about the mert !

[xiaodong]

hi, all
when I run the scirpt to tune the weights, I encounter a problem:
top@moses01:/media/sdb3/jane/corpus$ ../bin/localOptimization.sh --
method mert --janeConfig jane.config --dev input --reference
reference --init initial.lambda --optDir opt --randomRestarts 10
[error 9/6/2011 7:15:00 Jane] configuration error: parameter
"Jane.scalingFactors.s2tRatio" must be given a value
exiting...
gzip: opt/iter.1/merged/nBest.*.gz: No such file or directory
Iteration 1 result: 0
[error 9/6/2011 7:15:00 Jane] configuration error: parameter
"Jane.scalingFactors.s2tRatio" must be given a value
exiting...
gzip: opt/iter.2/generated/nBest.*.gz: No such file or directory
gzip: opt/iter.1/merged/nBest.*.gz: No such file or directory
gzip: opt/iter.1/merged/nBest.*.gz: No such file or directory
No new hypothesis in this iteration. Exiting...


my training set and dev set is very small , is that the reason ?

[David Vilar]

Hi!

This is due to a couple of new parameters that were added to the 1.1
release, but the manual was not updated accordingly :(

The complete list of (default) parameters is now

s2t
t2s
ibm1s2t
ibm1t2s
phrasePenalty
wordPenalty
s2tRatio
t2sRatio
isHierarchical
isPaste
glueRule

Both must be specified in the jane.config and initial.lambda files. In
your case, you are probably missing the s2tRation and t2sRatio.

Best regards,

David

--
David Vilar Torres
DFKI GmbH, Alt-Moabit 91c, 10559 Berlin
Tel. (+49) 30 238 95 1845

--------------- Legal Note ---------------
Deutsches Forschungszentrum fuer Kuenstliche Intelligenz GmbH
Firmensitz: Trippstadter Strasse 122, D-67663 Kaiserslautern
Geschaeftsfuehrung: Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster
(Vorsitzender), Dr. Walter Olthoff
Vorsitzender des Aufsichtsrats: Prof. Dr. h.c. Hans A. Aukes
Amtsgericht Kaiserslautern, HRB 2313