





          East-West Challenge: Results of Competition 2

Legend:
_______

Entry 		is a codeword used by us for scoring purposes
Oracle		is a codeword used to denote theories known to us  to  be  within
		the bottom quartile of complexity scores
Accuracy	is the accuracy of the subsymbolic entry, taking  as  oracle  the
		theory in the Oracle column
Contestant	is the name of the contestant

Notes:
______

1. Accuracy is measured as follows. For each oracle/entry  pair  the
   following 2x2 table is calculated using the file 100trains.pl:


			    Oracle class
		    |   East      |    West
	------------------------------------------
	            |             |
	       East |    N1       |     N2
	Entry	    |             |
	class ------------------------------------
	            |             |
	      West  |    N3       |     N4
		    |             |
	------------------------------------------
	
	Accuracy =  (N1 + N4) / (N1 + N2 + N3 + N4)

2. We know of 9 theories in the bottom quartile of complexity scores
   relevant  to  Competition  1.  Ordered by complexity score, these
   are:


	Theory		Size		Details
	______		____		_______

	mpage		16		Michie-Page inspired by Theory pfahr2
	pfahr2		16		Bernhard Pfahringer
	inglis		19		Stuart Inglis
	pfahr1		19		Bernhard Pfahringer
	turney		19		Peter D Turney
	weka		19		WEKA ML Project
	x		19		Theory X
	aqdt1		20		Ibrahim F Imam
	aqdt2		20		Ryszard S Michalski


3. Winner is decided on the basis of the highest accuracy using  any
   of  the  theories listed above as oracles. If two or more entries
   have the same accuracy, then a tie-break rule compares accuracies
   when  using  the winner of Competition 1, (Theory pfahr2) as ora-
   cle.

4. In the note "UPDATE ON INDUCTIVE LEARNING COMPETITIONS" issued by
   us  on  August 3 1994, we described necessary changes to the file
   100trains.pl used to score this competition. All entries received
   before   August   3  were  scored  using  the  older  version  of
   100trains.pl.  This affects Entry  hart  and  Entry  quin  listed
   below.


Winner:
_______
			Peter D. Turney 
			Knowledge Systems Laboratory
			National Research Council Canada
			Ottawa, Ontario, Canada, K1A 0R6
			peter@ai.iit.nrc.ca

Comments:
_________

Due to an oversight on our part, we failed to  realise  that  any
entry  that scored in the bottom quartile of Competition 1, could
be resubmitted to Competition 2 in the  guise  of  a  subsymbolic
learner.   Under  the  rules  of  engagement, this entry would be
guaranteed to have 100% agreement with at least one of  the  ora-
cles, namely itself. This error has been rightly punished by each 
of: Entry aqdt1, Entry aqdt2 and Entry turney.


Details:
________


Entry		Oracle		Accuracy	Contestant
_____		______		________	__________

aq17hci		aqdt1		57%		Janusz Wnek
		aqdt2		54%
		inglis		53%
		mpage		51%
		pfahr1		53%
		pfahr2		58%
		turney		60%
		weka		53%
		x		52%

aqdt1		aqdt1		100%		Ibrahim F. Imam
		aqdt2		81%
		inglis		69%
		mpage		54%
		pfahr1		69%
		pfahr2		61%
		turney		43%
		weka		69%
		x		47%

aqdt2		aqdt1		81%		Ryszard S Michalski
		aqdt2		100%
		inglis		60%
		mpage		57%
		pfahr1		60%
		pfahr2		60%
		turney		48%
		weka		60%
		x		52%

hart		aqdt1		65%		G R Hart
		aqdt2		61%
		inglis		68%
		mpage		60%
		pfahr1		68%
		pfahr2		61%
		turney		61%
		weka		68%
		x		58%

imam		aqdt1		69%		Ibrahim F Imam
		aqdt2		66%
		inglis		70%
		mpage		63%
		pfahr1		70%
		pfahr2		60%
		turney		56%
		weka		70%
		x		58%

quin		aqdt1		74%		Ross Quinlan
		aqdt2		78%
		inglis		47%
		mpage		53%
		pfahr1		47%
		pfahr2		62%
		turney		52%
		weka		47%
		x		53%

turney		aqdt1		43%		Peter D Turney
		aqdt2		48%
		inglis		52%
		mpage		85%
		pfahr1		52%
		pfahr2		72%
		turney		100%
		weka		52%
		x		90%
