| 1. Title: Small Soybean Database
| 
| 2. Sources:
|      (a) Michalski,R.S. Learning by being told and learning from
|          examples: an experimental comparison of the two methodes of knowledge
|          acquisition in the context of developing an expert system for soybean
|          desease diagnoiss", International Journal of Policy Analysis and
|          Information Systems, 1980, 4(2), 125-161.
|      (b) Donor: Doug Fisher (dfisher%vuse@uunet.uucp)
|      (c) Date: 1987
| 
| 3. Past Usage:
|      See the soybean-large.names
| 
| 4. Relevant Information Paragraph:
|      A small subset of the original soybean database.  See the reference
|      for Fisher and Schlimmer in soybean-large.names for more information.
|     
|      Steven Souders wrote:
| 
|     > Figure 15 in the Michalski and Stepp paper (PAMI-82) says that the
|     > discriminant values for the attribute CONDITION OF FRUIT PODS for the
|     > classes Rhizoctonia Root Rot and Phytophthora Rot are "few or none"
|     > and "irrelevant" respectively.  However, in the SOYBEAN-SMALL dataset
|     > I got from UCI, the value for this attribute is "dna" (does not apply)
|     > for both classes.  I show the actual data below for cases D3
|     > (Rhizoctonia Root Rot) and D4 (Phytophthora Rot).  According to the
|     > attribute names given in soybean-large.names, FRUIT-PODS is attribute
|     > #28.  If you look at column 28 in the data below (marked with arrows)
|     > you'll notice that all cases of D3 and D4 have the same value.  Thus,
|     > the SOYBEAN-SMALL dataset from UCI could NOT have produced the results
|     > in the Michalski and Stepp paper.
| 
|     I do not have that paper, but have found what is probably a later
|     variation of that figure in Stepp's dissertation, which lists the
|     value "normal" for the first 2 classes and "irrelevant" for the latter
|     2 classes.  I believe that "irrelevant" is used here as a synonym for
|     "not-applicable", "dna", and "does-not-apply".  I believe that there
|     is a mis-print in the figure he read in their PAMI-83 article.
|  
|     I have checked over each attribute value in this database.  It 
|     corresponds exactly with the copies listed in both Stepp's and Fisher's
|     dissertations.
|      
| 5. Number of Instances: 47
| 
| 6. Number of Attributes: 35 (all have been nominalized)
|      -- All attributes here appear with numeric values
|  
| 7. Attribute Information: 
|      -- derivable from soybean-large.names
| 
| 8. Number of Missing Attribute Values: 0
| 
| 9. Class Distribution: 
|    1. D1: 10
|    2. D2: 10
|    3. D3: 10
|    4. D4: 17

D1, D2, D3, D4.

Attribute 0:	continuous
Attribute 1:	continuous
Attribute 2:	continuous
Attribute 3:	continuous
Attribute 4:	continuous
Attribute 5:	continuous
Attribute 6:	continuous
Attribute 7:	continuous
Attribute 8:	continuous
Attribute 9:	continuous
Attribute 10:	continuous
Attribute 11:	continuous
Attribute 12:	continuous
Attribute 13:	continuous
Attribute 14:	continuous
Attribute 15:	continuous
Attribute 16:	continuous
Attribute 17:	continuous
Attribute 18:	continuous
Attribute 19:	continuous
Attribute 20:	continuous
Attribute 21:	continuous
Attribute 22:	continuous
Attribute 23:	continuous
Attribute 24:	continuous
Attribute 25:	continuous
Attribute 26:	continuous
Attribute 27:	continuous
Attribute 28:	continuous
Attribute 29:	continuous
Attribute 30:	continuous
Attribute 31:	continuous
Attribute 32:	continuous
Attribute 33:	continuous
Attribute 34:	continuous