Hi,
On 10.06.2013 14:17, Sondes Bannour wrote:
> I added this line:
>
> public WhiskRuleItem(TextRulerAnnotation tokenAnnotation) {
> super();
> setWordConstraint(new TextRulerWordConstraint(tokenAnnotation));
> * this.activeFeatures = new
> ArrayList<String>(tokenAnnotation.getFeatureMap().keySet());*
> }
This can be problematic, because conditions for all features are added
later on.
> and it worked, i get rules with FEATURE expression but when the FEATURE is
> in the context, the rule apply :
>
> *BEST EXTENSION IS: SW{REGEXP("humans")->MARKONCE(MBTO00001402)} ALL*?
> Lemma{FEATURE("lemma", ",")};
> Laplacian: 0.25 ; p=3; n=0
>
> *But when it is in the Target, the rule doesn't apply (p=0,n=0):*
>
> base1: Term{FEATURE("lemma", "human")->MARKONCE(MBTO00001402)};
> base2: Term{FEATURE("lemma", "human")->MARKONCE(MBTO00001402, 1, 2)} ALL*?
> SPACE;
> base1: p=0; n=0 --> laplacian = 1.0
> base2: p=0; n=0 --> laplacian = 1.0
>
> *
How are your types defined, what type has which feature, respectively?
Has Term the feature lemma?
You let the algorithm learn the rules sensitive to whitespaces? This
could maybe cause problems in the testing component. I haven't tested
that myself yet. Do you need the whitespaces? What is your exact
learning configuration, e.g., the "Filtered Feature Types"?
Best,
Peter
> Best,
> Sondes*
> *
>
