classifier4j-devel

RE: [Classifier4j-devel] How do I train it?

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

> -----Original Message-----
> From: moedusa [mailto:mo...@in...]
> Sent: Tuesday, 11 November 2003 7:46 AM
> To: cla...@li...
> Subject: [Classifier4j-devel] How do I train it?
> 
> 
> Hi! I am playing with classifier4j now, and one thing is 
> confusing me: 
> let's say I want to train C4J this way: feed it with some 
> text (a fiew 
> paragraphs) and category id, but I can not figure out how to 
> implement 
> this now... Looks like I must parse text to words, filter 
> them and train 
> it word-by-word now... Is it right, or there is another way and my 
> understanding of training such things is completely wrong?
> 
> Philipp.
> 


You don't need to train word-by-word, you can just use the
ITrainableClassifier.teachMatch(String) and teachNonMatch(String) methods
(See
<http://classifier4j.sourceforge.net/apidocs/net/sf/classifier4J/ITrainable.
html#teachMatch(java.lang.String)>). If you want to use non-default
categories then there are version of teachMatch & teachNonMatch that take a
category, too.

Have you looked at the code for the demos, in particular
net.sf.classifier4J.demo.Trainer
(<http://cvs.sourceforge.net/viewcvs.py/classifier4j/Classifier4J-Optional/s
rc/java/net/sf/classifier4J/demo/Trainer.java?view=markup>)?

Let me know if that doesn't make things clearer.

Nick

RE: [Classifier4j-devel] How do I train it?

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

> 
> 1. in JDBCWordsDataSource - constructor did not initialised 
> IJDBCConnectionManager connectionManager provided to it ( I added
> connectionManager = cm; below commented string //this(cm, 
> ICategorisedClassifier.DEFAULT_CATEGORY);).
> 

Yes, that is fixed in CVS (note to self - must do a release and stop wasting
people's time with this bug!)


> 2. in DriverMangerJDBCConnectionManager I added one more 
> constructor and 
> one private field dbDriver (String) and slightly changed 
> getConnection() 
> method to initialise driver class if dbDriver != null 
> (Class.forName(dbDriver)
> 

Hmm.. I had always thought it would make more sense to load the JDBC driver
class in your own code. After all, I would have thought that if you are
using a JDBC based database for Classifier4J then there is a good chance you
are using it for other things as well, so it makes sense for you to control
your own driver loading.

Do people disagree with this logic?

Nick

Re: [Classifier4j-devel] How do I train it?

[moedusa <mo...@in...>]

Nick Lothian wrote:
> I had always thought it would make more sense to load the JDBC driver
> class in your own code. ... Do people disagree with this logic?

Well, not. But when using this from inside IDE (NetBeans) you have to 
mount driver manually, and that is not how-we-used-to-do-it-before 
(usually we have to supply JDBC connection details with properties or 
smth., see http://www.quartzscheduler.org/features.jsp for exmp.)... But 
I think that is my own trick (I was playing with C4J alone, so it was 
the only thing required DB access), so there is no need to change 
something in there, but, maybe, make a note in javadoc about it...

Re: [Classifier4j-devel] How do I train it?

[moedusa <mo...@in...>]

Nick, I have understood how it works looking at unit tests, but to make 
it work I'v made some chages... I use custom categories, so it have to 
make me fix some inconsistencies:

1. in JDBCWordsDataSource - constructor did not initialised 
IJDBCConnectionManager connectionManager provided to it ( I added
connectionManager = cm; below commented string //this(cm, 
ICategorisedClassifier.DEFAULT_CATEGORY);).

2. in DriverMangerJDBCConnectionManager I added one more constructor and 
one private field dbDriver (String) and slightly changed getConnection() 
method to initialise driver class if dbDriver != null 
(Class.forName(dbDriver)

After this I could make things work as needed, so that's okay now :)

Philipp.