Output Validating Rules

Output Rules.   Output column validating rules can be used to override the validating controls.

 

The validating stops in the Controls dialog apply to every Output Column in the same way.   If more than one output is used it is possible to apply different validating rules to each column.   A score can be used with each rule so that training can be stopped as soon as a certain score is reached.   Rules can be defined to score points if the output value is within a range or if the value is correct after rounding it to a multiple.   Both methods can be used together.  

 

Consider predicting horse race results with a network trained from historical performance of named horses.   The output will most probably be the finishing place of each horse.   This may be an integer that is the actual place but it is more likely to be 3, 2, 1 for the first, second, and third.   A second output could be a boolean to indicate that the horse won the race.   With rules the first column could be set to score 1 if the place was correct.   The second column rule could be set to score 10 if the Win was correctly predicted.