MATLAB: Optimization Problem: Suggestions from the floor

interpn

Im running MATLAB 2010B and have the optimization toolbox (but not the global optimization toolbox).
I am not sure if I can use interpn.m for my problem.
I have a dependant variable, stored in a 1xK vector. (R)
I have a number (lets say 5) of independant variables stored in seperate 1xK vectors. (L1, L2, L3..L5)
I also have 5 "current" observations for the independant variables. (va, vb, vc…)
I would like to use somekind of optimization/ lookup routine to find the likely value of the dependant variable, r_now.
r_now = myFunc(L1, L2, L3, L4, L5, R, va, vb, vc, vd, ve);
I thought interpn.m might work for myFunc, but I cant get it to. Does anyone have any experince with this function?
None of my variables are montonically increasing, or on a regular grid, and the relationship is quite rough, so I thought the spline option from interpn.m would be useful.
When I place a breakpoint at the start of vi = interpn(varargin) I see (using 3 independants):
varargin =
    [8286x1 double]    [8286x1 double]    [8286x1 double]    [8286x1 double]    [0.5000]    [0.5000]    [0.5000]    'spline'
I get the error message
??? Error using ==> interpn at 155 Wrong number of input arguments or some dimension of V is less than 2.
If interpn.m is not a good approach, can anyone suggest one? It needs to be quick (i.e. not simulated annealing/ stochastic methods)

Best Answer

interpn.m assumes equally spaced data, as per ndgrid.m.
It does not support nonuniformly spaced data.
griddatan.m allows conversion from unequal to equally spaced data using linear interpolation or nearest neighbour.
Another way to approach such a problem is multi-dimensional kernel density estimation.
Related Question