Statistics FAQ
About fitness measures
These measure various qualities of each circuit. When sorting individuals for rank selection, a fitness measure lower down in this list will only be compared if all those above it are equal. Thus a dictionary order is established in which measures higher up in the list are more important.
- Task Fitness: How well it performs its main functionality. For eg. how good a multiplier circuit is at multiplying.
- Self-Testing: This is a loose self-checking metric. In some runs it relates to a circuits capacity of detecting a fault when all input vectors have been applied. In other runs it measures a circuits capability of diagnosing single faults instead of multiple. This fitness measure is akin to a general impression of the problem at hand giving an impression of the big picture. This is designed to avoid evolution getting stuck at local optima by optimizing for particular cases of the Fault Secure while ignoring others.
- Fault-Secure: A more complete self-checking metric than the above it drives the fine tuning of solutions so they meet the most stringent requirements. Following the examples above it would check for diagnosing capabilities under each input vector and under multiple faults.
- Parsimony: How many logic units it uses. All circuits are sythesised using two input look-up table elements.
- Transitions: This is a fitness metric which applies to Sequential self-checking. It measures how many incorrect state transitions are signalled. A fitness of one for this means that there are no unsignaled incorrect state transitions.
All fitnesses are normalized to the [0,1] range where 0 is worst and 1 is best. Parsimony never reaches 1 but the larger the value, the smaller the circuit.