Weka a tool_for_exploratory_data_mining

2,942 views

Published on

Published in: Technology
0 Comments
6 Likes
Statistics
Notes
  • Be the first to comment

No Downloads
Views
Total views
2,942
On SlideShare
0
From Embeds
0
Number of Embeds
5
Actions
Shares
0
Downloads
230
Comments
0
Likes
6
Embeds 0
No embeds

No notes for slide

Weka a tool_for_exploratory_data_mining

  1. 1.  WEKA: A MachineMachine Learning with Learning Toolkit WEKA  The Explorer • Classification and Regression • Clustering Bernhard Pfahringer • Association Rules(based on material by Eibe Frank, Mark • Attribute Selection Hall, and Peter Reutemann) • Data Visualization Department of Computer Science,  The Experimenter University of Waikato, New Zealand  The Knowledge Flow GUI  Other Utilities  Conclusions
  2. 2. WEKA: the bird The Weka or woodhen (Gallirallus australis) is an endemic bird of New Zealand. (Source: WikiPedia) Copyright: Martin Kramer (mkramer@wxs.nl)12/05/12 University of Waikato 2
  3. 3. WEKA: the software Machine learning/data mining software written in Java (distributed under the GNU Public License) Used for research, education, and applications Complements “Data Mining” by Witten & Frank Main features:  Comprehensive set of data pre-processing tools, learning algorithms and evaluation methods  Graphical user interfaces (incl. data visualization)  Environment for comparing learning algorithms12/05/12 University of Waikato 3
  4. 4. History Project funded by the NZ government since 1993  Develop state-of-the art workbench of data mining tools  Explore fielded applications  Develop new fundamental methods12/05/12 University of Waikato 4
  5. 5. History (2) Late 1992 - funding was applied for by Ian Witten 1993 - development of the interface and infrastructure  WEKA acronym coined by Geoff Holmes  WEKA’s file format “ARFF” was created by Andrew Donkin ARFF was rumored to stand for Andrew’s Ridiculous File Format Sometime in 1994 - first internal release of WEKA  TCL/TK user interface + learning algorithms written mostly in C  Very much beta software  Changes for the b1 release included (among others): “Ambiguous and Unsupported menu commands removed.” “Crashing processes handled (in most cases :-)” October 1996 - first public release: WEKA 2.112/05/12 University of Waikato 5
  6. 6. History (3) July 1997 - WEKA 2.2  Schemes: 1R, T2, K*, M5, M5Class, IB1-4, FOIL, PEBLS, support for C5  Included a facility (based on Unix makefiles) for configuring and running large scale experiments Early 1997 - decision was made to rewrite WEKA in Java  Originated from code written by Eibe Frank for his PhD  Originally codenamed JAWS (JAva Weka System) JA May 1998 - WEKA 2.3  Last release of the TCL/TK-based system Mid 1999 - WEKA 3 (100% Java) released  Version to complement the Data Mining book  Development version (including GUI)12/05/12 University of Waikato 6
  7. 7. The GUI back then… TCL/TK interface of Weka 2.112/05/12 University of Waikato 7
  8. 8. WEKA: versions There are several versions of WEKA:  WEKA 3.4: “book version” compatible with description in data mining book  WEKA 3.5.5: “development version” with lots of improvements This talk is based on a nightly snapshot of WEKA 3.5.5 (12-Feb-2007)12/05/12 University of Waikato 8
  9. 9. WEKA only deals with “flat” files@relation heart-disease-simplified@attribute age numeric@attribute sex { female, male}@attribute chest_pain_type { typ_angina, asympt, non_anginal, atyp_angina}@attribute cholesterol numeric@attribute exercise_induced_angina { no, yes}@attribute class { present, not_present}@data63,male,typ_angina,233,no,not_present67,male,asympt,286,yes,present67,male,asympt,229,yes,present38,female,non_anginal,?,no,not_present...12/05/12 University of Waikato 9
