[Ligncse256] More HW3 questions
Ben Cipollini
bcipolli at cogsci.ucsd.edu
Wed Mar 5 16:26:56 PST 2008
If anybody's started on HW3 and has comments / answers to these questions, it would be appreciated. I'm going to try and move ahead later this evening.
Thanks!
Ben
----- Original Message -----
From: Ben Cipollini
To: ligncse256 at ling.ucsd.edu
Sent: Saturday, March 01, 2008 9:00 AM
Subject: [Ligncse256] More HW3 questions
Hey all,
Some more questions from HW3:
1. Does anybody get performance on the BaselineParser like in the HW (F=30%)? I'm getting F1=16.43% on validation and 16.16% on test trees.
2. Is there any purpose to constructing a Grammar object? After being constructed, it doesn't seem used by the BaselineParser class at all. Commenting it out seems to have no effect.
3. Did we cover "Agenda-Based" parsers in class? If so, did we call them agenda-based parsers? If not, what did we call them?
4. You might think the BaselineParser would do better if it used HW2's HMM Tagger instead of the "unigram" tagger (which I think is really just another version of HW2's most frequent scorer + greedy decoder trellis combo).
Tagging performance (overall/unknown):
HW2 HMM perf: 95.8%/65%
HW unigram perf: 92%/40%
Instead I got a drop in performance from 15% to under 5%. Was anybody else able to hook up their HW2 tagger code to HW3? If so, what stats do you see?
------------------------------------------------------------------------------
_______________________________________________
Ligncse256 mailing list
Ligncse256 at ling.ucsd.edu
http://pidgin.ucsd.edu/mailman/listinfo/ligncse256
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://pidgin.ucsd.edu/pipermail/ligncse256/attachments/20080305/29a021bc/attachment.htm
More information about the Ligncse256
mailing list