You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@opennlp.apache.org by "Rodrigo Agerri (JIRA)" <ji...@apache.org> on 2014/06/04 09:55:01 UTC
[jira] [Closed] (OPENNLP-687) FMeasure change to avoid duplication
of true positives in the prediction Spans
[ https://issues.apache.org/jira/browse/OPENNLP-687?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Rodrigo Agerri closed OPENNLP-687.
----------------------------------
Resolution: Fixed
The calculation of FMeasure now removes a matched from the predictions array to avoid duplicating the true positives, as detected while solving OPENNLP-31.
All test pass, issue closed.
> FMeasure change to avoid duplication of true positives in the prediction Spans
> ------------------------------------------------------------------------------
>
> Key: OPENNLP-687
> URL: https://issues.apache.org/jira/browse/OPENNLP-687
> Project: OpenNLP
> Issue Type: Bug
> Affects Versions: tools-1.5.3
> Reporter: Rodrigo Agerri
> Assignee: Rodrigo Agerri
> Fix For: 1.6.0
>
>
> As a follow-up of OPENNLP-31, it has been noticed that the FMeasure is calculated in a way that the same prediction can be matched more than once.
> It is proposed to modify the FMeasure class to remove a prediction Span after being matched with a reference Span.
> This change will mean that the FMeasure class will look like the provisional ParseEval class, so that the ParseEval class can be removed.
> Before going ahead and change this we need to ascertain that the F results do not change for other components (chunker, namefinder, etc.)
--
This message was sent by Atlassian JIRA
(v6.2#6252)