Practical Neural Network Recipies in C++

382

Chapter

20

ing sumO and

sum1

by the prior probability and/or cost before return-

ing the ratio of

sum1

to their sum.

double parzconf ( // Returns conf in alternative hypothesis

int nO , // N of samples in null hypoth collection

int n1 , // And alternative

double *h0 , // Null hypothesis samples

double *h1 , // And alternative

double sigma , // Scale parameter (0.02 to 0.05 best)

double observed // Observed value to be classified

)

{

}

Multiple Classes

int i ;

double d, sumO, sum1 ;

sumO =

1.e-30

; // Insurance against dividing by 0 later

for (i=0 ; i<n0 ; i++) {

d = (observed - h0[i]) / sigma ;

sumO += exp ( - d * d ) ;

}

sumO /= (double ...