Orange Forum • View topic - C4.5 link is broken

C4.5 link is broken

Report bugs (or imagined bugs).
(Archived/read-only, please use our ticketing system for reporting bugs and their discussion.)
Forum rules
Archived/read-only, please use our ticketing system for reporting bugs and their discussion.

C4.5 link is broken

Postby roschler » Sun Apr 24, 2005 0:48

The link on the Orange + C4.5 page for finding the C4.5 files is broken (404 error):

http://www.cse.unsw.edu.au/~quinlan/c4.5r8.tar.gz

Is there a new one?

Also, you are running phpBB 2.0.8. This version has a big security vulnerability. Trust me, I and thousands of other phpBB board users know this the hard way.

Thanks.

Postby Blaz » Sun Apr 24, 2005 15:42

You're right, this page has moved. You now find the C4.5 sources at Quinlan's home page at Ruleqest at http://www.rulequest.com/Personal/.

Postby roschler » Mon Apr 25, 2005 13:54

Blaz,

Thanks. 8)

Postby Blaz » Tue Apr 26, 2005 7:40

Thanks to your post, we (Martin) have also updated phpBB and are now running the latest version.

Postby Guest » Tue Apr 26, 2005 13:49

Blaz,

Good. 8)

When a new update is available, you will see some text in bright red when you visit the administration panel, notifying you of the new version.

BTW, to distribute C4.5 you have to get permission from:

Business Liaison Office, University of Sydney, Sydney NSW 2006
Australia.

not from the author Ross Quinlan. He is the one that gave me this information.

I have sent them an e-mail, I'll let you know how it goes.

Thanks,
Robert
http://www.robodance.com/
Robosapien Dance Machine

Postby Blaz » Tue Apr 26, 2005 18:43

I've contacted Ross two years ago and got the same answer. We then decided not to include C45 in the distribution, but if you hear back from U of Sydney, let us know. Thanks!

Postby roschler » Thu Apr 28, 2005 19:35

Blaz,

I never thought to ask, but with all the existing techniques currently in Orange, is there really any pressing reason to still use C4.5?

Thanks.

Postby Blaz » Thu Apr 28, 2005 19:53

C4.5 has always been a good reference, and is considered in ML almost like a gold standard. For this reason, we would not like only to mimic it (through setting the right parameters for orange tree learner), but embed it just like it is in Quinlan's implementation [Janez never touched the functional code, just added an interface to oOrange and Python].

Postby roschler » Thu Apr 28, 2005 21:40

Blaz,

I just want to make sure I understand you.

Are you saying that C4.5 is much better at it's style of machine learning than the equivalent module in Orange?

Or, that there isn't really anything like the C4.5. technique in Orange at all?

The reason I'm asking is that sometime in the next two months, I am going to add Python scripting support to my open source Robosapien Dance Machine project; which is a Delphi app. I would like to provide some automatic sensor data learning / classification routines via Orange; so the robot I'm controlling can make tricky decisions. Since both our projects are GPL that should not be a problem.

But it seems like the C4.5 technique is a good one for object classification, based on sensor data pattern matching? (Please correct me if I'm wrong). And I don't want to spend too much time with it (C4.5) if I can't distribute it to my users. I'm a big fan of "one-click" installs and telling them to go download and install C4.5. by themselves, AFTER they have installed my app is not what I intend to do.

So if there is something that is in good enough in Orange for my purposes, then I'll just use that instead. If I ever get permission from Sydney I'll change things, but I'm not holding my breath.

Thanks.

Postby Blaz » Fri Apr 29, 2005 17:52

no, i just wanted to say that c4.5 is a nice reference. depending on classification task and peculiarities of your data, other machine learning algoritmh can perform much better (or much worse). there was a nice lesson at COIL 2000 challenge, where among many different algorithms, the simplest one (naive bayesian classifier) won [with a help of a little trick of constructing two new attributes, see http://www.liacs.nl/~putten/library/cc2000/].

in orange, classification tree learner should perform very similarly to c4.5 [use it through orngTree module, set sameMajorityPruning=1 and mForPruning=2].


Return to Bugs