\section{The TMVA Methods}

\label{sec:tmvaClassifiers}

All TMVA classification and regression methods (in most cases, a method serves both

analysis goals) inherit from \code{MethodBase}, which implements basic

functionality like the interpretation of common configuration options, the

interaction with the training and test data sets, I/O operations and common

performance evaluation calculus. The functionality each MVA method is required

to implement is defined in the abstract interface \code{IMethod}.\footnote

{

Two constructors are

implemented for each method: one that creates the method for

a first time for training with a configuration (``option'') string

among the arguments, and another that recreates a method from an

existing weight file. The use of the first constructor is

demonstrated in the example macros \code{TMVAClassification.C} and

\code{TMVARegression.C}, while the second one is employed by the Reader in

\code{TMVAClassificationApplication.C} and \code{TMVARegressionApplication.C}.

Other functions implemented by each methods are: \code{Train}

(called for training), \code{Write/ReadWeightsToStream} (I/O of

specific training results), \code{WriteMonitoringHistosToFile}

(additional specific information for monitoring purposes) and

\code{CreateRanking} (variable ranking).

}

Each MVA method provides a function that creates a rank object (of

type \code{Ranking}), which is an ordered list of the input variables

prioritised according to criteria specific to that method. Also

provided are brief method-specific help notes (option \code{Help},

switched off by default) with information on the adequate usage of

the method and performance optimisation in case of unsatisfying

results.

If the option \code{CreateMVAPdfs} is set TMVA creates signal and

background PDFs from the corresponding MVA response

distributions using the training sample (\cf\

Sec.~\ref{sec:usingtmva:training}). The binning and smoothing

properties of the underlying histograms can be customised via controls

implemented in the \code{PDF} class (\cf\ Sec.~\ref{sec:PDF} and Option

Table~\ref{opt:pdf} on page~\pageref{opt:pdf}). The options specific to

\code{MethodBase} are listed in Option Table~\ref{opt:mva::methodbase}.

They can be accessed by all MVA methods.

The following sections describe the methods implemented in TMVA. For each method we

proceed according to the following scheme: ($i$) a brief introduction, ($ii$) the description

of the booking options required to configure the method, ($iii$) a description of the

the method and TMVA implementation specifications for classification and -- where

available -- for regression, ($iv$) the properties of the

variable ranking, and ($v$) a few comments on performance, favourable (and

disfavoured) use cases, and comparisons with other methods.

% ======= input option table ==========================================

\begin{option}[!t]

\input optiontables/MVA__MethodBase.tex

\caption[.]{\optionCaptionSize

Configuration options that are common for all classifiers (but which can be controlled individually

for each classifier). Values given are defaults. If predefined categories exist, the default category

is marked by a ’*’. The lower options in the table control the PDF fitting of the classifiers (required,

\eg, for the Rarity calculation).

}

\label{opt:mva::methodbase}

\end{option}

% =====================================================================