UZH

Physik-Institut

CMS Wiki Pages

User Tools

Site Tools

Trace:
mva:mva

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revisionPrevious revision
Next revision		Previous revision
mva:mva [2016/08/06 12:34] – Application iwnmva:mva [2023/06/01 13:29] (current) – [Other information] iwn
Line 8: Line 8:
   * [[https://root.cern.ch/doc/v606/classTMVA_1_1Factory.html|Factory Class Reference]]   * [[https://root.cern.ch/doc/v606/classTMVA_1_1Factory.html|Factory Class Reference]]
   * [[https://root.cern.ch/doc/v606/classTMVA_1_1Reader.html|Reader Class Reference]]   * [[https://root.cern.ch/doc/v606/classTMVA_1_1Reader.html|Reader Class Reference]]
 +  * [[https://root.cern.ch/doc/master/group__tutorial__tmva.html|Official examples (C++)]]
  
 When using the neural netwerk method MLP, you might need ROOT 5.34.0.0 or newer, to have larger buffer for the xml reader, for example: When using the neural netwerk method MLP, you might need ROOT 5.34.0.0 or newer, to have larger buffer for the xml reader, for example:
-<code>+<code bash>
 . /afs/cern.ch/sw/lcg/app/releases/ROOT/5.34.26/x86_64-slc6-gcc48-opt/root/bin/thisroot.sh . /afs/cern.ch/sw/lcg/app/releases/ROOT/5.34.26/x86_64-slc6-gcc48-opt/root/bin/thisroot.sh
 </code> </code>
- 
  
  
Line 20: Line 20:
 To run the MVA in python on a signal tree ''treeS'' and background tree ''treeB'' with a list of variables, ''varNames'', that are available in the tree, you first need: To run the MVA in python on a signal tree ''treeS'' and background tree ''treeB'' with a list of variables, ''varNames'', that are available in the tree, you first need:
  
-<code>+<code python>
 from ROOT import TFile, TTree, TMVA, TCut from ROOT import TFile, TTree, TMVA, TCut
 f_out = TFile("MVA.root","RECREATE") f_out = TFile("MVA.root","RECREATE")
Line 34: Line 34:
 </code> </code>
  
-In the empty quote marks you can add cuts or options. More information on can be found on the [[https://root.cern.ch/doc/v606/classTMVA_1_1Factory.html|Factory Class Reference]].+In the empty quote marks you can add cuts or options. More information on can be found in section 3.1 in the [[http://tmva.sourceforge.net/docu/TMVAUsersGuide.pdf|manual]] and in the [[https://root.cern.ch/doc/v606/classTMVA_1_1Factory.html|Factory Class Reference]].
  
 Then you can book multiple methods like the BDT and MLP for different parameters: Then you can book multiple methods like the BDT and MLP for different parameters:
  
-<code>+<code python>
 factory.BookMethod( TMVA.Types.kBDT, "BDT", "!H:!V" ) factory.BookMethod( TMVA.Types.kBDT, "BDT", "!H:!V" )
 factory.BookMethod( TMVA.Types.kBDT, "BDTTuned", factory.BookMethod( TMVA.Types.kBDT, "BDTTuned",
Line 50: Line 50:
 Finally train, test and evaluate all the booked methods: Finally train, test and evaluate all the booked methods:
  
-<code>+<code python>
 factory.TrainAllMethods() factory.TrainAllMethods()
 factory.TestAllMethods() factory.TestAllMethods()
Line 56: Line 56:
 f_out.Close() f_out.Close()
 </code> </code>
- 
  
  
Line 63: Line 62:
 The factory will output weights in a XML file you can use to apply to a tree that contains the same variable. The factory will output weights in a XML file you can use to apply to a tree that contains the same variable.
  
-<code>+<code python>
 reader = TMVA.Reader() reader = TMVA.Reader()
 vars = [ ] vars = [ ]
Line 69: Line 68:
     vars.append(array('f',[0]))     vars.append(array('f',[0]))
     reader.AddVariable(name,vars[-1])     reader.AddVariable(name,vars[-1])
-reader.BookMVA(TMVAClassification.weights.xml")+reader.BookMVA("TMVAClassification.weights.xml")
     for i in range(len(config.varNames)):     for i in range(len(config.varNames)):
         tree.SetBranchAddress(config.varNames[i],vars[i])         tree.SetBranchAddress(config.varNames[i],vars[i])
Line 81: Line 80:
  ===== Parameters to tune =====  ===== Parameters to tune =====
  
-The parameters and options of the MVA method can be optimized from the default settings for better a performance, check out the [[http://tmva.sourceforge.net/optionRef.html|reference page]].+The parameters and options of the MVA method can be optimized from the default settings for better a performance, check out the [[http://tmva.sourceforge.net/optionRef.html|option reference page]].
  
