Thyroid
1. 5 features
2. number of train patterns = 140
3. number of test patterns = 75
4. How they do stuff:
- there are 100 test and 100 train samples http://ida.first.fraunhofer.de/projects/bench/
- pick 100 random samples and make them ur training samples. use the other 100 as test samples
- create a similarity matrix (doing what they did) - kernel params are determined using 5-fold cross validation on the training set
- run the classifier - classifier parameters are determined using 5-fold cross validation on the training set.
5. If we were to use our own similarity metric, features:
(from http://www.ics.uci.edu/~mlearn/MLSummary.html
Another thyroid database from Stefan Aeberhard

    * 3 classes, 215 instances, 5 attributes
    * No missing values )
called new-thyroid.names
all attributes are continuous :(
a. class attribute (1,2,3)
b. percentage
c. decimal value
d. decimal value
e. decimal value