class TMVA::MethodBDT: public TMVA::MethodBase

 the standard constructor for the "boosted decision trees"

 BDT can handle classification with multiple classes and regression with one regression-target

 define the options (their key words) that can be set in the option string
 know options:
 nTrees        number of trees in the forest to be created
 BoostType     the boosting type for the trees in the forest (AdaBoost e.t.c..)
                  known: AdaBoost
                         AdaBoostR2 (Adaboost for regression)
                         Bagging
                         GradBoost
 AdaBoostBeta     the boosting parameter, beta, for AdaBoost
 UseRandomisedTrees  choose at each node splitting a random set of variables
 UseNvars         use UseNvars variables in randomised trees
 UsePoission Nvars use UseNvars not as fixed number but as mean of a possion distribution
 SeparationType   the separation criterion applied in the node splitting
                  known: GiniIndex
                         MisClassificationError
                         CrossEntropy
                         SDivSqrtSPlusB
 MinNodeSize:     minimum percentage of training events in a leaf node (leaf criteria, stop splitting)
 nCuts:           the number of steps in the optimisation of the cut for a node (if < 0, then
                  step size is determined by the events)
 UseFisherCuts:   use multivariate splits using the Fisher criterion
 UseYesNoLeaf     decide if the classification is done simply by the node type, or the S/B
                  (from the training) in the leaf node
 NodePurityLimit  the minimum purity to classify a node as a signal node (used in pruning and boosting to determine
                  misclassification error rate)
 PruneMethod      The Pruning method:
                  known: NoPruning  // switch off pruning completely
                         ExpectedError
                         CostComplexity
 PruneStrength    a parameter to adjust the amount of pruning. Should be large enough such that overtraining is avoided.
 PruningValFraction   number of events to use for optimizing pruning (only if PruneStrength < 0, i.e. automatic pruning)
 NegWeightTreatment      IgnoreNegWeightsInTraining  Ignore negative weight events in the training.
                         DecreaseBoostWeight     Boost ev. with neg. weight with 1/boostweight instead of boostweight
                         PairNegWeightsGlobal    Pair ev. with neg. and pos. weights in traning sample and "annihilate" them
 MaxDepth         maximum depth of the decision tree allowed before further splitting is stopped

 options that are used ONLY for the READER to ensure backward compatibility

 the option string is decoded, for available options see "DeclareOptions"

 common initialisation with defaults for the BDT-Method

 reset the method, as if it had just been instantiated (forget all training etc.)

destructor
   Note: fEventSample and ValidationSample are already deleted at the end of TRAIN
         When they are not used anymore
   for (UInt_t i=0; i<fEventSample.size();      i++) delete fEventSample[i];
   for (UInt_t i=0; i<fValidationSample.size(); i++) delete fValidationSample[i];

 initialize the event sample (i.e. reset the boost-weights... etc)

 o.k. you know there are events with negative event weights. This routine will remove
 them by pairing them with the closest event(s) of the same event class with positive
 weights
 A first attempt is "brute force", I dont' try to be clever using search trees etc,
 just quick and dirty to see if the result is any good

 call the Optimzier with the set of paremeters and ranges that
 are meant to be tuned.

 set the tuning parameters accoding to the argument

 BDT training

returns MVA value: -1 for background, 1 for signal

Calculate residua for all events;

Calculate current residuals for all events and update targets for next iteration

calculates the quantile of the distribution of the first pair entries weighted with the values in the second pair entries

Calculate the desired response value for each region

 Implementation of M_TreeBoost using a Huber loss function as desribed by Friedman 1999

 initialize targets for first tree

 test the tree quality.. in terms of Miscalssification

 apply the boosting alogrithim (the algorithm is selecte via the the "option" given
 in the constructor. The return value is the boosting weight

 fills the ROCIntegral vs Itree from the testSample for the monitoring plots
 during the training .. but using the testing events

 the AdaBoost implementation.
 a new training sample is generated by weighting
 events that are misclassified by the decision tree. The weight
 applied is w = (1-err)/err or more general:
            w = ((1-err)/err)^beta
 where err is the fraction of misclassified events in the tree ( <0.5 assuming
 demanding the that previous selection was better than random guessing)
 and "beta" being a free parameter (standard: beta = 1) that modifies the
 boosting.

 the AdaCost boosting algorithm takes a simple cost Matrix  (currently fixed for
 all events... later could be modified to use individual cost matrices for each
 events as in the original paper...

                   true_signal true_bkg

     sel_signal |   Css         Ctb_ss    Cxx.. in the range [0,1]
     sel_bkg    |   Cts_sb      Cbb

    and takes this into account when calculating the misclass. cost (former: error fraction):

    err = sum_events ( weight* y_true*y_sel * beta(event)

 call it boot-strapping, re-sampling or whatever you like, in the end it is nothing
 else but applying "random" poisson weights to each event.

 fills fEventSample with fBaggedSampleFraction*NEvents random training events

 a special boosting only for Regression ...
 maybe I'll implement it later...

 adaption of the AdaBoost to regression problems (see H.Drucker 1997)

 write weights to XML

 reads the BDT from the xml file

 read the weights (BDT coefficients)

 Return the MVA value (range [-1;1]) that classifies the
 event according to the majority vote from the total number of
 decision trees.

 Return the MVA value (range [-1;1]) that classifies the
 event according to the majority vote from the total number of
 decision trees.

 get the multiclass MVA response for the BDT classifier

 get the regression value generated by the BDTs

 Here we could write some histograms created during the processing
 to the output file.

 Return the relative variable importance, normalized to all
 variables together having the importance 1. The importance in
 evaluated as the total separation-gain that this variable had in
 the decision trees (weighted by the number of events)

 Returns the measure for the variable importance of variable "ivar"
 which is later used in GetVariableImportance() to calculate the
 relative variable importances.

 Compute ranking of input variables

 Get help message text

 typical length of text line:
         "|--------------------------------------------------------------|"

 make ROOT-independent C++ class for classifier response (classifier-specific implementation)

 specific class header

 recursively descends a tree and writes the node instance to the output streem

 find useful preselection cuts that will be applied before
 and Decision Tree training.. (and of course also applied
 in the GetMVA .. --> -1 for background +1 for Signal
   /*

 aply the  preselection cuts before even bothing about any
 Decision Trees  in the GetMVA .. --> -1 for background +1 for Signal

 constructor for training and reading

 get the actual forest size (might be less than fNTrees, the requested one, if boosting is stopped early