TMVA Configuration Options Reference Reference version: TMVA-v4.2.0 TMVA-version @ ROOT

Reference for configuration options defined in the option string of each MVA method booking, and for the definition of data sets used for training and testing (Factory).

Table fields:

Option: The option identifier in the option string (given, e.g., in "factory->BookMethod(...)" call).
Array: Can the option be set individually for each input variable via the "[i]" tag, where "i" is the ith variable?
Default value: Value used if option is not explicitly set in the configuration option string.
Predefined values: Options can be categories of predefined values among which the user must choose.
Description: Info about the option.

Colour codes:

Greenish rows: Options shared by all MVA methods (through common base class).
Bluish rows: Specific MVA options.
Yellowish rows: Configuration options for minimiser (fitter) classes.
Redish rows: Options for other configurable classes.

Available MVA methods (1st row), minimisation tools (2nd row), and other configurables (3rd row):

[MVA::HMatrix] [MVA::Fisher] [MVA::PDERS] [MVA::FDA] [MVA::LD] [MVA::SVM] [MVA::CFMlpANN] [MVA::KNN] [MVA::BDT] [MVA::Boost] [MVA::RuleFit] [MVA::Likelihood] [MVA::MLP] [MVA::Cuts] [MVA::PDEFoam] [MVA::TMlpANN]
[Fitter_SA] [Fitter_MC] [Fitter_Minuit] [Fitter_GA]
[DataSetFactory] [PDF] [Factory]

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: HMatrix
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: Fisher
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
Method No Fisher Fisher, Mahalanobis Discrimination method

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: PDERS
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
VolumeRangeMode No Adaptive Unscaled, MinMax, RMS, Adaptive, kNN Method to determine volume size
KernelEstimator No Box Box, Sphere, Teepee, Gauss, Sinc3, Sinc5, Sinc7, Sinc9, Sinc11, Lanczos2, Lanczos3, Lanczos5, Lanczos8, Trim Kernel estimation function
DeltaFrac No 3 nEventsMin/Max for minmax and rms volume range
NEventsMin No 100 nEventsMin for adaptive volume range
NEventsMax No 200 nEventsMax for adaptive volume range
MaxVIterations No 150 MaxVIterations for adaptive volume range
InitialScale No 0.99 InitialScale for adaptive volume range
GaussSigma No 0.1 Width (wrt volume size) of Gaussian kernel estimator
NormTree No False Normalize binary search tree

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: FDA
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
Formula No (0) The discrimination formula
ParRanges No () Parameter ranges
FitMethod No MINUIT MC, GA, SA, MINUIT Optimisation Method
Converger No None None, MINUIT FitMethod uses Converger to improve result

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: LD
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: SVM
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
Gamma No 1 RBF kernel parameter: Gamma (size of the Kernel)
C No 1 Cost parameter
Tol No 0.01 Tolerance parameter
MaxIter No 1000 Maximum number of training loops

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: CFMlpANN
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
NCycles No 3000 Number of training cycles
HiddenLayers No N,N-1 Specification of hidden layer architecture

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: KNN
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
nkNN No 20 Number of k-nearest neighbors
BalanceDepth No 6 Binary tree balance depth
ScaleFrac No 0.8 Fraction of events used to compute variable width
SigmaFact No 1 Scale factor for sigma in Gaussian kernel
Kernel No Gaus Use polynomial (=Poln) or Gaussian (=Gaus) kernel
Trim No False Use equal number of signal and background events
UseKernel No False Use polynomial kernel weight
UseWeight No True Use weight to count kNN events
UseLDA No False Use local linear discriminant - experimental feature

