Search code examples
performancemachine-learningscikit-learnmultilabel-classificationsvmlight

Performance SVMlight Multilabel Classificaiton (Features: 1000)

I am currently working on a multi-label classification. As classifier I use SVMlight and binarized the multi-label problem. That means in my case more than 1000 classification runs.

I have a fixed feature size of 1000, the training of the models takes now more than 3 days (2 GHz Intel Core 2 Duo, 8 GB). Has anybody experience if it is common that it takes that long or can you give me any advice how to boost the performance? Do you think a change to a high-performance machine would have a great effect?

Thanks in advance!

Solution

  • Feature normalization solved my problem, the performance was much faster after min/max ([-1,1]) normalization in the feature space.