RE: [Classifier4j-devel] Bayesian Case Study

[Nick L. <nl...@es...>]

> -----Original Message-----
> From: Matt Collier [mailto:MCo...@my...]
> Sent: Friday, 14 November 2003 11:44 AM
> To: cla...@li...
> Subject: RE: [Classifier4j-devel] Bayesian Case Study
> 
> 
> Very nice.  Should we keep these in a flat file?  This would 
> make alot of 
> sense in my opinion.
> 

Tha makes sense to me.

> Do we want to modify the default tokenizer and stop list 
> provider, or do we 
> want to extend it?
> 

Create a new implemnetation of the IStopWordProvider interface that reads
from a resource. You might want to read a bit abotu java interfaces if you
haven't already.

> If we want to extend it, can you please shortcut me to doing 
> this.  I think I 
> understand that we will create a class that "extends default 
> tokenizer" etc, 
> but how will this new class be used by the other classes and 
> methods such as 
> bayesian.classify?  Surely we won't have to modify all this 
> code, or perhaps 
> we do.  I don't know... which is why I'm asking... :)
> 

Yes, it is a valid question. Fortuanly, we thought of this when we coded it
a while ago (pat myself on my back!).

There is a constructor for BayesianClassifier that looks like:

public BayesianClassifier(IWordsDataSource wd, ITokenizer tokenizer,
IStopWordProvider swp) 

Which allows you to specify your own stop-word provider. 

As a general rule most of Classifier4J is coded against interfaces, to make
this kind of change pretty easy. It means it is very flexible - it's just
that we don't have many non-standard implementations....