2009/2/20 Aditya :
> I'm trying to install moses onto a 64-bit linux machine. I have compiled
> srilm with the MACHINE_TYPE variable set to "i686-m64". Now, when compiling
as far as i remember from my last 64bit install, Moses gets the
architecture type from the result of an SRILM script.
in the S
Hi,
I'm trying to install moses onto a 64-bit linux machine. I have compiled srilm
with the MACHINE_TYPE variable set to "i686-m64". Now, when compiling moses, I
get this :
//Making all in moses/src
make[2]: Entering directory `/data/acp08avk/darwin/moses/moses/src'
make all-am
Just to close the loop on this, in case anyone else runs into this issue. I
simply hadn't deleted all the files in the training directories after the first
run in the opposite direction. The training script reuses some of the files if
they already exist, rather than overriding.
Now our Moses in
Hi James
There was an investigation into different methods of parallelising model 1 in
this paper:
Fully Distributed EM for Very Large Datasets
by: Jason Wolfe, Aria Haghighi, Dan Klein
http://www.cs.berkeley.edu/~aria42/pubs/icml08-distributedem
As for pseudo-code for the IBM models, check out
Another architecture to consider is storing/distributing the ttable
from a single central repository. Most of the ttable is full of crap,
and for each sentence, you know exactly what parameters will be
required in advance of running your E step. However, by not
distributing stuff that you don't n
Hi, James,
PGIZA++ is not using OpenMPI, and only use shared storage to transfer model
files, that could be a bottleneck, MGIZA++ is just using multi-thread. So
they are not quite complete and can be further improved, the advantage of
PGIZA++ is it already decomposed every training step (E/M, mode
if u're comparing the speed between (cube pruning on windows) and (normal
search on linux) the slowness is prob. due to the different language model
implmenetations used.
the windows version doesn't support SRILM or IRSTLM, instead it uses the the
built-in LM which is slow & memory hungry
Hieu
Yes, though not directly. PhraseDictionaryTreeAdaptor forwards its calls to
PDAImp, which is a wrapper on PhraseDictionaryTree, which uses a PrefixTreeF
template class. PrefixTreeF reads binary files using a smart pointer
defined in FilePtr.h.
Adam
On Fri, Feb 20, 2009 at 11:25 AM, tah...@gmail
Binarized tables are not read with zlib.
On Fri, Feb 20, 2009 at 9:38 AM, tah...@gmail.com wrote:
> I'm hoping someone can help me understand how Moses Decoder reads the
> binarized tables.
>
> 1) Does Moses perform all phrase table and LM reads through the zlib
> decompression interface? I.e.
I'm hoping someone can help me understand how Moses Decoder reads the
binarized tables.
1) Does Moses perform all phrase table and LM reads through the zlib
decompression interface? I.e. are phrase tables data decompressed using zlib
during runtime?
2) If yes, does Moses Decoder also read the bin
Documentation of the new mert implementation is here:
http://www.statmt.org/moses/?n=FactoredTraining.Tuning
The main script is
scripts/training/mert-moses-new.pl
This script is just an interface between Moses and the new mert code,
so you need to specify where mert software is located.
Use th
11 matches
Mail list logo
