[
https://issues.apache.org/jira/browse/OPENNLP-138?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13001203#comment-13001203
]
James Kosin commented on OPENNLP-138:
-------------------------------------
Generates better results also:
Before change
---------------------
499: .. loglikelihood=-2765.77789572618 0.9974904356623334
500: .. loglikelihood=-2763.104274308045 0.9974904356623334
Writing name finder model ... done (1.841s)
Wrote name finder model to
path: C:\Users\James
Kosin\Documents\NetBeansProjects\thesis\DocCompare\enNameFinder.model
Testing on eng.testa
Loading Token Name Finder model ... done (0.359s)
current: 193.2 sent/s avg: 193.2 sent/s total: 226 sent
current: 678.3 sent/s avg: 407.2 sent/s total: 851 sent
current: 624.6 sent/s avg: 476.7 sent/s total: 1465 sent
current: 916.8 sent/s avg: 584.6 sent/s total: 2380 sent
current: 668.7 sent/s avg: 601.2 sent/s total: 3048 sent
Average: 602.4 sent/s
Total: 3251 sent
Runtime: 5.397s
Precision: 0.799424509140149
Recall: 0.794850218781555
F-Measure: 0.7971308016877637
Testing on eng.testb
Loading Token Name Finder model ... done (0.359s)
current: 149.7 sent/s avg: 149.7 sent/s total: 154 sent
current: 736.7 sent/s avg: 438.9 sent/s total: 890 sent
current: 647.0 sent/s avg: 506.8 sent/s total: 1526 sent
current: 800.6 sent/s avg: 579.9 sent/s total: 2326 sent
current: 1119.2 sent/s avg: 687.6 sent/s total: 3443 sent
Average: 687.6 sent/s
Total: 3454 sent
Runtime: 5.023s
Precision: 0.75829636202307
Recall: 0.7565509915014165
F-Measure: 0.7574226712753699
After change
------------------
499: .. loglikelihood=-2765.77789572618 0.9974904356623334
500: .. loglikelihood=-2763.104274308045 0.9974904356623334
Writing name finder model ... done (1.794s)
Wrote name finder model to
path: C:\Users\James
Kosin\Documents\NetBeansProjects\thesis\DocCompare\enNameFinder.model
Testing on eng.testa
Loading Token Name Finder model ... done (0.375s)
current: 291.5 sent/s avg: 291.5 sent/s total: 309 sent
current: 809.9 sent/s avg: 536.8 sent/s total: 1080 sent
current: 1176.4 sent/s avg: 749.2 sent/s total: 2255 sent
Average: 830.4 sent/s
Total: 3251 sent
Runtime: 3.915s
Precision: 0.8666666666666667
Recall: 0.8444968024234265
F-Measure: 0.8554381179679509
Testing on eng.testb
Loading Token Name Finder model ... done (0.374s)
current: 130.2 sent/s avg: 130.2 sent/s total: 132 sent
current: 917.2 sent/s avg: 525.9 sent/s total: 1075 sent
current: 1042.4 sent/s avg: 691.8 sent/s total: 2083 sent
Average: 907.5 sent/s
Total: 3454 sent
Runtime: 3.806s
Precision: 0.8114532685035116
Recall: 0.7978045325779037
F-Measure: 0.8045710204446032
> The Name Finder always creates/uses two sets of feature generators
> ------------------------------------------------------------------
>
> Key: OPENNLP-138
> URL: https://issues.apache.org/jira/browse/OPENNLP-138
> Project: OpenNLP
> Issue Type: Bug
> Components: Name Finder
> Affects Versions: tools-1.5.0-sourceforge
> Reporter: Jörn Kottmann
> Assignee: Jörn Kottmann
> Priority: Blocker
> Fix For: tools-1.5.1-incubating
>
>
> The NameFinderME during initialization either creates default feature
> generators or uses a set of feature generators provided by the user. In both
> cases the NameFinderME code calls the DefaultNameContextGenerator()
> constructor and then adds the feature generators to the context generator.
> The code ignores the fact that DefaultNameContextGenerator creates default
> feature generators on its own. The add call does not replace the existing
> feature generation, but really adds them, so now there are two sets of
> feature generators.
--
This message is automatically generated by JIRA.
-
For more information on JIRA, see: http://www.atlassian.com/software/jira
