From: Raymond Mooney <mooney@cs.utexas.edu>
Date: Mon, 12 Mar 90 19:19:58 CST

Regarding the data we used.  We used a later version of the soybean
data sent to us by Bob Stepp which I believe to be the version used by
Reinke in his MS thesis on the GEM system (this has 17 diseases as
referenced by Spackman, ML88).  I also got the version you have (19
diseases, 35 features) from Bob Stepp which I believe is the version
used in the original experiments from Michalski & Chilausky (this has
missing features which is why we decided to use the 17 disease version
in our experiments).  The M&C paper references only 15 diseases (35
features) and the first 15 diseases in my 19 disease set match those
in the M&C paper.  I guess the last 4 diseases just weren't reported
in the paper (wonder why??).  The existence of the 4 disease
clustering set adds even more confusion to any reference to using
soybean data.

Hope this helps some.

-Ray