3. Creating a Feature Vector

Before you can do anything useful with a neural network, you need to have some feature vectors defined. Feature vectors must be of the class ExtInput or (more usually) a class which is derived from it. The ExtInput class is defined in `ann/ann.h' and if you have a look through that file you will see that it is (eventually) derived from the Standard Template Library class std::vector<float>. The following code fragment shows a na@"{i}ve way to create a feature vector.

#include <ann/ann.h> // Create a feature vector of length 4 ann::ExtInput featureVector(4); featureVector[0]=0.907; featureVector[1]=1.092; featureVector[2]=54.0; featureVector[3]=-3.409;

In real life examples however, your feature vector comprises elements of some physical quantity, such as pixel values of an image, a word frequency table from a passage of text, or some physical quantities. So you will probably want to define your own class to be used in place of ann::ExtInput. The beauty of the C++ inheritance mechanism is that this can be done very easily:

#include <ann/ann.h> // A class to represent physical properties of human individuals class VitalStatistics : public ann::ExtInput { public: // Construct a VitalStatistics object VitalStatistics(float height, float weight) : ann::ExtInput(2) { (*this)[0] = height; (*this)[1] = weight; } }

Now you can instantiate a VitalStatistics object and use it anywhere you could have used a ann::ExtInput:

int main(int argc, char **argv) { try { // Make a VitalStatistics VitalStatistics vs1(1.59,67,45); // and another VitalStatistics vs2(1.82,95,03); // rest of the program goes here } catch ( ... . . . }

These examples show the principle of how it works. In practice, you would be creating hundreds of instances, and would therefore probably read the data from a file or other external source.