Add metrics to assess and compare DOEs
DOELibrary
has a method compute_phip_criteria
to compute the \varphi_p
criterion of an input samples set; the smaller, the better.
Other metric could be useful to compare input samples sets, e.g. minimal distance criterion and centered L2-discrepancy, as proposed in OpenTURNS.
One could imagine a class method evaluate_quality
returning a kind of namedtuple equipped with comparison methods, to be used as:
first_quality = DOELibrary.evaluate_quality(a_first_doe)
is_first_doe_better = first_quality > DOELibrary.evaluate_quality(another_doe)
l2_discrepancy = first_quality.l2_discrepancy
Edited by Matthias De Lozzo