Hi, and it's also worthwhile checking if the optimization proceeds to something reasonable. Grep the log for 'BEST at' to see a summary of weights and devset BLEU scores. If the scores do not (slowly) grow, or if the absolute values of the BLEU scores are odd, something's wrong. You can also compare the various run.*.out files in mert-work dir.
Cheers, O. "Hieu Hoang" <[email protected]> wrote: >it can take a long time, it depends how large your tuning data is. >1000-2000 sentences is enough. Moses need to decoder this tuning set around >20 times. > >You should also filter and binarize your phrase table, reordering model and >language model so make the decoder run faster and take up less memory. > >you must set > --mertdir ~/mosesdecoder/bin >NOT > --mertdir ~/mosesdecoder/bin/mert > > >On 4 August 2013 04:02, Heidi Heweidy <[email protected]> wrote: > >> Hello, >> Is it possible that tuning takes more than 48 hours and still not yet >> finished? >> Ofcoarse I had to update the command for tuning: >> >> nohup nice ~/mosesdecoder/scripts/training/mert-moses.pl \ >> ~/corpus/news-test2008.true.fr ~/corpus/news-test2008.true.en \ >> ~/mosesdecoder/bin/moses train/model/moses.ini --mertdir >> ~/mosesdecoder/bin/ \ >> &> mert.out & >> >> by stating the --mertdir to be ~/mosesdecoder/bin/mert instead of >> ~/mosesdecoder/bin/ >> >> Is this correct? >> >> >> >> _______________________________________________ >> Moses-support mailing list >> [email protected] >> http://mailman.mit.edu/mailman/listinfo/moses-support >> >> > > >-- >Hieu Hoang >Research Associate >University of Edinburgh >http://www.hoang.co.uk/hieu >_______________________________________________ >Moses-support mailing list >[email protected] >http://mailman.mit.edu/mailman/listinfo/moses-support -- Ondrej Bojar http://www.cuni.cz/~obo _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
