Hi, can you create a patch of your changes and attach it to a (new) jira issue. If it is OK for you, then I would try to integrate your changes and investigate that problem. I have taken a look at the whisk implementation this weekend and the 0/0 problem only occurred for bad rule anchoring, which is not possible in your example. I see no other way to solve the problem than debugging the actual rule induction.
Best, Peter On 03.06.2013 10:34, Sondes Bannour wrote: > Hi peter, > > Yes actually i added the feature functionality (not really in a proper > way :) ) and i kept the default learning configuration (except that i > put my postag roottype instead of the default one). Everything works > well except that rules with FEATURE (not only Lemma but all rules that > contain FEATURE) don't apply well (always p=0 and n=0). Lemma is not a > target type, it is in fact an input annotation (an annotation that > assigns the lemmatised form to words). There are no exceptions and no > errors (have a look at my attached console file). > > It's really hard to give a simple example because i am working on the > source code of textruler itself (i only added a class named > TextRulerAnnotationWithFeatures that inherits from > TextRulerAnnotation) and made whisk work with it (of course i modified > the class whiskRuleItem to add the FEATURE thing), i can send it to > you if you want). I didn't change the way rules are learnt or applied, > only their expressions. > > PS: I am a little stressed because i am doing this work for my phd :) > > Cheers, > Sondes
