Pages: [1]
  Print  
Author Topic: Correlation operators with matrix output shall also provide weights output  (Read 707 times)
qwertz
Full Member
***
Posts: 126


« on: June 30, 2014, 11:29:29 AM »

Dear RapidMiner programmers,

referring to post http://rapid-i.com/rapidforum/index.php/topic,8048.0.html I was not able to manage to transform the actual matrix content of e.g. mutual information operator to weights.

As a partly work-around it is possible to write the matrix content via report to a file but still it seems impossible to apply this content as weight.

I would urgently appreciate a weight output which works as described in the following text:


Matrix   att1   att2   att3
att1        1       5        4
att2        6       1        5
att3        1       0        1

If this is the given matrix I wish to simply select an attribute e.g. "att3" and get the mutual information in this column as weights for each attribute, i.e. 4 as weight for att1 ... 1 as weight for att3.


PS: I could also imagine that an example to weights operator would come handy in general.


Best regards
Sachs
« Last Edit: June 30, 2014, 11:35:40 AM by qwertz » Logged
qwertz
Full Member
***
Posts: 126


« Reply #1 on: September 30, 2014, 10:01:01 AM »

Despite the issue mentioned above it would also be handy to have an operator that calculates correlation, mutual information, etc. but not as a matrix but just for chosen attributes, i.e.

cor (att1 and att2)
cor (att1 and att3)
cor (att1 and att4)

and not the correlation between each and every attribute like cor (att2 and att3).

With many attributes this could improve the speed tremendously. Furthermore, even with the mentioned workaround to read out the resulting values this method stops working if the matrix is too large. In this case it simply does not display all colums anymore.


Cheers
Sachs
Logged
Pages: [1]
  Print  
 
Jump to: