[Ligncse256] NaN in output - Update
Roger Levy
rlevy at ucsd.edu
Tue Feb 19 11:44:11 PST 2008
hbeecher at ling.ucsd.edu wrote:
> Hello again,
>
> I resolved 1 NaN error to get another! Turns out I hadn't transferred all
> the WSJ data from the data3.zip file to my data directory. So now the
> test harness reports Tag Accuracy: 0.945 (higher than the expected 92%).
> But I'm also getting
>
> Unknown Accuracy: NaN
>
> Could someone confirm the total files/directories in the data3.zip??
> My system shows 2313 files in 25 directories
Hi Henry,
These numbers are indeed different than what I get from the baseline
model. However, I also read 2313 files in 25 directories:
$ find . -type f | wc -l
2313
Other things to check: the program should be printing to STDERR that it
finds 39832 tagged sentences in the training set, 1700 in the validation
set, and 2416 in the test set.
Does anyone else have trouble of a sort similar to Henry's?
Roger
--
Roger Levy Email: rlevy at ucsd.edu
Assistant Professor Phone: 858-534-7219
Department of Linguistics Fax: 858-534-4789
UC San Diego Web: http://ling.ucsd.edu/~rlevy
More information about the Ligncse256
mailing list