Source

orange / Orange / datasets / bridges.htm

<html>
<head>
<title>Pittsburgh Bridges Data Base</title>
</head>
<body>
<h1>Info on Pittsburgh Bridges Data Base</h1>
<pre>
1. Title: Pittsburgh bridges

2. Sources:
   -- Yoram Reich & Steven J. Fenves
      Department of Civil Engineering
      and
      Engineering Design Research Center
      Carnegie Mellon University
      Pittsburgh, PA 15213

      Compiled from various sources.

   -- Donor: Yoram Reich (yoram.reich@cs.cmu.edu)
   -- Date: 1 August 1990

3. Past Usage:

   -- Reich & Fenves (1989). Incremental Learning for Capturing Design 
      Expertise. Technical Report: EDRC 12-34-89, Engineering Design
      Research Center, Carnegie Mellon University, Pittsburgh, PA.
      -- Qualitative results and runs with original ordering of examples.
         using COBWEB.

   -- Reich (1989). Converging to ``Ideal'' Design Knowledge by Learning,
      Proceedings of the First International Workshop on Formal Methods in
      Engineering Design, pp: 330-349, Colorado Springs, CO, January 1990.
      -- Describes a new design method with Bridger (variant of COBWEB) using
	 this domain. (Also an EDRC report: 12-35-89)

   -- Reich (1989) Combining Nominal and Continuous Properties in an 
      Incremental Learning System for Design. Technical Report: EDRC 12-33-89.
      -- Comparison of performance of Bridger when running on both versions
	 (V1 and V2) of the database

   -- Reich (1989) Incremental Concept Formation with Mixed Property Types
      Unpublished Manuscript.
      -- Results using 10 random 10-fold cross-validation test with Bridger
	 (relative error rate):
	 Version V1 of the database: 
	 MATERIAL 18.4%, REL-L 38.7%, SPAN 42.7%, T-OR-D 14.7%, TYPE 47.6%.
	 Version V2 of the database:
	 MATERIAL 24.2%, REL-L 41.7%, SPAN 39.9%, T-OR-D 14.7%, TYPE 56.5%.

   -- Quinlan (1989) Personal communication. 
      -- Results of a 10-fold cross-validation test with C4.5, and with
         a separate decision tree for each design property obtained the 
	 following error rates on version V1 of the database:
	 MATERIAL 15%, REL-L 32%, SPAN 32%, T-OR-D 15%, TYPE 44%.

4. Number of instances: 108

5. Relevant Information:
   
   There are two versions to the database: 
      V1 contains the original examples and 
      V2 contains descriptions after discretizing numeric properties.

   There are no ``classes'' in the domain. Rather this is a DESIGN domain where
   5 properties (design description) need to be predicted based on 7 
   specification properties.

6. Number of Attributes: 13: 7 specifications, 5 design description, and 1
   identifier (not used for the classification)

7. Attribute Information:  

   The type field state whether a property is continuous/integer (c) 
						      or nominal (n).
   For properties with c,n type, the range of continuous numbers is given 
   first and the possible values of the nominal follow the semi-colon.


       name     type    possible values		comments
   ------------------------------------------------------------------------
   1.  IDENTIF	-	-			identifier of the examples
   2.  RIVER	n	A, M, O
   3.  LOCATION	n       1 to 52
   4.  ERECTED	c,n	1818-1986 ; CRAFTS, EMERGING, MATURE, MODERN
   5.  PURPOSE	n	WALK, AQUEDUCT, RR, HIGHWAY
   6.  LENGTH	c,n	804-4558 ; SHORT, MEDIUM, LONG
   7.  LANES	c,n	1, 2, 4, 6 ; 1, 2, 4, 6
   8.  CLEAR-G	n	N, G
   9.  T-OR-D	n	THROUGH, DECK
   10. MATERIAL	n	WOOD, IRON, STEEL
   11. SPAN	n	SHORT, MEDUIM, LONG
   12. REL-L	n	S, S-F, F
   13. TYPE	n	WOOD, SUSPEN, SIMPLE-T, ARCH, CANTILEV, CONT-T


8. More complicated attributes:

   One can use a hierarchical structure for the Type property. There are two
   options.

	option 1 (use examples without modification)
        --------

			Type 
	     /      /	     \     \
	   /       /	      \	     \
	wood	suspen		arch	truss
				       /  |    \
				     /    |      \
				cantilev  cont-t   simple


	option 2 (requires changes in the Type property - specified bellow)
	--------

				Type 

		 /     	/        |	    	    \	
	   	/     /	         |   		     \
	    wood   suspen	arch		      truss
				/ \	    	   /  |  \    \
			      /	    \	  	 /    |   \	\
			tied-a    not-tied  cantilev cont-t simple arch-t


	Change the Type  property of the following examples (in both V1 and V2):
		E28   ->  arch-t
		E91,E90,E84,E83,E73  -> tied-a
		E97,E78,E77,E75,E66,E64,E43  -> not-tied


9. Missing Attribute Values:
   Attribute #: 	# instances with missing values:
    2			 1
    6			27
    7			16
    8			 2
    9			 6
   10			 2
   11			16
   12			 5
   13			 3
<pre>
</body>
</html>