  10. 10. WEKA only deals with “flat” files@relation heart-disease-simplified@attribute age numeric@attribute sex { female, male}@attribute chest_pain_type { typ_angina, asympt, non_anginal, atyp_angina}@attribute cholesterol numeric@attribute exercise_induced_angina { no, yes}@attribute class { present, not_present}@data63,male,typ_angina,233,no,not_present67,male,asympt,286,yes,present67,male,asympt,229,yes,present38,female,non_anginal,?,no,not_present...12/05/12 University of Waikato 10
  11. 11. java weka.gui.GUIChooser12/05/12 University of Waikato 11
  12. 12. 12/05/12 University of Waikato 12
  13. 13. 12/05/12 University of Waikato 13
  14. 14. java -jar weka.jar12/05/12 University of Waikato 14
  15. 15. Explorer: pre-processing the data Data can be imported from a file in various formats: ARFF, CSV, C4.5, binary Data can also be read from a URL or from an SQL database (using JDBC) Pre-processing tools in WEKA are called “filters” WEKA contains filters for:  Discretization, normalization, resampling, attribute selection, transforming and combining attributes, …12/05/12 University of Waikato 15
  16. 16. 12/05/12 University of Waikato 16
  17. 17. 12/05/12 University of Waikato 17
  18. 18. 12/05/12 University of Waikato 18
  19. 19. 12/05/12 University of Waikato 19
  20. 20. 12/05/12 University of Waikato 20
  21. 21. 12/05/12 University of Waikato 21
  22. 22. 12/05/12 University of Waikato 22
  23. 23. 12/05/12 University of Waikato 23
  24. 24. 12/05/12 University of Waikato 24
  25. 25. 12/05/12 University of Waikato 25
  26. 26. 12/05/12 University of Waikato 26
  27. 27. 12/05/12 University of Waikato 27
  28. 28. 12/05/12 University of Waikato 28
  29. 29. 12/05/12 University of Waikato 29
  30. 30. 12/05/12 University of Waikato 30
  31. 31. 12/05/12 University of Waikato 31
  32. 32. 12/05/12 University of Waikato 32
  33. 33. 12/05/12 University of Waikato 33
  34. 34. 12/05/12 University of Waikato 34
  35. 35. 12/05/12 University of Waikato 35
  36. 36. 12/05/12 University of Waikato 36
  37. 37. Explorer: building “classifiers” Classifiers in WEKA are models for predicting nominal or numeric quantities Implemented learning schemes include:  Decision trees and lists, instance-based classifiers, support vector machines, multi-layer perceptrons, logistic regression, Bayes’ nets, … “Meta”-classifiers include:  Bagging, boosting, stacking, error-correcting output codes, locally weighted learning, …12/05/12 University of Waikato 37
  38. 38. 12/05/12 University of Waikato 38
  39. 39. 12/05/12 University of Waikato 39
  40. 40. 12/05/12 University of Waikato 40
  41. 41. 12/05/12 University of Waikato 41
  42. 42. 12/05/12 University of Waikato 42
  43. 43. 12/05/12 University of Waikato 43
  44. 44. 12/05/12 University of Waikato 44
  45. 45. 12/05/12 University of Waikato 45
  46. 46. 12/05/12 University of Waikato 46
  47. 47. 12/05/12 University of Waikato 47
  48. 48. 12/05/12 University of Waikato 48
  49. 49. 12/05/12 University of Waikato 49
  50. 50. 12/05/12 University of Waikato 50
  51. 51. 12/05/12 University of Waikato 51
  52. 52. 12/05/12 University of Waikato 52
  53. 53. 12/05/12 University of Waikato 53
  54. 54. 12/05/12 University of Waikato 54
  55. 55. 12/05/12 University of Waikato 55
  56. 56. 12/05/12 University of Waikato 56
  57. 57. 12/05/12 University of Waikato 57
  58. 58. 12/05/12 University of Waikato 58
  59. 59. 12/05/12 University of Waikato 59
  60. 60. 12/05/12 University of Waikato 60
  61. 61. 12/05/12 University of Waikato 61
  62. 62. 12/05/12 University of Waikato 62
  63. 63. 12/05/12 University of Waikato 63
  64. 64. 12/05/12 University of Waikato 64
  65. 65. 12/05/12 University of Waikato 65
  66. 66. 12/05/12 University of Waikato 66
  67. 67. 12/05/12 University of Waikato 67
  68. 68. 12/05/12 University of Waikato 68
  69. 69. 12/05/12 University of Waikato 69
  70. 70. 12/05/12 University of Waikato 70
  71. 71. 12/05/12 University of Waikato 71
  72. 72. 12/05/12QuickTime™ and a T IFF (LZW) decompressor are needed to see this picture. University of Waikato 72
  73. 73. 12/05/12QuickTime™ and a T IFF (LZW) decompressor are needed to see this picture. University of Waikato 73
  74. 74. 12/05/12 University of Waikato 74
