Hi, all these threshold pruning settings are rather brittle, since they depend on what scaling is applied to the weights. That's why the "beam-threshold" is also defaulted to a value that makes it ineffective.
-phi On Thu, Mar 29, 2012 at 7:37 AM, Hieu Hoang <[email protected]> wrote: > no-one has ever mentioned it 'til now so unlikely to have be used a lot. > > with advent of cube pruning now, those limits and thresholds gets in the way > as much as they help > > > On 29/03/2012 21:17, Lane Schwartz wrote: > > So I just looked and here's what the Pharaoh manual says about > ttable-threshold: > > "-ttable-threshold - minimum probability for a phrase translation entry > (default 0, unlimited)" > > > When I run Moses with no parameters here's what it tells me about > ttable-limit: > > "-ttable-limit (ttl): maximum number of translation table entries per input > phrase" > > > I'm wondering if ttable-threshold was just dropped because no one ever set > it to anything other than zero. > > Lane > > > On Thu, Mar 29, 2012 at 10:10 AM, Hieu Hoang <[email protected]> > wrote: >> >> that ttable threhold is >> [ttable-limit] >> for as long as i remember. >> >> yep. I put in the check, couldn't believe it wasn't there before. >> >> >> >> On 29/03/2012 20:50, Lane Schwartz wrote: >> >> On Thu, Mar 29, 2012 at 9:24 AM, Lane Schwartz <[email protected]> wrote: >>> >>> A colleague of mine was just trying out a new git checkout of moses with >>> an old config file that he had. The old config file had a line for >>> "[ttable-threshold" and the new checkout of moses complained that that flag >>> didn't exist. >>> >>> I spent some time looking through the git logs and checking out old >>> versions of moses, and strangely I can't seem to find any old versions of >>> moses with that parameter present. Searching on google for "moses >>> ttable-threshold" reveals some hits, most notably the example Moses config >>> file from the JHU summer workshop >>> (http://www.statmt.org/jhuws/?n=Experiments.EPPS). >>> >>> Does anyone know what happened to the ttable-threshold parameter? Did it >>> get renamed, or removed? >> >> >> >> Well, I figured out why the error message is being thrown. On 2011-06-11 >> Hieu added a check in Parameter.cpp to verify that each parameter in the ini >> file is a real, valid, known parameter. >> >> Prior to that, I guess unknown parameters were silently ignored. >> >> I looked through the history of Parameter.cpp and StaticData.cpp going >> back to the first revision of Moses on sourceforge, and didn't find any >> mention of a ttable-threshold parameter. >> >> Based on that, and the fact that the parameter appears to have been >> present at the JHU summer workshop >> (http://www.statmt.org/jhuws/?n=Experiments.EPPS) I'm assuming that >> ttable-threshold was a Pharaoh parameter that was removed or renamed prior >> to Moses being put up on Sourceforge. >> >> If anyone knows, I would still be interested in knowing if >> ttable-threshold was removed or if it was renamed to something else. >> >> Thanks, >> Lane Schwartz >> >> >> >> _______________________________________________ >> 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 >> > > > > -- > When a place gets crowded enough to require ID's, social collapse is not > far away. It is time to go elsewhere. The best thing about space travel > is that it made it possible to go elsewhere. > -- R.A. Heinlein, "Time Enough For Love" > > > _______________________________________________ > 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
