Hi Kathy.
Why do you want to do that? The Chi2 Kernel is only defined for
non-negative data, so it makes sense that the approximation only works with
non-negative data. The Chi2 Kernel is mostly used for histogram data.
Cheers,
Andy
On 10/14/2014 09:13 PM, Kathy Hida wrote:
In Scikit-learn
In Scikit-learn, approximating the additive kernel of a matrix with
negative values using AdditiveChi2Sampler is not supported.
What is the best way to deal with this? Does it make sense if we just pass
the absolute of the original matrix to the AdditiveChi2Sampler method?
Thank you,
