I have been trying to use this package with nonlinear-optimization and ad for maximum likelihood computations, however this is highly non-practical since distributions always return Double -> Double and ad needs some kind of polymorphic interface to work. Are there any barriers for such a polymorphic interface?
One problem I see is the need for different contraints depending on the distribution (Erf for Normal, Num for others).
I have been trying to use this package with
nonlinear-optimizationandadfor maximum likelihood computations, however this is highly non-practical since distributions always returnDouble -> Doubleandadneeds some kind of polymorphic interface to work. Are there any barriers for such a polymorphic interface?One problem I see is the need for different contraints depending on the distribution (
Erffor Normal,Numfor others).