İlknur Durgar El_kahlout <[EMAIL PROTECTED]> writes:

> 
> Dear experts;
> 
> I am  making some experiments from English to Turkish with Moses. I face 
> the same error with Marco and Chez in the HMM training:
> 
> -----------
> Hmm: Iteration 2
> Reading more sentence pairs into memory ...
> ERROR2: nan nan nanN:
> 1.92727e-06 1.92727e-06 2.07238e-07 1.56484e-06
> 1.92727e-06 1.92727e-06 2.07238e-07 1.56484e-06
> 
> GIZA gives this error either at iteration 2 or iteration 4.
> 
> my original data is 241K with max. token length as 90 and max. fertility 
> as 6. I got the above error with 241K. When i used the first 100K as 
> data everything worked fine. If i choose the first 49K, i got the same 
> error again. I checked the \n and \r counts as Chez solved his problem 
> on that way. \n and \r counts are equal. Does anyone have an idea what 
> can be the reason of this error?.
> 
> Thanks
> 
> --ilknur
> 

I am also observing similar patterns (i.e. "ERROR: nan nan nan 66 36", "ERROR2:
nan nan nan") 
while trying to train on hung-train2008 corpus although the .hu
and .en files have the same number of lines. 
I retained sentences up to length 100.

Does this error correspond to a degradation of performance? 
Did anyone train on the hung-train2008 corpus without observing such errors? 

Ergun



_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to