I don't think this would work out-of-the-box.  The classic ball tree 
implementation depends on the metric satisfying the triangle 
inequality.  You may be able to cleverly modify the algorithm to work in 
other cases, but I'm not aware of any examples of that.  I think that 
approximate nearest neighbor methods would be a better bet.
   Jake

Andreas wrote:
> Hi everybody.
> While reviewing the label propagation PR, I thought about the pairwise 
> rbf functions.
> Would it be possible to compute an sparse, approximate RBF kernel matrix 
> using ball trees?
> The idea would be that if the distance between two points is some 
> "large" multiple of gamma, the kernel can be assumed
> to be zero.
> Do you think this is feasible to implement and helpful for real data?
>
> Cheers,
> Andy
>
> ------------------------------------------------------------------------------
> Try before you buy = See our experts in action!
> The most comprehensive online learning library for Microsoft developers
> is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3,
> Metro Style Apps, more. Free future releases when you subscribe now!
> http://p.sf.net/sfu/learndevnow-dev2
> _______________________________________________
> Scikit-learn-general mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
>   

------------------------------------------------------------------------------
Try before you buy = See our experts in action!
The most comprehensive online learning library for Microsoft developers
is just $99.99! Visual Studio, SharePoint, SQL - plus HTML5, CSS3, MVC3,
Metro Style Apps, more. Free future releases when you subscribe now!
http://p.sf.net/sfu/learndevnow-dev2
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to