 
  
  
  
  
 Next: Confid. Intervals for Disc. 
Up: Evaluating Hypothesis and Experimental 
 Previous: Question Assumptions
 
-  Given a large data set,   , suppose we apply learning
algorithm , suppose we apply learning
algorithm to to to construct classifier to construct classifier ; how
accurately will ; how
accurately will classify new examples? classify new examples?
-  The classifier   may have been constructed using part of the
data, but there is enough data remaining for a separate test set.
Measure the accuracy of may have been constructed using part of the
data, but there is enough data remaining for a separate test set.
Measure the accuracy of on the test set and construct a binomial
confidence interval.  Note on the test set and construct a binomial
confidence interval.  Note need not have been produced by a
learning algorithm. need not have been produced by a
learning algorithm.
 
Patricia Riddle 
Fri May 15 13:00:36 NZST 1998