"SVM design question"
Contributor II
hi all,
I'm using SVM to design a text classifier. Since the training corpus are large and the trained SVM model has tons of attributes. More than 50% of those attribute has a weight of 0.
My question is how can i remove those attributes whose weight is 0 to simply the SVM model and the wordlist?
The SVM model outputs model/estimate_performance/weight/example_set, from what i can see, none of them can be converted to wordlist, is that correct?
Thanks for ur help.
I'm using SVM to design a text classifier. Since the training corpus are large and the trained SVM model has tons of attributes. More than 50% of those attribute has a weight of 0.
My question is how can i remove those attributes whose weight is 0 to simply the SVM model and the wordlist?
The SVM model outputs model/estimate_performance/weight/example_set, from what i can see, none of them can be converted to wordlist, is that correct?
Thanks for ur help.
Tagged:
0
Answers
RapidMiner offers different approaches to reduce the dimensionality of data. In your case you may be interested in the Weight by SVM operator combined with Select by Weights. You will find a bunch of other statistics based weighting operators in group Modeling/Attribute Weighting. Other approaches to reduce dimensionality are feature selection algorithms, which you can find in Data Transformation/Attribute Set Reduction and Transformation/Selection/Optimization. Most notably here is the Forward Selection.
Cheers,
Marius