Train the classificator.

Features is a tuple of any floating point numbers or integers (attributes) which has to be assigned to the class Class. This class is specified by an integer. You may use procedure enquire_class_box later to find the most probable class for any array (=tupel). The algorithm tries to describe the set of arrays of one class by hyper cuboids in the feature space. On demand you may even create several cuboids per class. Hence it is possible to learn disjunct concepts, too. I.e such concepts which split in several "cluster" of points in the feature space. The data structure is hidden to the user and only accessible with such procedures which are described in this chapter.

It is possible to specify attributes as unknown by indicating the symbol '*' instead of a number. If you specify n values, then all following values, i.e. the attributes n+1 until 'max', are automatically supposed to be undefined.

You may call the procedures learn_class_box and enquire_class_box alternately, so that it is possible to classify already in the phase of learning. By this means you could see when a satisfying behavior had been reached.

The classificator is going to be bigger using further training. This means, that it is not advisable to continue training after reaching a satisfactory behavior.

learn_class_box returns 2 (H_MSG_TRUE) for a normal case. An exception handling is raised if there are memory allocation problems. The number of classes is constrained. If this limit is passed, an exception handling is raised, too.

learn_class_box is local and processed completely exclusively without parallelization.

create_class_box, enquire_class_box

test_sampset_box, learn_class_box, enquire_class_box, write_class_box, close_class_box, clear_sampset

test_sampset_box, close_class_box, create_class_box, enquire_class_box, learn_sampset_box

Foundation