
Need to fit some data to a surface (function of two independent variables) in Octave?
Matlab has a function to do it directly but Octave needs some help.
The following example demonstrates one possible method using the optimisation package.
# Perform surface fitting to data in octave using optimisation package
# In principle I can see no reason why this technique will not work for a
# higher number of dimensions of observed data.
# Example runs in Octave 5.2.0
# J E Green, Bear Instruments LTD © 2020.
# www.bearinstruments.co.uk
# Power Electronics, Embedded Systems & Instrumentation
clear
# required optimisation package
pkg load optim
# for demonstration, use the communications package to add noise to the
# contrived data
pkg load communications
# independents and observations
# Controlled variables (things you have adjusted to get your data)
x = [-5 -4 -3 -2 -1 0 1 2 3 4 5];
y = [-3 -2 -1 0 1 2 3]';
# dependent variable (i.e. your observed values, I made up in this case)
obs = [
42 31 22 15 10 7 6 7 10 15 22
33 23 15 9 5 3 3 5 9 15 23
26 17 10 5 2 1 2 5 10 17 26
21 13 7 3 1 1 3 7 13 21 31
18 11 6 3 2 3 6 11 18 27 38
17 11 7 5 5 7 11 17 25 35 47
18 13 10 9 10 13 18 25 34 45 58];
# add noise to the "observations" to give a signal to noise ratio of 0 dB.
obs = awgn(obs,0);
# Function you hope that the data will fit.
# p is the weights (coefficients of the function). In this case it is a
# quadratic in two variables. To generate the obs[] above I have used
# p(n) = 1 for n = 1 to 6.
# In other words all the weights are unity in the example.
function retval = myfunc (p,x,y)
retval = p(1).*x.^2 + p(2).*x.*y + p(3).*y.^2 + p(4).*x + p(5).*y + p(6);
endfunction
# initial values, just to throw it off I've made them negative and ten times too
# small.
init = [-0.1; -0.1; -0.1; -0.1; -0.1; -0.1;];
# Set some parameters of the optimisation problem
settings = optimset ("MaxIter", 100, "TolFun", 1e-6);
# start optimization
[p, residuals, cvg, outp] = nonlin_residmin( @(p) (myfunc(p,x,y) - obs), init, settings)
# Generate the 'ideal' data for plotting. This will be the same as obs[]
# because this problem is contrived to demonstrate the method.
for i=1:size(x)(2)
for j = 1:size(y)(1)
model_values(j,i) = myfunc(p,x(i),y(j));
endfor
endfor
# plot the fitted data
figure
surf(x,y,model_values)
title ("Fitted Values");
# plot the "experimental" data
figure
mesh(x,y,obs)
title ("Experimental Values");
# Dump the matrices to the command window
model_values
obs
# plot the residuals.
figure
surf(residuals)
title ("Residuals");
# total sum of squares
SST = sum(sum((obs - mean(mean(obs))).^2));
# sum of squared residuals
SSR = sum(sum(residuals.^2));
# R^2
Rsq = (1 - (SSR./SST))
# output the vector of coefficients (they should all be close to 1).
p
