That's an excellent question, Chris.  I would recommend making a  
second copy of mert-moses.pl and editing the following lines

my $additional_triples = {
     # if the more lambda parameters for the weights are needed
     # (due to additional tables) use the following values for them
     "d"  => [ [ 1.0, 0.0, 2.0 ],    # lexicalized reordering model
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ] ],
     "lm" => [ [ 1.0, 0.0, 2.0 ] ],  # language model
     "g"  => [ [ 1.0, 0.0, 2.0 ],    # generation model
               [ 1.0, 0.0, 2.0 ] ],
     "tm" => [ [ 0.3, 0.0, 0.5 ],    # translation model
               [ 0.2, 0.0, 0.5 ],
               [ 0.3, 0.0, 0.5 ],
               [ 0.2, 0.0, 0.5 ],
               [ 0.0,-1.0, 1.0 ] ],  # ... last weight is phrase penalty
};

So that they instead say

my $additional_triples = {
     # if more lambda parameters for the weights are needed
     # (due to additional tables) use the following values for them
     "d"  => [ [ 1.0, 0.0, 2.0 ],    # lexicalized reordering model
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ],
               [ 1.0, 0.0, 2.0 ] ],
     "lm" => [ [ 1.0, 0.0, 2.0 ] ],  # language model
     "g"  => [ [ 1.0, 0.0, 2.0 ],    # generation model
               [ 1.0, 0.0, 2.0 ] ],
     "tm" => [ [ 0.3, 0.0, 0.5 ],    # translation model
               [ 0.2, 0.0, 0.5 ],
               [ 0.3, 0.0, 0.5 ],
               [ 0.3, 0.0, 0.5 ],
               [ 0.2, 0.0, 0.5 ],
               [ 0.0,-1.0, 1.0 ] ],  # ... last weight is phrase penalty
};


That seems to work for the phrase table.

Note that doing a similar process for the re-ordering table causes  
moses to crash, since each of the re-ordering probs correspond to a  
particular orientation of phrases, so you can't just add feature  
functions willy-nilly.

--Chris

On Sep 3, 2008, at 6:12 PM, Chris Callison-Burch wrote:

> Hi guys,
>
> I'm trying to add another weight to the phrase table.  I get a message
> in MERT training that says "Your model tm needs 6 weights but we
> define the default ranges for only 5 weights.  Cannot use the default,
> you must supply lambdas by hand."    Can anyone tell me how to do so?
>
> Thanks a million.
>
> --Chris
> _______________________________________________
> 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

Reply via email to