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: BDT
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
NTrees No 800 Number of trees in the forest
MaxDepth No 3 Max depth of the decision tree allowed
MinNodeSize No 5% Minimum percentage of training events required in a leaf node (default: Classification: 5%, Regression: 0.2%)
nCuts No 20 Number of grid points in variable range used in finding optimal cut in node splitting
BoostType No AdaBoost AdaBoost, RealAdaBoost, Bagging, AdaBoostR2, Grad Boosting type for the trees in the forest
AdaBoostR2Loss No Quadratic Linear, Quadratic, Exponential Type of Loss function in AdaBoostR2
UseBaggedBoost No False Use only a random subsample of all events for growing the trees in each iteration.
Shrinkage No 1 Learning rate for GradBoost algorithm
AdaBoostBeta No 0.5 Learning rate for AdaBoost algorithm
UseRandomisedTrees No False Determine at each node splitting the cut variable only as the best out of a random subset of variables (like in RandomForests)
UseNvars No 2 Size of the subset of variables used with RandomisedTree option
UsePoissonNvars No True Interpret UseNvars not as fixed number but as mean of a Possion distribution in each split with RandomisedTree option
BaggedSampleFraction No 0.6 Relative size of bagged event sample to original size of the data sample (used whenever bagging is used (i.e. UseBaggedGrad, Bagging,)
UseYesNoLeaf No True Use Sig or Bkg categories, or the purity=S/(S+B) as classification of the leaf node -> Real-AdaBoost
NegWeightTreatment No InverseBoostNegWeights InverseBoostNegWeights, IgnoreNegWeightsInTraining, PairNegWeightsGlobal, Pray How to treat events with negative weights in the BDT training (particular the boosting) : IgnoreInTraining; Boost With inverse boostweight; Pair events with negative and positive weights in traning sample and *annihilate* them (experimental!)
NodePurityLimit No 0.5 In boosting/pruning, nodes with purity > NodePurityLimit are signal; background otherwise.
SeparationType No GiniIndex CrossEntropy, GiniIndex, GiniIndexWithLaplace, MisClassificationError, SDivSqrtSPlusB, RegressionVariance Separation criterion for node splitting
DoBoostMonitor No False Create control plot with ROC integral vs tree number
UseFisherCuts No False Use multivariate splits using the Fisher criterion
MinLinCorrForFisher No 0.8 The minimum linear correlation between two variables demanded for use in Fisher criterion in node splitting
UseExclusiveVars No False Variables already used in fisher criterion are not anymore analysed individually for node splitting
DoPreselection No False and and apply automatic pre-selection for 100% efficient signal (bkg) cuts prior to training
RenormByClass No False Individually re-normalize each event class to the original size after boosting
SigToBkgFraction No 1 Sig to Bkg ratio used in Training (similar to NodePurityLimit, which cannot be used in real adaboost
PruneMethod No NoPruning NoPruning, ExpectedError, CostComplexity Note: for BDTs use small trees (e.g.MaxDepth=3) and NoPruning: Pruning: Method used for pruning (removal) of statistically insignificant branches
PruneStrength No 0 Pruning strength
PruningValFraction No 0.5 Fraction of events to use for optimizing automatic pruning.
GradBaggingFraction No 0.6 deprecated: Use *BaggedSampleFraction* instead: Defines the fraction of events to be used in each iteration, e.g. when UseBaggedGrad=kTRUE.
UseNTrainEvents No 0 deprecated: Use *BaggedSampleFraction* instead: Number of randomly picked training events used in randomised (and bagged) trees

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: Boost
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
Boost_Num No 100 Number of times the classifier is boosted
Boost_MonitorMethod No True Write monitoring histograms for each boosted classifier
Boost_DetailedMonitoring No False Produce histograms for detailed boost-wise monitoring
Boost_Type No AdaBoost AdaBoost, Bagging Boosting type for the classifiers
Boost_BaggedSampleFraction No 0.6 Relative size of bagged event sample to original size of the data sample (used whenever bagging is used)
Boost_RecalculateMVACut No True Recalculate the classifier MVA Signallike cut at every boost iteration
Boost_AdaBoostBeta No 1 The ADA boost parameter that sets the effect of every boost step on the events' weights
Boost_Transform No step step, linear, log, gauss Type of transform applied to every boosted method linear, log, step
Boost_RandomSeed No 0 Seed for random number generator used for bagging

