incubator-ctakes-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Tim Miller <>
Subject Re: assistance with dictionary lookup issue
Date Tue, 05 Feb 2013 15:10:23 GMT
Yeah, if you mean just change the loop to iterate over the list instead 
of getting an iterator that makes sense.  There is still some logic in 
there to leave out punctuation tokens but I think you were implying that 
to be in your mockup diff.

As for sorting, the AnnotationIndex defines a sort order for its iterators:

so we are safe assuming that anything extending Annotation will be 
iterated in sorted order.  Does that answer the questions we had? Or was 
I missing something subtle in that discussion?


On 02/05/2013 09:44 AM, Masanz, James J. wrote:
> Looks good to me, with one question.
> Instead of getting an iterator and then building a new list, can we just skip getting
the iterator and use the list that selectCovered returns?
> I will mock up a diff here of what I mean:
> -	Iterator btaItr = org.uimafit.util.JCasUtil.selectCovered(jcas, BaseToken.class, covering).iterator();
> -	while (btaItr.hasNext())
> -		{
> - 			BaseToken bta = (BaseToken);
> -				ltList.add(lt);
> - 			}
> - 		}
> +	ltList = org.uimafit.util.JCasUtil.selectCovered(jcas, BaseToken.class, covering);
> 	return ltList;
> I know you said it was quick and dirty at the moment - my 2 cents - unless someone comes
up with a better engineered solution, I think we could add the new method (with a name like
getLookupTokens) and leave the old one so we don't have to deprecate anything. And phase in
the change to the various *LookupInitializerImpl classes if needed.
> -- James
>> -----Original Message-----
>> From:
>> []
>> On Behalf Of Masanz, James J.
>> Sent: Monday, February 04, 2013 4:01 PM
>> To:
>> Subject: RE: assistance with dictionary lookup issue
>> I'll take a look at the patch. Also be aware of
>> which talks about a way of
>> enhancing performance  -- if willing to assume annotations (BaseTokens
>> currently) are sorted. Currently it's always BaseToken and always sorted,
>> just not sure if we want to code to that assumption.
>> ________________________________________
>> From:
>> [] on
>> behalf of Tim Miller []
>> Sent: Monday, February 04, 2013 3:43 PM
>> To:
>> Subject: assistance with dictionary lookup issue
>> Pei helped me track down an issue with performance I'd noticed in the
>> dictionary annotator, and I have filed the issue here:
>> I implemented a quick and dirty proof of concept fix and noticed dramatic
>> performance improvement.  I attached the patch to the issue, but it
>> involves changing an interface (currently does not try to fix other
>> implementing classes so obviously not ready for primetime), so I wanted to
>> solicit the list first in case anyone with better knowledge of that module
>> has some better engineering ideas than what I came up with.
>> Thanks,
>> --
>> Tim Miller, PhD
>> Postdoctoral Research Fellow
>> Children's Hospital Informatics Program
>> Children's Hospital Boston and Harvard Medical School
>> 617-919-1223

View raw message