source: orange/orange/doc/datasets/monks-3_test.htm @ 1760:9d4bb141fb0e

Revision 1760:9d4bb141fb0e, 2.9 KB checked in by blaz <blaz.zupan@…>, 9 years ago (diff)

data info file

Line 
1<html>
2<head>
3<title>Monks Data Bases</title>
4</head>
5<body>
6<h1>Info on Monks Data Bases</h1>
7<pre>
81. Title: The Monk's Problems
9
102. Sources:
11    (a) Donor: Sebastian Thrun
12           School of Computer Science
13           Carnegie Mellon University
14           Pittsburgh, PA 15213, USA
15
16           E-mail: thrun@cs.cmu.edu
17
18    (b) Date: October 1992
19
203. Past Usage:
21
22   - See File: thrun.comparison.ps.Z
23
24   - Wnek, J., "Hypothesis-driven Constructive Induction," PhD dissertation,
25     School of Information Technology and Engineering, Reports of Machine
26     Learning and Inference Laboratory, MLI 93-2, Center for Artificial
27     Intelligence, George Mason University, March 1993.
28
29   - Wnek, J. and Michalski, R.S., "Comparing Symbolic and
30     Subsymbolic Learning: Three Studies," in Machine Learning: A
31     Multistrategy Approach, Vol. 4., R.S. Michalski and G. Tecuci (Eds.),
32     Morgan Kaufmann, San Mateo, CA, 1993.
33
344. Relevant Information:
35
36   The MONK's problem were the basis of a first international comparison
37   of learning algorithms. The result of this comparison is summarized in
38   "The MONK's Problems - A Performance Comparison of Different Learning
39   algorithms" by S.B. Thrun, J. Bala, E. Bloedorn, I.  Bratko, B.
40   Cestnik, J. Cheng, K. De Jong, S.  Dzeroski, S.E. Fahlman, D. Fisher,
41   R. Hamann, K. Kaufman, S. Keller, I. Kononenko, J.  Kreuziger, R.S.
42   Michalski, T. Mitchell, P.  Pachowicz, Y. Reich H.  Vafaie, W. Van de
43   Welde, W. Wenzel, J. Wnek, and J. Zhang has been published as
44   Technical Report CS-CMU-91-197, Carnegie Mellon University in Dec.
45   1991.
46
47   One significant characteristic of this comparison is that it was
48   performed by a collection of researchers, each of whom was an advocate
49   of the technique they tested (often they were the creators of the
50   various methods). In this sense, the results are less biased than in
51   comparisons performed by a single person advocating a specific
52   learning method, and more accurately reflect the generalization
53   behavior of the learning techniques as applied by knowledgeable users.
54
55   There are three MONK's problems.  The domains for all MONK's problems
56   are the same (described below).  One of the MONK's problems has noise
57   added. For each problem, the domain has been partitioned into a train
58   and test set.
59
605. Number of Instances: 432
61
626. Number of Attributes: 8 (including class attribute)
63
647. Attribute information:
65    1. class: 0, 1
66    2. a1:    1, 2, 3
67    3. a2:    1, 2, 3
68    4. a3:    1, 2
69    5. a4:    1, 2, 3
70    6. a5:    1, 2, 3, 4
71    7. a6:    1, 2
72    8. Id:    (A unique symbol for each instance)
73
748. Missing Attribute Values: None
75
769. Target Concepts associated to the MONK's problem:
77
78   MONK-1: (a1 = a2) or (a5 = 1)
79
80   MONK-2: EXACTLY TWO of {a1 = 1, a2 = 1, a3 = 1, a4 = 1, a5 = 1, a6 = 1}
81
82   MONK-3: (a5 = 3 and a4 = 1) or (a5 /= 4 and a2 /= 3)
83           (5% class noise added to the training set)
84
85</pre>
86</body>
87</html>
Note: See TracBrowser for help on using the repository browser.