Hi Amir.
1) As far as I know, the gradient boosting works only with trees using deviance 
or least squares regression.
I don't think it should be hard to add other losses, though.

2) There are at the moment no plans to add structured SVMs to the library. The 
reason is that structured
models usually are very problem specific. It is possible to build generic 
frameworks like Joachsim SVMstruct,
which works by the user specifying functions for features, inference and 
loss-augmented inference,
but this doesn't really fit well with the sklearn principle of using only 
arrays as data structures and
having a simple "fit/predict" interface.

What application did you have in mind?

In general I would love to have structured learning in sklearn, it just seems 
hard to integrate nicely.

Btw, I have some structured SVM code to play around in Python, if you want:
http://peekaboo-vision.blogspot.co.uk/2012/06/structured-svm-and-structured.html

Cheers,
Andy


----- Ursprüngliche Mail -----
Von: "amir rahimi" <[email protected]>
An: [email protected]
Gesendet: Mittwoch, 8. August 2012 12:40:52
Betreff: [Scikit-learn-general] GradientBoostingRegression loss function and    
Structured svm



Hi all, 
I have two questions/requests 

Is there any way to define arbitrary loss function for gradient boosting 
regression? e.g. using huber penalty 
My request is about adding structured output prediction for SVM in the library. 
Is there any plan for adding that? 


-- 
---------------------------------------------------------------------- 
#include <stdio.h> 
double d[]={9299037773.178347,2226415.983937417,307.0}; 
main(){d[2]--?d[0]*=4,d[1]*=5,main():printf((char*)d);} 
---------------------------------------------------------------------- 

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

------------------------------------------------------------------------------
Live Security Virtual Conference
Exclusive live event will cover all the ways today's security and 
threat landscape has changed and how IT managers can respond. Discussions 
will include endpoint security, mobile security and the latest in malware 
threats. http://www.accelacomm.com/jaw/sfrnl04242012/114/50122263/
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to