There were also some experiments during the CLSP workshop that built Moses (Fig. 6.8): http://www.clsp.jhu.edu/ws2006/groups/ossmt/documents/jhu-summer-workshop-final-report.pdf
Since MERT is only tuning a handful of parameters, it would be somewhat surprising if a drastically large tuning set made a difference. 2011/5/30 marco turchi <[email protected]>: > Hi Miles, > I check how tuning time changes according to the tuning set size. I do not > use a "drastically large tuning set" but I guess it can give an idea. > Section 4.5 of this technical report: > https://patterns.enm.bris.ac.uk/files/LearningCurveMain.pdf > > Cheers > Marco > > > On Mon, May 30, 2011 at 11:08 AM, Miles Osborne <[email protected]> wrote: >> >> well spotted. also, the OP needs to make sure that MERT is not run on >> the training set. it needs to be disjoint >> >> as a matter of interest, need this be a Bad Thing? (apart from using >> resources). it would be interesting for someone with the time to run >> MERT with a drastically large tuning set. >> >> Miles >> >> On 30 May 2011 10:01, Adam Lopez <[email protected]> wrote: >> > From the error it appears that you are running MERT on hundreds of >> > thousands of sentences. It generally only needs ~1000 sentences. >> > >> > 2011/5/30 [Intra] Mariusz Hawryłkiewicz <[email protected]>: >> >> Hello Moses team! >> >> >> >> I just wanted to ask you if it’s normal for mert to run over 10 days on >> >> a >> >> multicore machine (2 processors with 8 cores – AMD Opteron) and 24GB of >> >> RAM? >> >> The bilingual corpora is about 400k sentences large, and I am running >> >> mert >> >> with the following command: >> >> >> >> >> >> >> >> /moses/install/skrypty/scripts-20110516-1443/training/mert-moses.pl >> >> /moses/korpusy/ERP_test/lowercased/erp.lowercased.en >> >> /moses/korpusy/ERP_test/lowercased/erp.lowercased.pl >> >> /moses/install/moses/moses-cmd/src/moses /moses/install/model/moses.ini >> >> --working-dir /moses/korpusy/ERP_test/tuning/ --rootdir >> >> /moses/install/skrypty/scripts-20110516-1443/ >> >> --mertdir=/moses/install/moses/mert/ --decoder-flags "-v 0 -threads 16" >> >> >& >> >> /moses/korpusy/ERP_test/tuning/mert.out & >> >> >> >> >> >> >> >> Over these 10 days mert was able to run just 2 iterations and generated >> >> some >> >> moses.ini file with weights, but mert.out file shows that the process >> >> wasn’t >> >> finished at all. It ended up suddenly and the last message is: >> >> >> >> >> >> >> >> Translating line 246297 in thread id >> >> >> >> >> >> >> >> Isuppose this is not normal mert behavior.. what am I doing wrong then? >> >> >> >> >> >> >> >> Thanks for any hint on the issue! >> >> >> >> Best regards >> >> >> >> Mariusz >> >> >> >> _______________________________________________ >> >> Moses-support mailing list >> >> [email protected] >> >> http://mailman.mit.edu/mailman/listinfo/moses-support >> >> >> >> >> > >> > _______________________________________________ >> > Moses-support mailing list >> > [email protected] >> > http://mailman.mit.edu/mailman/listinfo/moses-support >> > >> > -- >> > The University of Edinburgh is a charitable body, registered in >> > Scotland, with registration number SC005336. >> > >> > >> >> >> >> -- >> The University of Edinburgh is a charitable body, registered in >> Scotland, with registration number SC005336. >> >> _______________________________________________ >> Moses-support mailing list >> [email protected] >> http://mailman.mit.edu/mailman/listinfo/moses-support > > _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
