You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@opennlp.apache.org by "Jörn Kottmann (JIRA)" <ji...@apache.org> on 2011/02/01 14:02:29 UTC

[jira] Created: (OPENNLP-124) Maxent/Perceptron training should report progess back via an API

Maxent/Perceptron training should report progess back via an API
----------------------------------------------------------------

                 Key: OPENNLP-124
                 URL: https://issues.apache.org/jira/browse/OPENNLP-124
             Project: OpenNLP
          Issue Type: Improvement
          Components: Maxent
            Reporter: Jörn Kottmann


Currently any training progress is printed to the console. The code should be changed to report the training progress back via an API. A command line training tool could use this API to print the status messages to the console. Other applications, e.g. a training server could use the reported results to display them to a user in a ui interface.

-- 
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira

       

[jira] [Commented] (OPENNLP-124) Maxent/Perceptron training should report progess back via an API

Posted by "Jörn Kottmann (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/OPENNLP-124?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13027657#comment-13027657 ] 

Jörn Kottmann commented on OPENNLP-124:
---------------------------------------

The following interface could be defined to report the current training progress:

interface TrainingProgessMonitor {
  finishedIteration(int iteration, int numberCorrectEvents, int totalEvents);
  finishedTraining(int iterations, int numberCorrectEvents, int totalEvents, StopCriteria stopCriteria);
}

The following training methods will be added to maintain API backward compatibility:
GISModel GISTrainer.trainModel(int iterations, DataIndexer di, Prior modelPrior, int cutoff, TrainingProgressMonitor monitor)
AbstractModel PerceptronTrainer.trainModel(int iterations, DataIndexer di, int cutoff, boolean useAverage, TrainingProgressMonitor monitor)

The existing implementations will call the newly defined methods and pass a default monitor which uses the display method to print to the console as done by the current implementation.

Maybe a result object should be defined which can also compute the accuracy and log-likelihood.

> Maxent/Perceptron training should report progess back via an API
> ----------------------------------------------------------------
>
>                 Key: OPENNLP-124
>                 URL: https://issues.apache.org/jira/browse/OPENNLP-124
>             Project: OpenNLP
>          Issue Type: Improvement
>          Components: Maxent
>            Reporter: Jörn Kottmann
>
> Currently any training progress is printed to the console. The code should be changed to report the training progress back via an API. A command line training tool could use this API to print the status messages to the console. Other applications, e.g. a training server could use the reported results to display them to a user in a ui interface.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira

[jira] [Commented] (OPENNLP-124) Maxent/Perceptron training should report progess back via an API

Posted by "Jason Baldridge (JIRA)" <ji...@apache.org>.
    [ https://issues.apache.org/jira/browse/OPENNLP-124?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13027807#comment-13027807 ] 

Jason Baldridge commented on OPENNLP-124:
-----------------------------------------

Yes, I very much agree that not having everything printed out by default is good. The TrainingProgressMonitor interface looks good, and yes, it would be better to have a Result object rather than each of the values (such that Result is an interface that could have different implementations, etc).

> Maxent/Perceptron training should report progess back via an API
> ----------------------------------------------------------------
>
>                 Key: OPENNLP-124
>                 URL: https://issues.apache.org/jira/browse/OPENNLP-124
>             Project: OpenNLP
>          Issue Type: Improvement
>          Components: Maxent
>            Reporter: Jörn Kottmann
>
> Currently any training progress is printed to the console. The code should be changed to report the training progress back via an API. A command line training tool could use this API to print the status messages to the console. Other applications, e.g. a training server could use the reported results to display them to a user in a ui interface.

--
This message is automatically generated by JIRA.
For more information on JIRA, see: http://www.atlassian.com/software/jira