalArgumentException: c = 14, t = 23, d = NaN
at

edu.columbia.stat.wood.pub.sequencememoizer.util.SeatingArranger.getSeatingArrangement(SeatingArranger.java:
44)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
318)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
326)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sampleSeatingArrangements(IntSequenceMemoizer.java:
285)
at
edu.columbia.stat.wood.pub.sequencememoizer.IntSequenceMemoizer.sample(IntSequenceMemoizer.java:
272)

Error in ==> runSMExperiment at 45
sm.sample(max(0,burnin-interval));


Any thoughts as to what might be going on? Querying parameters isn't
currently supported, so I wasn't have to track the evolution of the
discount parameters.

Thanks,
Jonathan

[Nicholas Bartlett]

not really sure, looks like the discount parameter is NaN obviously, which isn't good.  that is maybe happening in the gradient updates which happen as you load the model up with data, but I'm not sure.  these kinds of headaches usually required me to just do some conditional printing and then maybe also use the debugger to step through code.  hope it works out

nick

--
BARTLETT Nick

207.233.7017

[sunyamingdacong]

Hi all:

  I am a new learner of SM model, but the java version doesn't provide any example, so

I need your help. If anyone of you have used the java version, could you please give me

an example about how to use the SM model to train and test data.

  Thank you very much!

Yaming Sun