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: RuleFit
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
GDTau No -1 Gradient-directed (GD) path: default fit cut-off
GDTauPrec No 0.01 GD path: precision of tau
GDStep No 0.01 GD path: step size
GDNSteps No 10000 GD path: number of steps
GDErrScale No 1.1 Stop scan when error > scale*errmin
LinQuantile No 0.025 Quantile of linear terms (removes outliers)
GDPathEveFrac No 0.5 Fraction of events used for the path search
GDValidEveFrac No 0.5 Fraction of events used for the validation
fEventsMin No 0.1 Minimum fraction of events in a splittable node
fEventsMax No 0.9 Maximum fraction of events in a splittable node
nTrees No 20 Number of trees in forest.
ForestType No AdaBoost AdaBoost, Random Method to use for forest generation (AdaBoost or RandomForest)
RuleMinDist No 0.001 Minimum distance between rules
MinImp No 0.01 Minimum rule importance accepted
Model No ModRuleLinear ModRule, ModRuleLinear, ModLinear Model to be used
RuleFitModule No RFTMVA RFTMVA, RFFriedman Which RuleFit module to use
RFWorkDir No ./rulefit Friedman's RuleFit module (RFF): working dir
RFNrules No 2000 RFF: Mximum number of rules
RFNendnodes No 4 RFF: Average number of end nodes

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: Likelihood
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
TransformOutput No False Transform likelihood output by inverse sigmoid function

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: MLP
Option Array Default value Predefined values Description
NCycles No 500 Number of training cycles
HiddenLayers No N,N-1 Specification of hidden layer architecture
NeuronType No sigmoid Neuron activation function type
RandomSeed No 1 Random seed for initial synapse weights (0 means unique seed for each run; default value '1')
EstimatorType No MSE MSE, CE, linear, sigmoid, tanh, radial MSE (Mean Square Estimator) for Gaussian Likelihood or CE(Cross-Entropy) for Bernoulli Likelihood
NeuronInputType No sum sum, sqsum, abssum Neuron input function type
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
TrainingMethod No BP BP, GA, BFGS Train with Back-Propagation (BP), BFGS Algorithm (BFGS), or Genetic Algorithm (GA - slower and worse)
LearningRate No 0.02 ANN learning rate parameter
DecayRate No 0.01 Decay rate for learning parameter
TestRate No 10 Test for overtraining performed at each #th epochs
EpochMonitoring No False Provide epoch-wise monitoring plots according to TestRate (caution: causes big ROOT output file!)
Sampling No 1 Only 'Sampling' (randomly selected) events are trained each epoch
SamplingEpoch No 1 Sampling is used for the first 'SamplingEpoch' epochs, afterwards, all events are taken for training
SamplingImportance No 1 The sampling weights of events in epochs which successful (worse estimator than before) are multiplied with SamplingImportance, else they are divided.
SamplingTraining No True The training sample is sampled
SamplingTesting No False The testing sample is sampled
ResetStep No 50 How often BFGS should reset history
Tau No 3 LineSearch size step
BPMode No sequential sequential, batch Back-propagation learning mode: sequential or batch
BatchSize No -1 Batch size: number of events/batch, only set if in Batch Mode, -1 for BatchSize=number_of_events
ConvergenceImprove No 1e-30 Minimum improvement which counts as improvement (<0 means automatic convergence check is turned off)
ConvergenceTests No -1 Number of steps (without improvement) required for convergence (<0 means automatic convergence check is turned off)
UseRegulator No False Use regulator to avoid over-training
UpdateLimit No 10000 Maximum times of regulator update
CalculateErrors No False Calculates inverse Hessian matrix at the end of the training to be able to calculate the uncertainties of an MVA value
WeightRange No 1 Take the events for the estimator calculations from small deviations from the desired value to large deviations only over the weight range

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: Cuts
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
FitMethod No GA GA, SA, MC, MCEvents, MINUIT, EventScan Minimisation Method (GA, SA, and MC are the primary methods to be used; the others have been introduced for testing purposes and are depreciated)
EffMethod No EffSel EffSel, EffPDF Selection Method
CutRangeMin Yes -1 Minimum of allowed cut range (set per variable)
CutRangeMax Yes -1 Maximum of allowed cut range (set per variable)
VarProp Yes NotEnforced NotEnforced, FMax, FMin, FSmart Categorisation of cuts

