This exception occurred when I by mistake executed the training with 250K
sentences :(
I tried again with fewer sentences and it worked. I tried to change the Xmx
argument and it worked with 250k sentences.
It was probably caused by running out of memory because I couldn't reproduce
it.


On Thu, Feb 24, 2011 at 9:45 PM, [email protected] <
[email protected]> wrote:

> I got the following exception while training NameFinder with Portuguese
> corpus. I could train it before. I'll try to debug it.
>
> 495:  .. loglikelihood=-291913.4407221578    0.9836559881410069
> 496:  .. loglikelihood=-291852.86554366164    0.9836612515061119
> 497:  .. loglikelihood=-291792.45105346554    0.9836648305943834
> 498:  .. loglikelihood=-291732.1964646411    0.9836660938020086
> 499:  .. loglikelihood=-291672.10099584213    0.9836681991480507
> 500:  .. loglikelihood=-291612.16387121956    0.9836711466325095
> Writing name finder model ... Exception in thread "main"
> java.lang.ArrayIndexOutOfBoundsException: 1
>     at java.lang.String.charAt(String.java:688)
>     at java.io.DataOutputStream.writeUTF(DataOutputStream.java:336)
>     at java.io.DataOutputStream.writeUTF(DataOutputStream.java:306)
>     at
> opennlp.maxent.io.BinaryGISModelWriter.writeUTF(BinaryGISModelWriter.java:73)
>     at opennlp.maxent.io.GISModelWriter.persist(GISModelWriter.java:103)
>     at opennlp.model.GenericModelWriter.persist(GenericModelWriter.java:91)
>     at opennlp.tools.util.model.ModelUtil.writeModel(ModelUtil.java:68)
>     at
> opennlp.tools.util.model.GenericModelSerializer.serialize(GenericModelSerializer.java:39)
>     at
> opennlp.tools.util.model.GenericModelSerializer.serialize(GenericModelSerializer.java:31)
>     at opennlp.tools.util.model.BaseModel.serialize(BaseModel.java:317)
>     at opennlp.tools.cmdline.CmdLineUtil.writeModel(CmdLineUtil.java:187)
>     at
> opennlp.tools.cmdline.namefind.TokenNameFinderTrainerTool.run(TokenNameFinderTrainerTool.java:103)
>     at opennlp.tools.cmdline.CLI.main(CLI.java:183)
>
>
>
> On Wed, Feb 23, 2011 at 8:37 PM, Jörn Kottmann <[email protected]> wrote:
>
>> Hello everyone,
>>
>> our first OpenNLP 1.5.1 Release Candidate is ready for testing.
>>
>> It can be found here:
>> http://people.apache.org/~joern/releases/opennlp-1.5.1-incubating/rc1/
>>
>> Please test it as good as you can and check that it meets all the Apache
>> rules.
>> I will do all the historic regression testing with my private data
>> against this candidate.
>>
>> Thanks for your help.
>>
>> Jörn
>>
>
>

Reply via email to