  75. 75. 12/05/12 University of Waikato 75
  76. 76. 12/05/12 University of Waikato 76
  77. 77. 12/05/12 University of Waikato 77
  78. 78. Qui c k Ti m e ™ a nd a TIFF (LZW) d ec om pre s s o r a re n e ed e d to s e e th i s p i c tu re .12/05/12 University of Waikato 78
  79. 79. 12/05/12 University of Waikato 79
  80. 80. 12/05/12 University of Waikato 80
  81. 81. 12/05/12 University of Waikato 81
  82. 82. 12/05/12 University of Waikato 82
  83. 83. 12/05/12 University of Waikato 83
  84. 84. 12/05/12 University of Waikato 84
  85. 85. 12/05/12 University of Waikato 85
  86. 86. 12/05/12 University of Waikato 86
  87. 87. 12/05/12 University of Waikato 87
  88. 88. 12/05/12 University of Waikato 88
  89. 89. 12/05/12 University of Waikato 89
  90. 90. 12/05/12 University of Waikato 90
  91. 91. 12/05/12 University of Waikato 91
  92. 92. 12/05/12 University of Waikato 92
  93. 93. Explorer: clustering data WEKA contains “clusterers” for finding groups of similar instances in a dataset Some implemented schemes are:  k-Means, EM, Cobweb, X-means, FarthestFirst Clusters can be visualized and compared to “true” clusters (if given) Evaluation based on loglikelihood if clustering scheme produces a probability distribution12/05/12 University of Waikato 93
  94. 94. 12/05/12 University of Waikato 94
  95. 95. 12/05/12 University of Waikato 95
  96. 96. 12/05/12 University of Waikato 96
  97. 97. 12/05/12 University of Waikato 97
  98. 98. 12/05/12 University of Waikato 98
  99. 99. 12/05/12 University of Waikato 99
  100. 100. 12/05/12 University of Waikato 100
  101. 101. 12/05/12 University of Waikato 101
  102. 102. 12/05/12 University of Waikato 102
  103. 103. Explorer: finding associations WEKA contains the Apriori algorithm (among others) for learning association rules  Works only with discrete data Can identify statistical dependencies between groups of attributes:  milk, butter ⇒ bread, eggs (with confidence 0.9 and support 2000) Apriori can compute all rules that have a given minimum support and exceed a given confidence12/05/12 University of Waikato 103
  104. 104. 12/05/12 University of Waikato 104
  105. 105. 12/05/12 University of Waikato 105
  106. 106. 12/05/12 University of Waikato 106
  107. 107. Explorer: attribute selection Panel that can be used to investigate which (subsets of) attributes are the most predictive ones Attribute selection methods contain two parts:  A search method: best-first, forward selection, random, exhaustive, genetic algorithm, ranking  An evaluation method: correlation-based, wrapper, information gain, chi-squared, … Very flexible: WEKA allows (almost) arbitrary combinations of these two12/05/12 University of Waikato 107
  108. 108. 12/05/12 University of Waikato 108
  109. 109. 12/05/12 University of Waikato 109
  110. 110. 12/05/12 University of Waikato 110
  111. 111. 12/05/12 University of Waikato 111
  112. 112. 12/05/12 University of Waikato 112
  113. 113. 12/05/12 University of Waikato 113
  114. 114. Explorer: data visualization Visualization very useful in practice: e.g. helps to determine difficulty of the learning problem WEKA can visualize single attributes (1-d) and pairs of attributes (2-d)  To do: rotating 3-d visualizations (Xgobi-style) Color-coded class values “Jitter” option to deal with nominal attributes (and to detect “hidden” data points) “Zoom-in” function12/05/12 University of Waikato 114
  115. 115. 12/05/12 University of Waikato 115
  116. 116. 12/05/12 University of Waikato 116
  117. 117. 12/05/12 University of Waikato 117
  118. 118. 12/05/12 University of Waikato 118
  119. 119. 12/05/12 University of Waikato 119
  120. 120. 12/05/12 University of Waikato 120
  121. 121. 12/05/12 University of Waikato 121
  122. 122. 12/05/12 University of Waikato 122
  123. 123. 12/05/12 University of Waikato 123
  124. 124. 12/05/12 University of Waikato 124
