Re: [TMVA-users] Multiple targets in Boosted Regression Trees

[Quentin B. <que...@ce...>]

Hi Helge,

Thanks for confirming !

I may give it a try. I think I have a pretty clear idea of what to do for the ‘easy’ part of the modifications: there is a few located place in MethodBDT.cxx where I could introduce an iteration over the targets.
However I have to admit I am a bit fuzzy on the output writing and reading part. Any recommendations on how to tackle that ?
I see that multiple target is already implemented in PDEFoam but I don’t know if this really helps to modify the BRT output.

Cheers,
Quentin

> On Mar 3, 2015, at 12:37 AM, Helge Voss <Hel...@ce...> wrote:
> 
> Hello Quentin,
> 
> you are right, TMVA does not support boosted regression trees with
> multiple target outputs. A standard way how to implement this would be
> "simply" to:
> 
> - replace for each leaf node the target value by a 'target-vector'
> - replace the regression separation calulator (used to determine the
> split) from a simple
>   sum_events (y_event - <y>)^2
> 
>   by
> 
>   sum_events ( sum_targets ( y_event_target - <y>_target)^2
> 
> it shouldn't be 'too' much work but unfortunatly I don't have the time
> to do that. So if you'd be willing to give it a go ?
> 
> (note: while 'multiclass' is implemented in TMVA as a 'special' case
> of classification, the 'regression' interface seems to directly allow
> for the implementation of multi-dimensional output, hence the main
> work would simply be the modification of the writing of the weight
> file to 'know' about the multiple
> output values)
> 
> Thinking about it .. i really doesn't seem to be a major work, but still ..
> 
> Cheers,
> 
> helge
> 
> 
> On 3 March 2015 at 03:18, Quentin Buat <que...@ce...> wrote:
>> Dear TMVA developers,
>> 
>> I’m writing to seek information about the Boosted Regression Tree part of
>> the package.
>> As far as I understood from the documentation and some tests, the method BDT
>> is not capable to handle regression with multiple targets.
>> 
>> (--- <WARNING> Factory                  : Method BDT is not capable of
>> handling regression with 4 targets.)
>> 
>> I know this feature is available in scikit-learn
>> http://scikit-learn.org/stable/auto_examples/tree/plot_tree_regression_multioutput.html#example-tree-plot-tree-regression-multioutput-py
>> and I was wondering if this is something you were considering adding to TMVA
>> in the near future ? Also, if you don’t have any plans to implement this,
>> some guidance on how to do it and how involved it is would be great !
>> 
>> Thanks in advance.
>> 
>> Cheers,
>> Quentin
>> 
>> 
>> 
>> ------------------------------------------------------------------------------
>> Dive into the World of Parallel Programming The Go Parallel Website,
>> sponsored
>> by Intel and developed in partnership with Slashdot Media, is your hub for
>> all
>> things parallel software development, from weekly thought leadership blogs
>> to
>> news, videos, case studies, tutorials and more. Take a look and join the
>> conversation now. http://goparallel.sourceforge.net/
>> _______________________________________________
>> TMVA-users mailing list
>> TMV...@li...
>> https://lists.sourceforge.net/lists/listinfo/tmva-users
>>