"Zero Weights in LibSVMLearner / JMySVMLearner"

Shubha

Hi,

I get all weights to be zero (i have used 200 attributes) in the LibSVM / JMySVM classification procedures. And this happens only for certain data. What are the circumstances where the weights corresponding to all the attributes will be zero? (Any specific cases?)

For example: I used LibSVMLearner for a data and I get non-zero weights for most of the attributes. But if I use JMySVMLearner for the same data, I get the weights of all 200 attributes to be zero. What could be the problem?....


Thanks,
Shubha

fischer

Shubha,

please attach a process files. Are you referring to AttributeWeights? The LibSVM doesn't provide any.

Could it be the weights are only displayed as 0, simply because they are so small, probably below <0.001? You can specify the number of displayed digits in the RapidMiner preferences.

Cheers,
Simon

Shubha

Hi Simon, Thanks.

Since the weights given in the 'AttributeWeights' object and the 'Text View / Weight Ttable' of the Kernel model object are the same, I was referring to these weights.

These weights in the text view has the values 0.000. And in the Weight Table view, they are displayed as 0. I even put this into excel to see the actual decimal points, but they are again displayed as 0.

So, at what circumstances all the weights will be zero according to the internal codings of the learners?

Thanks a lot,
Shubha