I have 1,000 points of grid (equally spaced) measurement data that consists of input/stimulus (triplet) and output/result (triplet).
This data is populated in a 10x10x10x3 data structure, basically a cube (or 3D LUT) of size 10 and each element is a result/output triplet.
This maps input/stimulus->output/result and is used to simulate or predict the measurement source.
This works well including for interpolated input values that were not part of the original 10 grid data set, using trilinear or tetrahedral interpolation.
The problem that I'm running into is that the results become very inaccurate for low stimulus triplet inputs that contain non-grid values (non-integer).
example:
The 10 cube has indices 0-9.
Mid/high stimulus inputs always produce okay results.
Low stimulus inputs work well if the input triplet contains grid points (integers), e.g. [1.0, 1.0, 2.0] - hence points that were part of the source grid data set.
But a low stimulus input triplet such as [0.574596, 1.12419, 0.0] produces a very inaccurate result/prediction to what the source actually measures at that stimulus.
Is this a limitation of trilinear/tetrahedral interpolation that depends on surrounding points and in low stim there is not much data? what do I need to change to improve the low stimulus interpolation results?
I've tried tricubic interpolation, but unfortunately it seems this algorithm does not work with points that are located on the face/edge of the cube.
I'm working in Python.
Thank you!
