Every parameter spans lower to upper (the search is not restricted). A parameter whose interval has 0 as one bound and whose other bound is at least logSpanMin from 0 (the fuel coefficients, 0-100, and youngAge, -50-0; also CMD, 0-50) is searched on a signed log1p scale, sign(x) * log1p(|x|), so the values near its 0 bound, a sign constraint where fitted values sit (1-2 against a bound of 100), get as much of the cube as the far end. The others (bounds off 0, or a narrow range such as yearSpreadSD) are linear. The covariates are scaled, so a coefficient's units are comparable across parameters and logSpanMin is in them.

emulatorSpace(lower, upper, logSpanMin = 20)

Arguments

lower, upper

the bounds, named.

logSpanMin

numeric; see above.

Value

a list: logPars (names), toUnit(P) and fromUnit(U) (matrices, columns names(lower); fromUnit() stays within lower/upper).