Dear Barry, Thank you very much for your answer. How do i know which set of weights give maximum performance on the Dev set? Is this determined by BLUE score? In all the weights files i have, BLUE sores are all 0.
Best, Thuong. On Sun, Jun 17, 2012 at 4:04 PM, Barry Haddow <[email protected]>wrote: > Hi Minh > > 1. You'll probably get better results by choosing the weights that give > maximum performance on the Dev set. > > 2. Yes, you can restart mira from the intermediate weights of a previous > run. > > 3. In the Moses.ini file, the first weight of the [weight-d] section is > the distortion weight, the other 6 are Lexicalised rendering, > > Cheers - Barry > > Sent from my ZX81 > > > ----- Reply message ----- > From: "Minh Thuong Bui" <[email protected]> > Date: Sat, Jun 16, 2012 09:23 > Subject: [Moses-support] MIRA for MOSES > To: <[email protected]> > > Dear all, > I am running MIRA and have three questions about MIRA: > > 1. The output of MIRA are different files containing tuned weights after > each epoch. I assume I should use the weights from the last epoch. Is this > correct? > > 2. There is no --continue option for 'mira', so when the process got > stopped due to some reason if I run mira with the a new moses.ini > containing weights from the last epoch of the killed mira, do the output > weights of the new mira are exactly as they were expected from the killed > process? For example, suppose that my killed mira produced weights for the > 3rd epoch; I put the weights in a new moses.ini file and run a new mira > with this moses.ini as input configuration file, are the weights from the > first epoch of the new mira are as the weights from the 4th epoch of the > killed mira? > > 3. I am a bit confused about the structure of the file containing the > weights and not sure how to put the weights into the moses.ini format. In > the case i am running now, there are 6 LexcialReodering weights and one > Distortion weight in the weights file but in the moses.ini i think these > weights were put together under the* # distortion (reordering) > weight [weight-d]* section. > > > I don't know which weight in this section should be the distortion and the > rest should be Lexical Reordering. Could someone please explain this to > me. > > Distortion 0.357801 > WordPenalty -2.60219 > !UnknownWordPenalty 1 > LM 0.665965 > LexicalReordering_wbe-msd-bidirectional-fe-allff_1 0.774182 > LexicalReordering_wbe-msd-bidirectional-fe-allff_2 0.458457 > LexicalReordering_wbe-msd-bidirectional-fe-allff_3 1.06073 > LexicalReordering_wbe-msd-bidirectional-fe-allff_4 0.961528 > LexicalReordering_wbe-msd-bidirectional-fe-allff_5 0.295955 > LexicalReordering_wbe-msd-bidirectional-fe-allff_6 0.5965 > PhraseModel_1 0.383676 > PhraseModel_2 0.482894 > PhraseModel_3 0.47771 > PhraseModel_4 0.811175 > PhraseModel_5 1.66881 > BleuScore 0 > > *In the moses.ini: * > * > * > > # distortion (reordering) weight > [weight-d] > 0.3 > 0.3 > 0.3 > 0.3 > 0.3 > 0.3 > 0.3 > > > Best regards, > Thuong. > > > > 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
