Hi,
you do not use the latest version of Moses. In the latest version,
there are three numbers at the end, as produced by code in
phrase-extract/consolidate-main.cpp:
// counts, for debugging
fileConsolidated << "||| " << countE << " " << countF << " " << countEF;
Note that these numbers are not used for anything currently.
-phi
On Fri, Apr 5, 2013 at 1:01 PM, Prashant Mathur <[email protected]> wrote:
> Hi All,
>
> I have a background phrase table, and I want to add more phrases from
> the new data in that table on the fly.
> To do that, I have to extract and score the new phrases and redistribute
> the probabilities.
> In order to do that I would need sufficient stats to compute them.
>
> I am looking at the phrase table, and the last field of the table should
> be count and total_count as I understood from the code but I am confused...
> Is this correct?
> count => count of target phrase given the source phrase,
> total_count => count of source phrase
>
> if this is true then I have some entries in phrase table which are
> contradictory
>
> ! " And I started ||| ! » Et j' ai commencé à ||| 1 0.213791 0.5
> 1.71596e-05 2.718 ||| ||| 1 2
> ! " And I went , " ||| ! » Je me suis dit « ||| 1 4.9134e-05 1
> 1.43432e-09 2.718 ||| ||| 1 1
> ! " And I went , ||| ! » Je me suis dit ||| 0.25 5.76287e-05 1
> 2.20446e-08 2.718 ||| ||| 4 1
> ! " And I went ||| ! » Je me suis dit ||| 0.25 0.000656861 1 2.20446e-08
> 2.718 ||| ||| 4 1
>
> there are some entries in which count is greater than total_count... how
> is this possible?
>
> sorry if I understood everything wrong.
>
> Ciao,
> Prashant
> _______________________________________________
> 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
