Rene Pickhardt
Rene Pickhardt
First of all I guess you mean `Alpha` apears at the start of a line in our training file instead of testing file. I assumed that these cases existed but...
yes and no. I have empirical evidence but couldn't find a proof. Everyone I talked to has the same problem. Everyone believes it works noone is completely sure. The idea...
we should do a backoff if c(a b) was not seen (if we do proper backoffs this should add to one but it is not MLE)
I agree using is fine. installing and retrieving the toolkit is still a problem. On Tue, Jan 6, 2015 at 3:43 PM, Lukas Schmelzeisen wrote: > > We have a...
agree. optimizing is also something that partly relates to your bachlor thesis On Tue, Jan 6, 2015 at 3:39 PM, Lukas Schmelzeisen wrote: > > There has been quite some...
as we saw yesterday: disc i/o is not the limiting factor. we should still calculate the size of buffers and the maxcount divider dynamically in accordance with the memory size...
fine for me currently On Tue, Jan 6, 2015 at 3:38 PM, Lukas Schmelzeisen wrote: > > I had dynamically calculated buffer sizes in the programm for some time >...
correct On Tue, Jan 6, 2015 at 3:35 PM, Lukas Schmelzeisen wrote: > > We won't have this for stable release, right? > > — > Reply to this email...
I would like to have it in order to verify. Modified kneser ney was probosed by chen / goodman and their values cannot be reproduced and we don't know if...
yes we decided to not alow UNK tokens yet. On Tue, Jan 6, 2015 at 3:15 PM, Lukas Schmelzeisen wrote: > > I guess we won't have this for release?...