Page: 17 prev next

"Real" vs. "Constructed" Datasets

Philosophically, real datasets, taken from physical science applications, seem preferable to simulated datasets because they focus validation efforts in the most important areas.

Constructed datasets, however, can be focused on known software weaknesses, and can offer more specific tests of software properties, than real data can.

Constructed datasets also have sufficient structure to allows analytical computation of correct results.

Finally, constructed datasets have historical precedent.