Skip to first unread message
Georgi Parlichev
Jul 1, 2015, 2:30:33 PM7/1/15
to wekamooc...@googlegroups.com
Dear Ian,
I wonder if it is possible to make WEKA predict a set of 6 numeric attributes based on classification of these sets into a new attribute a7. The 7th attributes is generated based on some repeated pattern of combining the 6 numerical attributes. In the example below a1-a6 are numeric attributes, a7 is a nominal attribute, c1...cn are classes, and the question marks "?" are the values I might try to predict. In short, I want to use a7 to predict a1-a6, i.e. the 7th attribute explain the first 6. Perhaps this problem can be approached with some kind of regression/neural-network/SVM. Or, should I make WEKA predict one attribute at a time or using multilable implementation of WEKA (MEKA)?
Another problem a run into is actually the generation of the 7th class. The 7th class should be generated by binning the numbers in the first six attributes into a smaller number of "bins" and use the pattern those bins follow to generate a7. I looked at the unsupervised.instance.SubsetByExpression filter but it might not work due the the length of expression I might have to write. Maybe its time to write some code?
a1 | a2| a3| a4| a5| a6| a7|
num|num|num|num|num|num|nom|
1| 2| 15| 22| 26| 32| c1|
3| 9| 18| 24| 32| 38| c2|
?| ?| ?| ?| ?| ?| ?|
I wonder if it is possible to make WEKA predict a set of 6 numeric attributes based on classification of these sets into a new attribute a7. The 7th attributes is generated based on some repeated pattern of combining the 6 numerical attributes. In the example below a1-a6 are numeric attributes, a7 is a nominal attribute, c1...cn are classes, and the question marks "?" are the values I might try to predict. In short, I want to use a7 to predict a1-a6, i.e. the 7th attribute explain the first 6. Perhaps this problem can be approached with some kind of regression/neural-network/SVM. Or, should I make WEKA predict one attribute at a time or using multilable implementation of WEKA (MEKA)?
Another problem a run into is actually the generation of the 7th class. The 7th class should be generated by binning the numbers in the first six attributes into a smaller number of "bins" and use the pattern those bins follow to generate a7. I looked at the unsupervised.instance.SubsetByExpression filter but it might not work due the the length of expression I might have to write. Maybe its time to write some code?
a1 | a2| a3| a4| a5| a6| a7|
num|num|num|num|num|num|nom|
1| 2| 15| 22| 26| 32| c1|
3| 9| 18| 24| 32| 38| c2|
?| ?| ?| ?| ?| ?| ?|
Reply all
Reply to author
Forward
0 new messages