Search Mailing List Archives
[java-nlp-user] Ask about update conditions in iis
horatio at gmail.com
Fri May 13 14:08:07 PDT 2011
To tell you the truth, we haven't used iterative scaling in the tagger
in many years, so there's a chance this particular code doesn't work
any more. In general, these days people use other methods such as
conjugate gradient or quasi newton methods (we use quasi newton) to
train models. If you find the code that is still there does what you
want, that's great, but we probably aren't going to support it much...
On Fri, May 13, 2011 at 1:16 AM, Arif N <aku at arifn.web.id> wrote:
> Good day,
> I wanna ask about the update condition for iis in stanford pos tagger..
> In class LambdaSolve, method improvedIterative(iter),
> the updateConds() method are immediately called after finding deltaL for
> one feature,
> So for finding deltaL for the next feature (still at the same iteration)
> the we're using the new conditional probability (because we have changed it
> on the previous feature)..
> Am I right?
> Is it the same if the updateConds are called for every feature after one
> iteration of finding deltaL?
> Thank you so much..
> Best Regards
> Arif N
> java-nlp-user mailing list
> java-nlp-user at lists.stanford.edu
More information about the java-nlp-user