JATOON - Documentation

4. Examples of data sets

Objects in data sets for JATOON should be separated by a newline, ie. one object, one line. Within each line, variables can be space- or tab- separated.

In training sets for Kohonen NNs, every object should include a class label after the last variable. Class labels must be A, B, C, D, E, F, G, H, or I. The labels will be used to color the Kohonen surface. In training sets for CPG NNs, the output for every object should be included after the last input variable, and a string at the end.

Objects that are to be mapped by a trained Kohonen NN must be labeled with a string after the last variable. The string will be used to identify the object in the map and can't include spaces.

Objects that are to be mapped by a trained CPG NN must be labeled with a string after the last input variable. The string will be used to identify the object in the map and can't include spaces. If an output variable (optional) is included after the last input variable and before the label it will not be considered for the computation. (after all, one of the motivations for mapping objects on a trained SOM is to obtain predictions for new objects, for which no output is known!).

In training and test sets for BPG NNs every object must include the outputs after the inputs (no labels!). The number of input values in the data sets must match the number of input neurons and the number of output values must match the number of output neurons. Objects to be predicted by a trained BPG NN should include the input variables (output variables at the end will not be considered for the computation).

Object formats for JATOON data sets (click on the links to get sample data sets)

Kohonen NN Training set input(1) input(2) input(3) ... input(m) A
Kohonen NN Objects for mapping input(1) input(2) input(3) ... input(m) label
CPG NN Training set input(1) input(2) input(3) ... input(m) output label
CPG NN Objects for mapping input(1) input(2) input(3) ... input(m) label
BPG NN Training and test sets input(1) input(2) ... input(m) output(1) output(2)... output(p)
BPG NN Objects for predicting input(1) input(2) ... input(m)


© 2001-2009 João Aires de Sousa