Pages: [1]
  Print  
Author Topic: Variable Importance  (Read 1020 times)
DDelen
Newbie
*
Posts: 9


« on: November 27, 2013, 09:37:44 PM »

RapidMiner is great for many things except variable importance! All other major DM tools have it. That way, you can also assess the ranked ordered importance of variables in a predictive model. PLEASE!
Logged
JEdward
Full Member
***
Posts: 145


« Reply #1 on: November 28, 2013, 10:42:24 AM »

Hmmm... yes, there's nothing obvious that I can see.  Some Weka operators might have it built in.  Otherwise you may need to use something like the R caret package integrated into your RM process.  http://www.jstatsoft.org/v28/i05
Logged
DDelen
Newbie
*
Posts: 9


« Reply #2 on: November 29, 2013, 04:24:43 AM »

I am not aware of variable importance on Weka or in R. I think this is what differentiates a professional tool from an experimental one. I cannot understanding the reasoning behind not including Variable Importance in RapidMiner. IT IS UNFORTUNATE! Sad
Logged
awchisholm
Sr. Member
****
Posts: 411


WWW
« Reply #3 on: November 29, 2013, 03:03:51 PM »

The Weight By operators allow the influence of attributes on a label to be determined. The output is a set of weights where 1 means the attribute is relevant to the label and 0 means irrelevant.

There is also the Optimize Weights (Forward) operator which can contain a model and this might give something that is useful.

Andrew
Logged

JEdward
Full Member
***
Posts: 145


« Reply #4 on: November 29, 2013, 06:17:25 PM »

Hi Andrew & apologies DDelen,

At first I thought I did a quick search on the forum I found this post relating to RandomForests and Variable Importance http://rapid-i.com/rapidforum/index.php?topic=417.0 which sent me off track. 

Hopefully that's helped you out. 
Logged
Pages: [1]
  Print  
 
Jump to: