Download data visualization


Machine Learning with
WEKA

WEKA: A Machine
Learning Toolkit
The Explorer
•
•
Eibe Frank
•
•
Department of Computer Science,
University of Waikato, New Zealand
•



Classification and
Regression
Clustering
Association Rules
Attribute Selection
Data Visualization
The Experimenter
The Knowledge
Flow GUI
Conclusions
WEKA: the bird
Copyright: Martin Kramer ([email protected])
5/2/2017
University of Waikato
2
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 algorithms

5/2/2017
University of Waikato
3
WEKA: versions

There are several versions of WEKA:
WEKA 3.0: “book version” compatible with
description in data mining book
 WEKA 3.2: “GUI version” adds graphical user
interfaces (book version is command-line only)
 WEKA 3.3: “development version” with lots of
improvements


This talk is based on the latest snapshot of WEKA
3.3 (soon to be WEKA 3.4)
5/2/2017
University of Waikato
4
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}
@data
63,male,typ_angina,233,no,not_present
67,male,asympt,286,yes,present
67,male,asympt,229,yes,present
38,female,non_anginal,?,no,not_present
...
5/2/2017
University of Waikato
5
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}
@data
63,male,typ_angina,233,no,not_present
67,male,asympt,286,yes,present
67,male,asympt,229,yes,present
38,female,non_anginal,?,no,not_present
...
5/2/2017
University of Waikato
6
5/2/2017
University of Waikato
7
5/2/2017
University of Waikato
8
5/2/2017
University of Waikato
9
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:

5/2/2017
Discretization, normalization, resampling, attribute
selection, transforming and combining attributes, …
University of Waikato
10
5/2/2017
University of Waikato
11
5/2/2017
University of Waikato
12
5/2/2017
University of Waikato
13
5/2/2017
University of Waikato
14
5/2/2017
University of Waikato
15
5/2/2017
University of Waikato
16
5/2/2017
University of Waikato
17
5/2/2017
University of Waikato
18
5/2/2017
University of Waikato
19
5/2/2017
University of Waikato
20
5/2/2017
University of Waikato
21
5/2/2017
University of Waikato
22
5/2/2017
University of Waikato
23
5/2/2017
University of Waikato
24
5/2/2017
University of Waikato
25
5/2/2017
University of Waikato
26
5/2/2017
University of Waikato
27
5/2/2017
University of Waikato
28
5/2/2017
University of Waikato
29
5/2/2017
University of Waikato
30
5/2/2017
University of Waikato
31
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:

5/2/2017
Bagging, boosting, stacking, error-correcting output
codes, locally weighted learning, …
University of Waikato
32
5/2/2017
University of Waikato
33
5/2/2017
University of Waikato
34
5/2/2017
University of Waikato
35
5/2/2017
University of Waikato
36
5/2/2017
University of Waikato
37
5/2/2017
University of Waikato
38
5/2/2017
University of Waikato
39
5/2/2017
University of Waikato
40
5/2/2017
University of Waikato
41
5/2/2017
University of Waikato
42
5/2/2017
University of Waikato
43
5/2/2017
University of Waikato
44
5/2/2017
University of Waikato
45
5/2/2017
University of Waikato
46
5/2/2017
University of Waikato
47
5/2/2017
University of Waikato
48
5/2/2017
University of Waikato
49
5/2/2017
University of Waikato
50
5/2/2017
University of Waikato
51
5/2/2017
University of Waikato
52
5/2/2017
University of Waikato
53
5/2/2017
University of Waikato
54
5/2/2017
University of Waikato
55
5/2/2017
University of Waikato
56
5/2/2017
University of Waikato
57
5/2/2017
University of Waikato
58
5/2/2017
University of Waikato
59
5/2/2017
University of Waikato
60
5/2/2017
University of Waikato
61
5/2/2017
University of Waikato
62
5/2/2017
University of Waikato
63
5/2/2017
University of Waikato
64
5/2/2017
QuickTime™ and a TI FF (LZW) decompressor are needed to see this picture.
University of Waikato
65
5/2/2017
QuickTime™ and a TI FF (LZW) decompressor are needed to see this picture.
University of Waikato
66
5/2/2017
QuickTime™ and a TI FF (LZW) decompressor are needed to see this picture.
University of Waikato
67
5/2/2017
University of Waikato
68
5/2/2017
University of Waikato
69
5/2/2017
University of Waikato
70
5/2/2017
University of Waikato
71
5/2/2017
University of Waikato
72
5/2/2017
University of Waikato
73
5/2/2017
University of Waikato
74
Qu i ck Ti me ™a nd a TIFF (LZW)d ec omp res so ra re ne ed ed to s ee th i s pi c tu re.
5/2/2017
University of Waikato
75
5/2/2017
University of Waikato
76
5/2/2017
University of Waikato
77
5/2/2017
University of Waikato
78
5/2/2017
University of Waikato
79
QuickTime™ and a TI FF (LZW) decompressor are needed to see this picture.
5/2/2017
University of Waikato
80
QuickTime™ and a TI FF (LZW) decompressor are needed to see this picture.
5/2/2017
University of Waikato
81
5/2/2017
University of Waikato
82
QuickTime™ and a TI FF (LZW) decompressor are needed to see this picture.
5/2/2017
University of Waikato
83
5/2/2017
University of Waikato
84
5/2/2017
University of Waikato
85
5/2/2017
University of Waikato
86
5/2/2017
University of Waikato
87
5/2/2017
University of Waikato
88
5/2/2017
University of Waikato
89
5/2/2017
University of Waikato
90
5/2/2017
University of Waikato
91
Explorer: clustering data


