[
https://issues.apache.org/jira/browse/OPENNLP-759?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14343755#comment-14343755
]
Daniel Russ commented on OPENNLP-759:
-------------------------------------
Oh I forgot to mention. I also checked that the Log Likelihoods are equal. I
could try to see if the parameters are exactly the same, but that might be
difficult as equals() was overridden on GISModel.
> Speed up GIS training by saving Executor in the GISTrainer
> ----------------------------------------------------------
>
> Key: OPENNLP-759
> URL: https://issues.apache.org/jira/browse/OPENNLP-759
> Project: OpenNLP
> Issue Type: Improvement
> Components: Machine Learning
> Affects Versions: maxent-3.0.3
> Reporter: Daniel Russ
> Priority: Minor
>
> In GISTrainer.nextIteration(double) an ExecutorService is created and
> shutdown. I don't see a reason to create a ExecutorService for each
> iteration.
> If you create the ExecutorService in the TrainModels(int, dataindexer, Prior,
> int, int) method you can save it as a field in GISTrainer or pass it as an
> argument to findParameters(int, double).
> To test it out, I made a MyGIS and MyGISTrainer classes with the fixes.
> There was a 5% speedup with 100 my dataset and 100 iterations of GIS.
> I would be happy to share the code with you. (I can not share my data
> though, sorry data-use agreements).
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
