| 1 |
benhoob |
1.1 |
Dear TMVA User,
|
| 2 |
|
|
|
| 3 |
|
|
this README contains three sections
|
| 4 |
|
|
|
| 5 |
|
|
+ Overview
|
| 6 |
|
|
+ How to run the example
|
| 7 |
|
|
+ The TMVA GUI
|
| 8 |
|
|
|
| 9 |
|
|
Overview
|
| 10 |
|
|
========
|
| 11 |
|
|
|
| 12 |
|
|
This directory contains two files that illustrate the usage of the
|
| 13 |
|
|
TMVA package to perform a MultiVariate Analysis. It also contains a
|
| 14 |
|
|
GUI for easy access to the training result.
|
| 15 |
|
|
|
| 16 |
|
|
An MV analysis is executed in two steps, the training of the MVA
|
| 17 |
|
|
methods, and the application of the best suited MVA method to your
|
| 18 |
|
|
dataset. These two steps are shown in the files TMVAClassification.C and
|
| 19 |
|
|
TMVAClassificationApplication.C, respectively.
|
| 20 |
|
|
|
| 21 |
|
|
The example uses some toy data (tmva_example.root) that comes with
|
| 22 |
|
|
the sourceforge distribution or is available on the ROOT web. The
|
| 23 |
|
|
example files contains a signal and a background tree with four
|
| 24 |
|
|
variables.
|
| 25 |
|
|
|
| 26 |
|
|
|
| 27 |
|
|
TMVA_HWW.C
|
| 28 |
|
|
-------------
|
| 29 |
|
|
|
| 30 |
|
|
TMVA_HWW.C explains the usage of the TMVA::Factory class, which
|
| 31 |
|
|
is the framework for the whole training process. The following steps
|
| 32 |
|
|
are taken:
|
| 33 |
|
|
|
| 34 |
|
|
1st: The factory is created (a name for the project is specified,
|
| 35 |
|
|
which will help with the bookkeeping of different training cycles)
|
| 36 |
|
|
|
| 37 |
|
|
2nd: The input data trees are made available to the factory ( one
|
| 38 |
|
|
can also use text files as input, also shown in this example ). The
|
| 39 |
|
|
sizes of the training and the testing data samples are also specified.
|
| 40 |
|
|
|
| 41 |
|
|
3rd: The variables that are selected to be used in the MVA are
|
| 42 |
|
|
declared to the factory
|
| 43 |
|
|
|
| 44 |
|
|
4th: The factory needs to transfer the input data into the local
|
| 45 |
|
|
data structure used for the training and testing
|
| 46 |
|
|
|
| 47 |
|
|
5th: Methods are than booked with the factory. For each method
|
| 48 |
|
|
number of options can be set that steer the training behavior. The
|
| 49 |
|
|
options are described in greater detail in the method classes and
|
| 50 |
|
|
can be looked at in the CVS web-viewer
|
| 51 |
|
|
|
| 52 |
|
|
6th: The training of all methods on the training data sample is run
|
| 53 |
|
|
by the factory. The training results for each method are stored in
|
| 54 |
|
|
the directory weights/
|
| 55 |
|
|
|
| 56 |
|
|
7th: The testing of all methods on the testing data sample is run by
|
| 57 |
|
|
the factory. The test results are stored in a root file (filename
|
| 58 |
|
|
specified in the Factory constructor)
|
| 59 |
|
|
|
| 60 |
|
|
8th: The performance of all methods is analyzed by the factory. The
|
| 61 |
|
|
performance results are stored in the root file (see last step)
|
| 62 |
|
|
|
| 63 |
|
|
|
| 64 |
|
|
Classify_HWW.C
|
| 65 |
|
|
----------------
|
| 66 |
|
|
|
| 67 |
|
|
TMVAClassificationApplication.C explains the usage of the TMVA::Reader class, which
|
| 68 |
|
|
is to be used to evaluate your data with the trained MVA
|
| 69 |
|
|
methods. The following steps are taken:
|
| 70 |
|
|
|
| 71 |
|
|
1st: The reader is created.
|
| 72 |
|
|
|
| 73 |
|
|
2nd: A set of local variables is created and declared to the reader.
|
| 74 |
|
|
|
| 75 |
|
|
3rd: The method(s) - which means the name of the weight files
|
| 76 |
|
|
created in the training process - are booked with the reader.
|
| 77 |
|
|
|
| 78 |
|
|
4th: Your event loop:
|
| 79 |
|
|
- The values of the local variables are set (either by assignment
|
| 80 |
|
|
of, as in the example, by setting the BranchAddresses of your
|
| 81 |
|
|
datatree)
|
| 82 |
|
|
- the reader is asked to evaluate the MVA, the result of which is
|
| 83 |
|
|
used to separate signal from background
|
| 84 |
|
|
|
| 85 |
|
|
|
| 86 |
|
|
How to run the example
|
| 87 |
|
|
======================
|
| 88 |
|
|
|
| 89 |
|
|
First, TMVA_HWW.C needs to be run to perform the training. You
|
| 90 |
|
|
can edit TMVA_HWW.C (boolean flags at the head of the file) to
|
| 91 |
|
|
enable (disable) the methods you would (not) like to use. Then run
|
| 92 |
|
|
|
| 93 |
|
|
> root TMVA_HWW.C
|
| 94 |
|
|
|
| 95 |
|
|
Add the end a file TMVA.root is written, and a GUI opens. Once
|
| 96 |
|
|
TMVA.root exists, the GUI can also be started via
|
| 97 |
|
|
|
| 98 |
|
|
> root TMVAGui.C
|
| 99 |
|
|
|
| 100 |
|
|
You can also run
|
| 101 |
|
|
|
| 102 |
|
|
> root Classify_HWW.C
|
| 103 |
|
|
|
| 104 |
|
|
which evaluates the methods on the signal data from the toy
|
| 105 |
|
|
tmva_example.root and writes the result to TMVApp.root. However,
|
| 106 |
|
|
TMVAClassificationApplication is more of a pedagogical example for
|
| 107 |
|
|
the usage of TMVA inside your own analysis framework.
|
| 108 |
|
|
|
| 109 |
|
|
The TMVA GUI
|
| 110 |
|
|
============
|
| 111 |
|
|
|
| 112 |
|
|
The GUI provides easy access to a large number of macros that plot
|
| 113 |
|
|
various distributions and correlations of the input data, the MVA
|
| 114 |
|
|
method output distributions and the performance comparison plot
|
| 115 |
|
|
(background rejection versus efficiency). It can be used once the
|
| 116 |
|
|
training is run, and is started with
|
| 117 |
|
|
|
| 118 |
|
|
> root TMVAGui.C
|
| 119 |
|
|
|
| 120 |
|
|
or
|
| 121 |
|
|
|
| 122 |
|
|
> root TMVAGui.C\(\"MyTMVA.root\"\)
|
| 123 |
|
|
|
| 124 |
|
|
in case you had directed your training output to a different root
|
| 125 |
|
|
file (2nd argument of the TMVA::Factory constructor)
|
| 126 |
|
|
|
