CODEWITHSUNDEEP
javaclassificationwekasvm
Francesco
Francesco

Reputation: 75

Weka Attribute Evaluator with multiple-attributes features

I am classifying tweets using single-valued features (e.g. the number of followers of the user) and multiple-valued features (e.g. some long histogram, from LDA or Bag-of-Words for example).

I simply concatenate the features, modelling each component of each multiple-valued feature as a Weka Attribute. I am using SVM and Naive Bayes.

The issue is this: I want to evaluate the attributes with Weka classes, I want to rank the multiple-valued feature as a single attribute (no sense get that BoF_1342 is better then LDA_4103 and BoF_242, and I only want know that BoF is better then LDA).

Does Weka support this kind of evaluation?

Upvotes: 1

Views: 325

Answers (1)

Oscar Picchi
Oscar Picchi

Reputation: 174

I'm not sure WEKA supports this kind of aggregation. The solution in this case would be for you to create a script that does it for you (not sure if you can even do that in you scenario) Ex: you have 4 attributes and three instances (BoF_1342,LDA_4103,BoF_242)

BoF_1342 0 - 0 - 0 - 1

LDA_4103 1 - 1 - 0 - 1

BoF_242 0 - 1 - 1 - 0

it would become

BoF 0 - 1 - 1 - 1

LDA 1 - 1 - 0 - 1

Upvotes: 0

Related Questions