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: PDEFoam
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
SigBgSeparate No False Separate foams for signal and background
TailCut No 0.001 Fraction of outlier events that are excluded from the foam in each dimension
VolFrac No 0.0666667 Size of sampling box, used for density calculation during foam build-up (maximum value: 1.0 is equivalent to volume of entire foam)
nActiveCells No 500 Maximum number of active cells to be created by the foam
nSampl No 2000 Number of generated MC events per cell
nBin No 5 Number of bins in edge histograms
Compress No True Compress foam output file
MultiTargetRegression No False Do regression with multiple targets
Nmin No 100 Number of events in cell required to split cell
MaxDepth No 0 Maximum depth of cell tree (0=unlimited)
FillFoamWithOrigWeights No False Fill foam with original or boost weights
UseYesNoCell No False Return -1 or 1 for bkg or signal like events
DTLogic No None None, GiniIndex, MisClassificationError, CrossEntropy, GiniIndexWithLaplace, SdivSqrtSplusB Use decision tree algorithm to split cells
Kernel No None None, Gauss, LinNeighbors Kernel type used
TargetSelection No Mean Mean, Mpv Target selection method

Configuration options for MVA method : Information on method tuning
Configuration options reference for MVA method: TMlpANN
Option Array Default value Predefined values Description
V No False Verbose output (short form of VerbosityLevel below - overrides the latter one)
VerbosityLevel No Default Default, Debug, Verbose, Info, Warning, Error, Fatal Verbosity level
VarTransform No None List of variable transformations performed before training, e.g., D_Background,P_Signal,G,N_AllClasses for: Decorrelation, PCA-transformation, Gaussianisation, Normalisation, each for the given class of events ('AllClasses' denotes all events of all classes, if no class indication is given, 'All' is assumed)
H No False Print method-specific help message
CreateMVAPdfs No False Create PDFs for classifier outputs (signal and background)
IgnoreNegWeightsInTraining No False Events with negative weights are ignored in the training (but are included for testing and performance evaluation)
NCycles No 200 Number of training cycles
HiddenLayers No N,N-1 Specification of hidden layer architecture (N stands for number of variables; any integers may also be used)
ValidationFraction No 0.5 Fraction of events in training tree used for cross validation
LearningMethod No Stochastic Stochastic, Batch, SteepestDescent, RibierePolak, FletcherReeves, BFGS Learning method

Configuration options for setup and tuning of specific fitter :
Configuration options reference for fitting method: Simulated Annealing (SA)
Option Array Default value Predefined values Description
MaxCalls No 100000 Maximum number of minimisation calls
InitialTemp No 1e+06 Initial temperature
MinTemp No 1e-06 Mimimum temperature
Eps No 1e-10 Epsilon
TempScale No 1 Temperature scale
AdaptiveSpeed No 1 Adaptive speed
TempAdaptiveStep No 0.009875 Step made in each generation temperature adaptive
UseDefaultScale No False Use default temperature scale for temperature minimisation algorithm
UseDefaultTemp No False Use default initial temperature
KernelTemp No IncAdaptive IncAdaptive, DecAdaptive, Sqrt, Log, Sin, Homo, Geo Temperature minimisation algorithm

Configuration options for setup and tuning of specific fitter :
Configuration options reference for fitting method: Monte Carlo sampling (MC)
Option Array Default value Predefined values Description
SampleSize No 100000 Number of Monte Carlo events in toy sample
Sigma No -1 If > 0: new points are generated according to Gauss around best value and with Sigma in units of interval length
Seed No 100 Seed for the random generator (0 takes random seeds)

Configuration options for setup and tuning of specific fitter :
Configuration options reference for fitting method: TMinuit (MT)
Option Array Default value Predefined values Description
ErrorLevel No 1 TMinuit: error level: 0.5=logL fit, 1=chi-squared fit
PrintLevel No -1 TMinuit: output level: -1=least, 0, +1=all garbage
FitStrategy No 2 TMinuit: fit strategy: 2=best
PrintWarnings No False TMinuit: suppress warnings
UseImprove No True TMinuit: use IMPROVE
UseMinos No True TMinuit: use MINOS
SetBatch No False TMinuit: use batch mode
MaxCalls No 1000 TMinuit: approximate maximum number of function calls
Tolerance No 0.1 TMinuit: tolerance to the function value at the minimum

