Methods that return expectation values or variance can be thought of as on a continuum of "model moments." Expectation value (mean) is "the first model moment"; variance (centered on mean) is "the second model moment"; besides the first model moment, for a general real "x", "the x model moment" is the x-th power of the per-dimension eigenvalue less the first model moment or mean, like $y _x= \sum_n (v_n - y_1)^x$.
Expectation value and variance methods should be generalized as instances of statistical "moment", in unified output methods. This entails QInterface implementation, shared library wrapping, and maintenance of existing ad hoc optimizations on expectation value and variance queries.
Methods that return expectation values or variance can be thought of as on a continuum of "model moments." Expectation value (mean) is "the first model moment"; variance (centered on mean) is "the second model moment"; besides the first model moment, for a general real "
x
", "thex
model moment" is thex
-th power of the per-dimension eigenvalue less the first model moment or mean, like $y _x= \sum_n (v_n - y_1)^x$.Expectation value and variance methods should be generalized as instances of statistical "moment", in unified output methods. This entails
QInterface
implementation, shared library wrapping, and maintenance of existing ad hoc optimizations on expectation value and variance queries.