2d interpolation of n-dimensional data

1 view (last 30 days)
Curtis Wen on 24 Aug 2021
Commented: Curtis Wen on 25 Aug 2021
I have this set of data (simplified)
x = [1 2 3]
y = [3 7 2]
temp = [22 25 23; 33 25 17; 19 25 26];
each X Y coordinate pair corresponds to a column in temp. So (X,Y) = (1,3) -> [22; 33; 19]. I'm trying to setup a 2D interpolation which gives me interpolated values of temp from input XY. My solution which works, but is slow, is to create a scatteredinterpolant for every row of temp ie
int1 = scatteredinterpolant([1 2 3]', [3 7 2]', [22 25 23]')
int2 = scatteredinterpolant([1 2 3]', [3 7 2]', [33 25 17]')
int3 = scatteredinterpolant([1 2 3]', [3 7 2]', [19 25 26]')
and then using the scatteredinterpolants 3 times. My actual dataset of temp has 30000 elements per coordinate pair, so this goes pretty slow. Any suggestions?

Sean de Wolski on 24 Aug 2021
You can just change the Values of one scatteredInterpolant object. It won't have to redo the expensive underlying triangulation of x/y then. Example using zeros and ones for your temp.
s = scatteredInterpolant(rand(3,1),rand(3,1),zeros(3,1))
s =
scatteredInterpolant with properties: Points: [3×2 double] Values: [3×1 double] Method: 'linear' ExtrapolationMethod: 'linear'
s(0.3,0.2)
ans = 0
s.Values = ones(3,1) % swap for your second set of values
s =
scatteredInterpolant with properties: Points: [3×2 double] Values: [3×1 double] Method: 'linear' ExtrapolationMethod: 'linear'
s(0.3,0.2)
ans = 1
Curtis Wen on 25 Aug 2021
Its not too much of a time savings since the evaluation of each interp still takes as long, but it does reduce the number of interpolant objects created and reduces my filesize so I'll take it. Thank you!

R2019b

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!