Configuration options for setup and tuning of specific fitter :
Configuration options reference for fitting method: Genetic Algorithm (GA)
Option Array Default value Predefined values Description
PopSize No 300 Population size for GA
Steps No 40 Number of steps for convergence
Cycles No 3 Independent cycles of GA fitting
SC_steps No 10 Spread control, steps
SC_rate No 5 Spread control, rate: factor is changed depending on the rate
SC_factor No 0.95 Spread control, factor
ConvCrit No 0.001 Convergence criteria
SaveBestGen No 1 Saves the best n results from each generation. They are included in the last cycle
SaveBestCycle No 10 Saves the best n results from each cycle. They are included in the last cycle. The value should be set to at least 1.0
Trim No False Trim the population to PopSize after assessing the fitness of each individual
Seed No 100 Set seed of random generator (0 gives random seeds)

Configuration options given in the "PrepareForTrainingAndTesting" call; these options define the creation of the data sets used for training and expert validation by TMVA :
Configuration options reference for class: DataSetFactory
Option Array Default value Predefined values Description
SplitMode No Random Random, Alternate, Block Method of picking training and testing events (default: random)
MixMode No SameAsSplitMode SameAsSplitMode, Random, Alternate, Block Method of mixing events of differnt classes into one dataset (default: SameAsSplitMode)
SplitSeed No 100 Seed for random event shuffling
NormMode No EqualNumEvents None, NumEvents, EqualNumEvents Overall renormalisation of event-by-event weights used in the training (NumEvents: average weight of 1 per event, independently for signal and background; EqualNumEvents: average weight of 1 per event for signal, and sum of weights for background equal to sum of weights for signal)
nTrain_Signal No 0 Number of training events of class Signal (default: 0 = all)
nTest_Signal No 0 Number of test events of class Signal (default: 0 = all)
nTrain_Background No 0 Number of training events of class Background (default: 0 = all)
nTest_Background No 0 Number of test events of class Background (default: 0 = all)
V No False Verbosity (default: true)
VerboseLevel No Info Debug, Verbose, Info VerboseLevel (Debug/Verbose/Info)

Configuration options for the PDF class :
Configuration options reference for class: PDF
Option Array Default value Predefined values Description
NSmooth No 0 Number of smoothing iterations for the input histograms
MinNSmooth No -1 Min number of smoothing iterations, for bins with most data
MaxNSmooth No -1 Max number of smoothing iterations, for bins with least data
NAvEvtPerBin No 50 Average number of events per PDF bin
Nbins No 0 Defined number of bins for the histogram from which the PDF is created
CheckHist No False Whether or not to check the source histogram of the PDF
PDFInterpol No Spline2 Spline0, Spline1, Spline2, Spline3, Spline5, KDE Interpolation method for reference histograms (e.g. Spline2 or KDE)
KDEtype No Gauss Gauss KDE kernel type (1=Gauss)
KDEiter No Nonadaptive Nonadaptive, Adaptive Number of iterations (1=non-adaptive, 2=adaptive)
KDEFineFactor No 1 Fine tuning factor for Adaptive KDE: Factor to multyply the width of the kernel
KDEborder No None None, Renorm, Mirror Border effects treatment (1=no treatment , 2=kernel renormalization, 3=sample mirroring)

Configuration options for Factory running :
Configuration options reference for class: Factory
Option Array Default value Predefined values Description
V No False Verbose flag
Color No True Flag for coloured screen output (default: True, if in batch mode: False)
Transformations No List of transformations to test; formatting example: Transformations=I;D;P;U;G,D, for identity, decorrelation, PCA, Uniform and Gaussianisation followed by decorrelation transformations
Silent No False Batch mode: boolean silent flag inhibiting any output from TMVA after the creation of the factory class object (default: False)
DrawProgressBar No True Draw progress bar to display training, testing and evaluation schedule (default: True)
AnalysisType No Auto Classification, Regression, Multiclass, Auto Set the analysis type (Classification, Regression, Multiclass, Auto) (default: Auto)


Page created on Mon Jul 29 00:06:19 2013 (© TMVA, 2006−2013)