Re: [TMVA-users] Problem with BDTG in TMVA v4.1.2

[Olivier D. <dav...@lp...>]

Hello Helge,

Thank you for your help!
I checked the category definitions and the number of entries in the 
original trees. There are events in there.
I also tried the other modifications, without any success... Finally I 
replaced my signal tree by the background tree (again), just to see what 
happens, and I get the same error...

Cheers,

-Olivier.

Le 28/11/11 19:36, Helge Voss a écrit :
> Hi Olivier,
>
> I would guess from from the log output, that one of the categories
> you've defined simply does
> not have any events in it. Guess that the other classifiers don't
> worriy about this as their initialisation
> does not require at least "some" events...
>
> Another hint: You should NOT use pruning (you've explicitely specified
> ExpectedError  Pruning, why?)
> and ... it's better to specify "MaxDepth=3"  oder 2  anstatt
> NNodesMax=5 (as the latter gives lopsided
> triess..)
>
>                   cheers,
>
>                          Helge
>
>
>
> On 28 November 2011 18:36, Olivier Davignon<dav...@lp...>  wrote:
>> Dear TMVA-users,
>>
>> I'm currently trying to use a BDTG discriminant, in the context of
>> categorization (TMVAClassificationCategory.cxx), using TMVA v4.1.2.
>>
>> I get an error while Training the BDT:
>>
>> ---<WARNING>  BDTGCat                  : Method Category_BDTG_1 not
>> trained (training tree has less entries [1] than required [10]
>> ---<WARNING>  BDTGCat                  : Method Category_BDTG_2 not
>> trained (training tree has less entries [1] than required [10]
>> --- BDTGCat                  : Begin ranking of input variables...
>> --- BDTGCat                  : End of training
>> --- BDTGCat                  : Elapsed time for training with 31476
>> events: 0.000193 sec
>> --- BDTGCat                  : Create MVA output for classification on
>> training sample
>> --- BDTGCat                  : Evaluation of BDTGCat on training sample
>> (31476 events)
>> ---<FATAL>  Rearrange                : Transformation not yet created
>>
>> Here is how I build the discriminant:
>>
>>    TMVA::MethodBase* BDTCat = factory->BookMethod(
>> TMVA::Types::kCategory, "BDTGCat","" );
>>    mcat = dynamic_cast<TMVA::MethodCategory*>(BDTCat);
>>    mcat->AddMethod("NumberJets==1", theCat1Vars, TMVA::Types::kBDT,
>> "Category_BDTG_1",
>> "H:NTrees=1000:BoostType=Grad:Shrinkage=0.10:UseBaggedGrad:GradBaggingFraction=0.5:nCuts=20:NNodesMax=5:PruneMethod=ExpectedError");
>>    mcat->AddMethod("NumberJets>=2", theCat2Vars, TMVA::Types::kBDT,
>> "Category_BDTG_2",
>> "H:NTrees=1000:BoostType=Grad:Shrinkage=0.10:UseBaggedGrad:GradBaggingFraction=0.5:nCuts=20:NNodesMax=5:PruneMethod=ExpectedError");
>>
>> Do you have any idea of what could be the problem? I tried to train a
>> KNN or Cuts, and it works fine with the same inputs. I also tried a BDTG
>> a few weeks ago, using something really similar, without any problems.
>> Maybe there's something obvious for you, but I don't see it.
>>
>> Thank you,
>>
>> -Olivier.
>>
>>
>> ------------------------------------------------------------------------------
>> All the data continuously generated in your IT infrastructure
>> contains a definitive record of customers, application performance,
>> security threats, fraudulent activity, and more. Splunk takes this
>> data and makes sense of it. IT sense. And common sense.
>> http://p.sf.net/sfu/splunk-novd2d
>> _______________________________________________
>> TMVA-users mailing list
>> TMV...@li...
>> https://lists.sourceforge.net/lists/listinfo/tmva-users
>>