TMVA
The Toolkit for Multivariate Data Analysis with ROOT (TMVA) provides a machine learning environment for the processing and evaluation of multivariate classification, both binary and multiclass, and regression techniques targeting applications in highenergy physics. The package includes:
 Neural networks
 Deep networks
 Multilayer perceptron
 Boosted/Bagged decision trees
 Function discriminant analysis (FDA)
 Multidimensional probability density estimation (PDE  rangesearch approach)
 Multidimensional knearest neighbor classifier
 Predictive learning via rule ensembles (RuleFit)
 Projective likelihood estimation (PDE approach)
 Rectangular cut optimisation
 Support Vector Machine (SVM)
Topical manuals
For TMVA, topical manuals are available at Topical Manuals  TMVA.
They contain indepth information about TMVA.
TMVA in the ROOT forum
Discuss TMVA in the ROOT forum.
Using TMVA
The ROOT tutorials for TMVA, available in $ROOTSYS/tutorials/tmva
, provide example jobs for the training phase and the application of the training results in a classification or regression analysis using the TMVA Reader.
Training examples

TMVAClassification.C provides examples for the training and testing of TMVA classifiers.

TMVAMulticlass.C provides an example for the training and testing of a TMVA multiclass classification.

TMVARegression.C provides examples for the training and testing of TMVA classifiers.
Application examples

TMVAClassificationApplication.C provides an example on how to use the trained classifiers within an analysis module.

TMVAMulticlassApplication.C provides an example on how to use the trained multiclass classifiers within an analysis module.

TMVARegressionApplication.C provides an example on how to use the trained regression MVAs within an analysis module.
Example
TMVAClassification.C uses an academic toy data set for training and testing. It consists of four linearly correlated, Gaussian distributed discriminating input variables, with different sample means for signal and background.
The training job provides a formatted output logging that contains the following information:
 Linear correlation matrices for the input variables.
 Correlation ratios and mutual information between input variables and regression targets.
 Variable ranking.
 Summaries of the MVA configurations.
 Goodnessoffit evaluation for PDFs.
 Signal and background (or regression target) correlations between the various MVA methods.
 Decision overlaps.
 Signal efficiencies at benchmark background rejection rates.
 Other performance estimators.
After a successful training, TMVA provides so called “weight”files (here in the TMVA.root
ROOT file) that contain all information necessary to recreate the method without retraining.
In addition, a GUI is provided to execute macros for displaying training, test and evaluation results.
Figure: GUI for TMVA.
You can, for example, plot input variable distributions.
Figure: Example plots for input variable distributions.