  125. 125. Performing experiments Experimenter makes it easy to compare the performance of different learning schemes For classification and regression problems Results can be written into file or database Evaluation options: cross-validation, learning curve, hold-out Can also iterate over different parameter settings Significance-testing built in!12/05/12 University of Waikato 125
  126. 126. 12/05/12 University of Waikato 126
  127. 127. 12/05/12 University of Waikato 127
  128. 128. 12/05/12 University of Waikato 128
  129. 129. 12/05/12 University of Waikato 129
  130. 130. 12/05/12 University of Waikato 130
  131. 131. 12/05/12 University of Waikato 131
  132. 132. 12/05/12 University of Waikato 132
  133. 133. 12/05/12 University of Waikato 133
  134. 134. 12/05/12 University of Waikato 134
  135. 135. 12/05/12 University of Waikato 135
  136. 136. The Knowledge Flow GUI Java-Beans-based interface for setting up and running machine learning experiments Data sources, classifiers, etc. are beans and can be connected graphically Data “flows” through components: e.g., “data source” -> “filter” -> “classifier” -> “evaluator” Layouts can be saved and loaded again later cf. Clementine ™12/05/12 University of Waikato 136
  137. 137. 12/05/12 University of Waikato 137
  138. 138. 12/05/12 University of Waikato 138
  139. 139. 12/05/12 University of Waikato 139
  140. 140. 12/05/12 University of Waikato 140
  141. 141. 12/05/12 University of Waikato 141
  142. 142. 12/05/12 University of Waikato 142
  143. 143. 12/05/12 University of Waikato 143
  144. 144. 12/05/12 University of Waikato 144
  145. 145. 12/05/12 University of Waikato 145
  146. 146. 12/05/12 University of Waikato 146
  147. 147. 12/05/12 University of Waikato 147
  148. 148. 12/05/12 University of Waikato 148
  149. 149. 12/05/12 University of Waikato 149
  150. 150. 12/05/12 University of Waikato 150
  151. 151. 12/05/12 University of Waikato 151
  152. 152. Sourceforge.net – Downloads12/05/12 University of Waikato 152
  153. 153. Sourceforge.net – Web Traffic WekaDoc Wiki introduced – 12/2005 WekaWiki launched – 05/200512/05/12 University of Waikato 153
  154. 154. Projects based on WEKA 45 projects currently (30/01/07) listed on the WekaWiki Incorporate/wrap WEKA  GRB Tool Shed - a tool to aid gamma ray burst research  YALE - facility for large scale ML experiments  GATE - NLP workbench with a WEKA interface  Judge - document clustering and classification  RWeka - an R interface to Weka Extend/modify WEKA  BioWeka - extension library for knowledge discovery in biology  WekaMetal - meta learning extension to WEKA  Weka-Parallel - parallel processing for W EKA  Grid Weka - grid computing using WEKA  Weka-CG - computational genetics tool library12/05/12 University of Waikato 154
  155. 155. WEKA and PENTAHO Pentaho – The leader in Open Source Business Intelligence (BI) September 2006 – Pentaho acquires the Weka project (exclusive license and SF.net page) Weka will be used/integrated as data mining component in their BI suite Weka will be still available as GPL open source software Most likely to evolve 2 editions:  Community edition  BI oriented edition12/05/12 University of Waikato 155
  156. 156. Limitations of WEKA Traditional algorithms need to have all data in main memory ==> big datasets are an issue Solution:  Incremental schemes  Stream algorithms MOA “Massive Online Analysis” (not only a flightless bird, but also extinct!)12/05/12 University of Waikato 156
  157. 157. Conclusion: try it yourself! WEKA is available at http://www.cs.waikato.ac.nz/ml/weka Also has a list of projects based on WEKA (probably incomplete list of) WEKA contributors: Abdelaziz Mahoui, Alexander K. Seewald, Ashraf M. Kibriya, Bernhard Pfahringer, Brent Martin, Peter Flach, Eibe Frank, Gabi Schmidberger, Ian H. Witten, J. Lindgren, Janice Boughton, Jason Wells, Len Trigg, Lucio de Souza Coelho, Malcolm Ware, Mark Hall, Remco Bouckaert, Richard Kirkby, Shane Butler, Shane Legg, Stuart Inglis, Sylvain Roy, Tony Voyle, Xin Xu, Yong Wang, Zhihai Wang12/05/12 University of Waikato 157

×