Lsqcurvefit - problem with matrix input

Hi everyone,

I'm trying to use lsqcurvefit to optimize one set of data to look as close as another set.

For this purpose I try to optimize the difference between a fixed vector empty(:,2) and another vector, which I know can be quite nicely tuned using the function (var(:,1) - x(1).*var(:,2) + x(2)). So, the Ydata is obviously a matrix and even though I saw examples of functions being fitted using lsqcurvefit with matrices and also the manual says it's possible, I still get an error. The sizes and dimensions of all inputs are correct, so it can't be the problem.

Here's my code:

%Uploading data from other sources:
var(:,1) = abs_full;
var(:,2) = absorbance_molecule;

%Defining the optim. function + fitting
myfun = @(x,var) empty(:,2) - (var(:,1) - x(1).*var(:,2) + x(2));
x0 = [0.53,0.1];
x = lsqcurvefit(mufyn,x0,wav_full,var);

Here's the error:

Index in position 2 exceeds array bounds. Index must not exceed 1.

Error in kinetics_UV_shifting_background>@(x,var)empty(:,2)-(var(:,1)-x(1).*var(:,2)+x(2)) (line 290)

myfun = @(x,var) empty(:,2) - (var(:,1) - x(1).*var(:,2) + x(2));

Error in lsqcurvefit (line 225)

initVals.F = feval(funfcn_x_xdata{3},xCurrent,XDATA,varargin{:});

Error in kinetics_UV_shifting_background (line 293)

x = lsqcurvefit(myfun,x0,wav_full,var);

Caused by:

Failure in initial objective function evaluation. LSQCURVEFIT cannot continue.

I would be grateful for any help!

Answer 1

[Star Strider]

Index in position 2 exceeds array bounds. Index must not exceed 1.

Check to see how many columns ‘empty’ has. It may only have one.

[Maciej Piejko]

I tried to reduce this two-column vector to only one column (the one I need), but I get the same error.

[Maciej Piejko]

When I change the function so that it uses only one column of the var variable (for example only var(:,2)) it works without error. But of course I need both to do anything useful. So it would seem the problem comes from using a matrix as input, even though it should be fine for lsqcurvefit.

[Star Strider]

Having all the data and all the necessary code to work with would definitely help.

EDIT — (9 Jul 2023 at 13:44)

You are creating the ‘var’ matrix, so I doubt that it is the problem. The problem appears to be the ‘empty’ variable, since it appears to have only one column. That throws the error.

.

[Maciej Piejko]

The rest of the code has 200 lines, but I will just show the essential parts and I highlighted the parts where the code has an issue:

I include the rest of the related code (still not everything, the rest is irrelevant), and I highlight the most essential part that has an issue somewhere.

prompt = {'Wavelength used for kinetics'};

dlgtitle = 'Select wavelength [nm]';

dims = [1 50];

answer = inputdlg(prompt,dlgtitle,dims);

obs_wav = str2double(answer{1});

file_cavity_bg = dlmread('CM-CC-024D cavoffNoOH100eqlutidine20%ACN_RawData-070.txt','');

wavelength = file_cavity_bg(:,1);

files = dir('*.txt');

file_number = 0;

for i=1:length(files)

if length(files(i).name()) > 35;

file_number = file_number + 1;

end

end

min_lim = 610;

max_lim = 636;

initial_spectrum=readmatrix(files(1).name());

wav_initial = initial_spectrum(:,1);

absorbance_initial = initial_spectrum(:,2);

molecule = dlmread('molecule.txt','');

wav_molecule = molecule(:,1);

absorbance_molecule = molecule(:,2);

final = dlmread('final.txt','');

empty = final(:,2);

for i=1:file_number

spectra=readmatrix(files(i).name());

wav = spectra(:,1);

abs = spectra(:,2);

wav_full = spectra(:,1);

abs_full = spectra(:,2);

var(:,1) = abs_full;

var(:,2) = absorbance_molecule;

belowrange = (wav < min_lim);

aboverange = (wav > max_lim);

remove = or(belowrange,aboverange);

wav(remove) = [];

abs(remove) = [];

[~,idx] = max(abs);

maximum(i,1) = wav(idx,1);

myfun = @(x,var) empty(:,1) - (var(:,1) - x(1).*var(:,2) + x(2));

x0 = [0.53,0.1];

x = lsqcurvefit(myfun,x0,wav_full,var);

tmp_bg = abs_full - param(1,1)*absorbance_molecule+param(1,2);

%Find current maximum and phase shift

phase_diff = maximum - max_bg;

abs_bg_shift = zeros(length(abs_full));

if phase_diff > 0

abs_bg_shift = circshift(test,phase_diff);

%abs_bg_shift(phase_diff:end) = abs_bg(length(abs_full));

abs_final = abs_full - abs_bg_shift - tmp_bg;

else

abs_final = abs_full - tmp_bg;

end

a = nlinfit(x,y,@(a0, x) myfun(a0, constants),a0);

idx_kin = find(wav_full == obs_wav);

absorbance_phase_bg(i,1) = abs_final(idx_kin,1);

plot(wav_full,abs_final)

end

[Star Strider]

You define ‘empty’ here:

empty = final(:,2);

so it only has one column (as I suspected).

Just use:

myfun = @(x,var) empty - (var(:,1) - x(1).*var(:,2) + x(2));

instead, without the subscript reference to ‘empty’, although this:

myfun = @(x,var) empty(:,1) - (var(:,1) - x(1).*var(:,2) + x(2));

should also work.

I would replace the dlmread calls with readmatrix however without the data and the rest of the code, I cannot determine what any remaining problems are, if any remain, or obviously how to fix them.

[Maciej Piejko]

Thank you for all the help, it still did not work but I found a workaround.

[Star Strider]

What was the workaround?