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 2014/10/06 13:18:33 UTC
[jira] [Created] (OPENNLP-717) NameFinder trainer creates model
always with default feature generator
Rodrigo Agerri created OPENNLP-717:
--------------------------------------
Summary: NameFinder trainer creates model always with default feature generator
Key: OPENNLP-717
URL: https://issues.apache.org/jira/browse/OPENNLP-717
Project: OpenNLP
Issue Type: Bug
Components: Name Finder
Affects Versions: 1.6.0
Reporter: Rodrigo Agerri
Assignee: Rodrigo Agerri
Fix For: 1.6.0
While adding new features to name finder (e.g., OPENNLP-714) it was noticed that the NameFinder trainer performance degraded after adding more features than those contained in the default generator.
It turned out that at the moment of creating the TokenNameFinderModel, the feature generator parameter was not the one used for the training, but a default one (null). As a result, the init() method in TokenNameFinderModel always created the model with the default feature generator.
Solution: add a getter in the TokenNameFinderFactory class to have access to the featureGenerator created for each implementation of the factory, and then use that getter as a parameter when creating the models after training. It has been implemented and tested and it works.
If you find this solution ok, I will push the commit.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)