the other things you might be interested in:
   1. A feature function can give a -infinity score. If it's weight is
positive, then the hypothesis is guaranteed to be pruned

https://github.com/moses-smt/mosesdecoder/commit/5ebb81a17af4008768f8fd93d7ad31442325f47c
   2. There can be more than 3 Evaluate() functions in the framework. For
example, Evaluate-After-All-Rules-Have-Been-Loaded(),
Evaluate-After-All-Rules-Have-Been-Applied-But-Before-Search(). Fabienne
knows more about those



On 3 December 2013 13:18, Ales Tamchyna <[email protected]> wrote:

> I will, thanks!
>
> Ales
>
>
> On 12/03/2013 01:55 PM, Hieu Hoang wrote:
>
>> yes, of course. The current set of FF only calculate future cost in the
>> 1st Evaluate. However, there's no reason why future cost should be
>> updated in the 2nd. Update it if you think it would be useful
>>
>>
>> On 3 December 2013 12:47, Ales Tamchyna <[email protected]
>> <mailto:[email protected]>> wrote:
>>
>>     Hi,
>>
>>     I'm working on a feature function that depends on the source and I
>> would
>>     like to add its score to future cost estimates.
>>
>>     I noticed that FFs can only affect future score in the first overload
>> of
>>     the Evaluate() method which does not have access to the input. Is
>> there
>>     a reason for this limitation?
>>
>>     As far as I can tell from the code (in TranslationOptionCollection.
>> cpp),
>>     pruning only occurs *after* EvaluateWithSource() is called. So would
>> it
>>     break anything if FFs which are called in this method could also
>> affect
>>     future cost estimates?
>>
>>     If not, is it OK if I add a reference to estimatedFutureScore as a
>>     parameter to the second overload?
>>
>>     Thanks,
>>     Ales
>>     _______________________________________________
>>     Moses-support mailing list
>>     [email protected] <mailto:[email protected]>
>>
>>     http://mailman.mit.edu/mailman/listinfo/moses-support
>>
>>
>>
>>
>> --
>> Hieu Hoang
>> Research Associate
>> University of Edinburgh
>> http://www.hoang.co.uk/hieu
>>
>>
>


-- 
Hieu Hoang
Research Associate
University of Edinburgh
http://www.hoang.co.uk/hieu
_______________________________________________
Moses-support mailing list
[email protected]
http://mailman.mit.edu/mailman/listinfo/moses-support

Reply via email to