Pages: [1]
  Print  
Author Topic: M5 Prime?  (Read 3399 times)
daviddonohue
Newbie
*
Posts: 8


« on: June 03, 2008, 11:18:17 PM »

Ingo and friends,
I am quite fond of the Weka M5P learner.  I find it to be fast and highly flexible.  Specifically, none or very few of the RapidMiner learners can handle numeric or nominal attributes with a numeric label. 
Am I missing something in the standard RapidMiner learners?
Any plans to implement a RapidMiner version of M5P?
Thanks,
David
Logged
Ingo Mierswa
Administrator
Hero Member
*****
Posts: 1221



WWW
« Reply #1 on: June 04, 2008, 11:22:40 AM »

Hi David,

you are right, natively only the NeuralNet and NearestNeighbors support numerical predictions for nominal attributes. We will therefore add a M5P implementation to our todo list. Of course we also highly appreciate any contribution if someone is interested in implementing this learner.

We actually had a regression tree some time ago in our developer version but it turned out to perform less good on a large set of test sets and so we decided to stick to appropriate preprocessing operators (for example turning the nominal attributes into numerical ones) and apply superior learning schemes.

Cheers,
Ingo
Logged

Did you try our new Marketplace? Upload or download new Extensions, add comments, and organize your operators. Have a look at  http://marketplace.rapid-i.com
daviddonohue
Newbie
*
Posts: 8


« Reply #2 on: June 05, 2008, 09:41:18 AM »

Thanks, Ingo!
David
Logged
Pages: [1]
  Print  
 
Jump to: