#naive bayes classifier in gawk
#usage:  gawk -F, -f nbc.awk Pass=1 train.csv Pass=2 test.csv

Pass==1 {train()}
Pass==2 {print $NF "|" classify()}

function train(    i,h) { 
   Total++;
   h=$NF;     # the hypotheis is in the last column
   H[h]++;    # remember how often we have seen "h"
   for(i=1;i<=NF;i++) {
     if ($i=="?") 
           continue;         # skip unknown values
     Freq[h,i,$i]++
     if (++Seen[i,$i]==1) 
           Attr[i]++}  # remember unique values
}
function classify(         i,temp,what,like,h) {  
   like = -100000;         # smaller than any log
   for(h in H) {           # for every hypothesis, do...  
     temp=log(H[h]/Total); # logs stop numeric errors
     for(i=1;i<NF;i++) {  
       if ( $i=="?" ) 
             continue;     # skip unknwon values
       temp += log((Freq[h,i,$i]+1)/(H[h]+Attr[NF])) }
     if ( temp >= like ) { #  better hypothesis
        like = temp
        what=h}
   }
   return what;