WEKA contains “clusterers” for finding groups of
similar instances in a dataset
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 distribution
5/2/2017
University of Waikato
92
5/2/2017
University of Waikato
93
5/2/2017
University of Waikato
94
5/2/2017
University of Waikato
95
5/2/2017
University of Waikato
96
5/2/2017
University of Waikato
97
5/2/2017
University of Waikato
98
5/2/2017
University of Waikato
99
5/2/2017
University of Waikato
100
5/2/2017
University of Waikato
101
5/2/2017
University of Waikato
102
5/2/2017
University of Waikato
103
5/2/2017
University of Waikato
104
5/2/2017
University of Waikato
105
5/2/2017
University of Waikato
106
5/2/2017
University of Waikato
107
Explorer: finding associations

WEKA contains an implementation of the Apriori
algorithm for learning association rules


Can identify statistical dependencies between
groups of attributes:


Works only with discrete data
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 confidence
5/2/2017
University of Waikato
108
5/2/2017
University of Waikato
109
5/2/2017
University of Waikato
110
5/2/2017
University of Waikato
111
5/2/2017
University of Waikato
112
5/2/2017
University of Waikato
113
5/2/2017
University of Waikato
114
5/2/2017
University of Waikato
115
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 two
5/2/2017
University of Waikato
116
5/2/2017
University of Waikato
117
5/2/2017
University of Waikato
118
5/2/2017
University of Waikato
119
5/2/2017
University of Waikato
120
5/2/2017
University of Waikato
121
5/2/2017
University of Waikato
122
5/2/2017
University of Waikato
123
5/2/2017
University of Waikato
124
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” function
5/2/2017
University of Waikato
125
5/2/2017
University of Waikato
126
5/2/2017
University of Waikato
127
5/2/2017
University of Waikato
128
5/2/2017
University of Waikato
129
5/2/2017
University of Waikato
130
5/2/2017
University of Waikato
131
5/2/2017
University of Waikato
132
5/2/2017
University of Waikato
133
5/2/2017
University of Waikato
134
5/2/2017
University of Waikato
135
5/2/2017
University of Waikato
136
5/2/2017
University of Waikato
137
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!
5/2/2017
University of Waikato
138
5/2/2017
University of Waikato
139
5/2/2017
University of Waikato
140
5/2/2017
University of Waikato
141
5/2/2017
University of Waikato
142
5/2/2017
University of Waikato
143
5/2/2017
University of Waikato
144
5/2/2017
University of Waikato
145
5/2/2017
University of Waikato
146
5/2/2017
University of Waikato
147
5/2/2017
University of Waikato
148
5/2/2017
University of Waikato
149
5/2/2017
University of Waikato
150
5/2/2017
University of Waikato
151
The Knowledge Flow GUI





New graphical user interface for WEKA
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
5/2/2017
University of Waikato
152
5/2/2017
University of Waikato
153
5/2/2017
University of Waikato
154
5/2/2017
University of Waikato
155
5/2/2017
University of Waikato
156
5/2/2017
University of Waikato
157
5/2/2017
University of Waikato
158
5/2/2017
University of Waikato
159
5/2/2017
University of Waikato
160
5/2/2017
University of Waikato
161
5/2/2017
University of Waikato
162
5/2/2017
University of Waikato
163
5/2/2017
University of Waikato
164
5/2/2017
University of Waikato
165
5/2/2017
University of Waikato
166
5/2/2017
University of Waikato
167
5/2/2017
University of Waikato
168
5/2/2017
University of Waikato
169
5/2/2017
University of Waikato
170
5/2/2017
University of Waikato
171
5/2/2017
University of Waikato
172
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
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 Wang
5/2/2017
University of Waikato
173