Information Technology Reference
In-Depth Information
training set classification error of 2.857% and a classification accuracy of
97.143%, and a testing set classification error of 1.724% and a classification
accuracy of 98.276%. Thus, this model is a slightly worse predictor than the
model (6.8) created without random numerical constants. More formally, the
model (6.9a) can be automatically translated into the following C++ model:
double ADF0(double d[])
{
double dblTemp = 0.0;
dblTemp = (((d[8]/(d[6]+d[0]))*(d[1]+d[5]))-(-0.586578));
return dblTemp;
}
double ADF1(double d[])
{
double dblTemp = 0.0;
dblTemp = ((((d[6]/d[4])*d[3])+d[2])*d[8]);
return dblTemp;
}
double ADF2(double d[])
{
double dblTemp = 0.0;
dblTemp = (((d[2]+(d[6]+d[0]))*(d[1]+d[5]))/0.316253);
return dblTemp;
}
double apsModel(double d[])
{
double ROUNDING_THRESHOLD = 0.5;
double dblTemp = 0.0;
dblTemp = ((ADF2(d)+(ADF2(d)*ADF2(d)))*
((ADF1(d)/ADF2(d))*(ADF2(d)+ADF0(d))));
return (dblTemp >= ROUNDING_THRESHOLD ? 1:0);
} (6.9b)
Note that different linking functions are used to connect the ADFs in the
main program. Note also that, although ADF 2 is called from five different
places in the main program, this program could be simplified so that this
ADF would be called just twice. Nonetheless, this model is much more ver-
bose than the model (6.8) designed without RNCs as all the ADFs are in-
voked in the main program (ADF 0 and ADF 1 are both used just once). And as
you can see, in this case, with the exception of one (normal nucleoli), all the
analyses are used to make the diagnosis of breast cancer.
Search WWH ::




Custom Search