Hi Marcin, It stops the loop for producing n-best translations from the search graph after a maximum of (nBestFactor * desiredNBestSize) iterations. Depending on the value of nBestFactor and the amount of identical candidates that are seen, you might end up with only a few distinct entries in your n-best list even for longer input sentences.
Cf. Manager::CalcNBest() in moses/Manager.cpp This is purely about the n-best list output and does not affect search in any way. If your problem is caused by pruning, then you need to look into other parameters. Cheers, Matthias On Thu, 2014-06-05 at 20:15 +0200, Marcin Junczys-Dowmunt wrote: > Hi Matthias, > this doesn't do much for me. What exactly is this parameter intended > for? I admit the option description is not telling me anything. > > Setting "-sd 10" for stack decoding with "-n-best-list file 100 > distinct" seems to help. I am getting distinct n-best lists with at most > 100 items and they seem to be a little bit better filled than with -sd 0. > > With cube pruning, "-cbd some_number" does not seem to do anything, I > also tried to increase the pop limit with no success. > Best, > Marcin > > W dniu 05.06.2014 19:17, Matthias Huck pisze: > > Hi Marcin, > > > > You can increase the n-best factor: > > > > -n-best-factor 100 > > > > Or just set it to 0 (= no limit). The default is 20. > > > > Cheers, > > Matthias > > > > > > On Thu, 2014-06-05 at 17:23 +0200, Marcin Junczys-Dowmunt wrote: > >> Hi list, > >> is there a way to make the n-best list in Moses a little bit more > >> diverse without increasing the maximum n-best list size? Let's say I am > >> willing to sacrifice decoding speed. > >> > >> For instance > >> > >> -n-best-list file 100 distinct > >> > >> generates 100 and collapses them, so sometimes there may be only one > >> hypothesis. I can get more distinct hypotheses by setting the size to > >> e.g. 1000, but then some sentences might actually come up with 1000 > >> distinct hypotheses which is overkill. > >> > >> The options -cube-pruning-diversity and -stack-diversity sound > >> promising. I should add that this is a minolingual task with reordering > >> switched off. I understand this has some impact on these options? > >> Best, > >> Marcin > >> _______________________________________________ > >> 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. _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
