{\rm# GLOBALS: ``F'': frequeny tables;     ``I'' : number of instances; 
#               ``C'': how many classes?; ``N'': instances per class}
{\bf function} update(class,train) 
   {\rm# OUTPUT: changes to the globals.}
   {\rm# INPUT:   a ``train''ing example containing attribute/value pairs 
     #             plus that case's ``class''}
   I++; {\bf if} (++N[class]==1){\bf then} C++{\bf fi} 
  {\bf for} <attr,value> {\bf in} train 
     {\bf if} (value != "?") {\bf then}
        F[class,attr,range]++{\bf fi} 
{\bf function} classify(test)
  {\rm# OUTPUT: ``what'' is the most likely hypothesis for the test case.}
   {\rm# INPUT:    a ``test'' case containing attribute/value pairs.}
   k=1; m=2          {\rm # Control for Laplace and M-estimates.}
   like = -100000    {\rm # Initial, impossibly small likelihood.}
   {\bf for} H {\bf in} N         {\rm# Check all  hypotheses.}
   \{ prior = (N[H]+k)/(I+(k*C))                 {\rm #\(\Leftarrow\)\(P(H)\)}
     temp  = log(prior)
     {\bf for} <attr,value> {\bf in} attributes 
     \{{\bf if} (value != "?") {\bf then}       
       inc= F[H,attr,value]+(m*prior))/(N[H]+m) {\rm #\(\Leftarrow\)\(P(E\sb{i}{\given}H)\)} 
       temp += log(inc) {\bf fi}
     \}
     {\bf if} (temp >= like){\bf then} like = temp; what=class{\bf fi}
   \}
   {\bf return} what