 In particular, for the **BDT**, important parameters are the the learning rate, number of boost steps and maximal tree depth: In particular, for the **BDT**, important parameters are the the learning rate, number of boost steps and maximal tree depth:
  
-  * ''AdaBoostBeta=0.5'': [[https://en.wikipedia.org/wiki/AdaBoost|learning rate]], smaller (~0.1) is better, but takes longer+  * ''AdaBoostBeta=0.5'': [[https://en.wikipedia.org/wiki/AdaBoost|learning rate for AdaBoost]], smaller (~0.1) is better, but takes longer
   * ''nTrees=800'': number of boost steps, too large mainly costs time and can cause overtraining   * ''nTrees=800'': number of boost steps, too large mainly costs time and can cause overtraining
   * ''MaxDepth=3'': maximum tree depth, ~2-5 depending on interaction of the variables   * ''MaxDepth=3'': maximum tree depth, ~2-5 depending on interaction of the variables
   * ''nCuts=20'': grid points in variable range to find the optimal cut in node splitting   * ''nCuts=20'': grid points in variable range to find the optimal cut in node splitting
-  * ''MinNodeSize=5%''+  * ''SeparationType=GiniIndex'': separating criterion at each splitting node to select best variable. The [[https://en.wikipedia.org/wiki/Gini_coefficient|Gini index]] is one often used measure 
 +  * ''MinNodeSize=5%'': minimum percentage of training events required in a leaf node
  
 Important **MLP**  parameters to tune are the number of neurons on each hidden layer, learning rate and the activation function. Important **MLP**  parameters to tune are the number of neurons on each hidden layer, learning rate and the activation function.
Line 98: Line 98:
      * ''N+2,N'' = two hidden layers, with N+2 nodes in the first      * ''N+2,N'' = two hidden layers, with N+2 nodes in the first
   * ''LearningRate=0.02''   * ''LearningRate=0.02''
 +  * ''NeuronType=sigmoid'': other neuron activation function is ''tanh''
  
  
Line 106: Line 106:
   * [[https://aholzner.wordpress.com/2011/08/27/a-tmva-example-in-pyroot/|TMVA in PyRoot tutorial]]   * [[https://aholzner.wordpress.com/2011/08/27/a-tmva-example-in-pyroot/|TMVA in PyRoot tutorial]]
   * [[mva:mvaexample|This python example]] shows how you can run over multiple trees and different sets of variables. It also includes functions to apply the MVA output to trees and make background rejection vs. signal efficiency plots and correlation plots.   * [[mva:mvaexample|This python example]] shows how you can run over multiple trees and different sets of variables. It also includes functions to apply the MVA output to trees and make background rejection vs. signal efficiency plots and correlation plots.
 +  * [[https://root.cern/doc/master/group__tutorial__tmva.html|ROOT official TMVA tutorials]]
 +
 +
 + ===== Other information =====
 +
 +  * The working principles of a [[https://www.physik.uzh.ch/~grazzini/teaching/higgsnotes/lecture6.pdf|neural network]] and a [[https://www.physik.uzh.ch/~grazzini/teaching/higgsnotes/lecture10.pdf|BDT]] are visually explained in the UZH's [[https://www.physik.uzh.ch/~grazzini/teaching/higgs.html|Higgs Physics course]] by Mauro Donegà.
 +  * [[https://arogozhnikov.github.io/2016/07/05/gradient_boosting_playground.html|Gradient Boosting Interactive Playground]] with interactive visuals
 +  * TMVA BibTex reference:
 +<code latex>
 +@article{TMVA, 
 +  title         = {TMVA: Toolkit for Multivariate Data Analysis},
 +  author        = {Hoecker, Andreas and Speckmayer, Peter and
 +                   Stelzer, Joerg and Therhaag, Jan and
 +                   von Toerne, Eckhard and Voss, Helge},
 +  journal       = {PoS},
 +  volume        = {ACAT},
 +  year          = {2007},
 +  month         = {Mar},
 +  pages         = {040},
 +  url           = {http://inspirehep.net/record/746087/},
 +  reportNumber  = {CERN-OPEN-2007-007},
 +  eprint        = {physics/0703039},
 +  archivePrefix = {arXiv},
 +  primaryClass  = {physics},
 +  SLACcitation  = {%%CITATION = arXiv:physics/0703039;%%},
 +}
 +</code>
mva/mva.1470479653.txt.gz · Last modified: 2016/08/06 12:34 by iwn

Page Tools

Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 4.0 International
CC Attribution-Share Alike 4.0 International Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki