#15 POSTaggerMETest Patch

closed-fixed
nobody
None
5
2010-08-07
2010-08-06
James Kosin
No

I found an issue with the test.. When it fails the expected and actual output results are reversed in the assert() statements. This could cause confustion if the test ever fails... as it did in my case.
I'm now trying to find out who broke the models. I'm guessing the changes in Maxent recently may have broken me [OT].

Discussion

  • James Kosin

    James Kosin - 2010-08-06

    POSTaggerMETest Patch

     
  • Joern Kottmann

    Joern Kottmann - 2010-08-06

    Yes the change in GISModel.eval breaks the test. Jason can you please comment ?

     
  • James Kosin

    James Kosin - 2010-08-06

    I found the issue, line 174 for version 1.3 of the file. The commit comment says it removed multiplying by constant inverse when correlation constant is 0 or not used. Unfortunately, it blows up the models when we try to train.

    I'm reverting my version to version 1.2 for this line of code, that fixes the models and our tests and training. I'm also interested in why Jason may have changed this.

     
  • James Kosin

    James Kosin - 2010-08-06

    Jorn,

    To help, I filed a bug on Maxent's Sourceforge page for this.
    https://sourceforge.net/tracker/?func=detail&aid=3040940&group_id=5961&atid=105961

    Do look over the patch for the asserts to this file. POSTaggerMETest.java that are with this patch tracker. It will help in the future; since, I was a little confused by the output of the expected and actual values.

     
  • Joern Kottmann

    Joern Kottmann - 2010-08-07
    • status: open --> closed-fixed
     
  • Joern Kottmann

    Joern Kottmann - 2010-08-07

    Applied the patch. Thanks James. The maxent problem through Jason fix will be discussed over at the maxent issue James created.

     

Log in to post a comment.

Get latest updates about Open Source Projects, Conferences and News.

Sign up for the SourceForge newsletter:





No, thanks