How to suppress this error?

Hi there

I am receiving the following error message:

Warning: Matrix is singular, close to singular or badly scaled. Results may be inaccurate. RCOND = NaN.
> In mpower>integerMpower (line 80)
  In ^ (line 49)
  In gaussfit (line 111)

The line in question is:

  a = (F2)^(-1)*F'*(y-f0) + a0;

I wish to suppress, not solve, this problem and so I added the following to the beginning of my script, and yet still receive the error message. What am I doing wrong?

warning('off','MATLAB:singularMatrix')

[Sudhakar Shinde]

If you are not bothered about other warnings too, does it make sense to suppress all warnings.

warning off;

[Walter Roberson]

Why are using matrix inversion and matrix multiplication instead of F2\F1 ?

[Jakob Sievers]

This is not a script I originally wrote so I couldn't say. It just worked for my purpose. Though if you have a better way, not to mention a faster way, of doing it I am all ears. My process involves using this equation quite often and would love to shave some time off of it if possible.

[Jakob Sievers]

I just had a look and the file (which fits a gaussian curve) is from 2012, so it might indeed be overdue for an update:

function [sigma, mu,error] = gaussfit( x, y, sigma0, mu0,runmode,idix)
% [sigma, mu] = gaussfit( x, y, sigma0, mu0 )
% Fits a guassian probability density function into (x,y) points using iterative 
% LMS method. Gaussian p.d.f is given by: 
% y = 1/(sqrt(2*pi)*sigma)*exp( -(x - mu)^2 / (2*sigma^2))
% The results are much better than minimazing logarithmic residuals
%
% INPUT:
% sigma0 - initial value of sigma (optional)
% mu0 - initial value of mean (optional)
%
% OUTPUT:
% sigma - optimal value of standard deviation
% mu - optimal value of mean
%
% REMARKS:
% The function does not always converge in which case try to use initial
% values sigma0, mu0. Check also if the data is properly scaled, i.e. p.d.f
% should approx. sum up to 1
% 
% VERSION: 23.02.2012
% 
% EXAMPLE USAGE:
% x = -10:1:10;
% s = 2;
% m = 3;
% y = 1/(sqrt(2*pi)* s ) * exp( - (x-m).^2 / (2*s^2)) + 0.02*randn( 1, 21 );
% [sigma,mu] = gaussfit( x, y )
% xp = -10:0.1:10;
% yp = 1/(sqrt(2*pi)* sigma ) * exp( - (xp-mu).^2 / (2*sigma^2));
% plot( x, y, 'o', xp, yp, '-' );
warning off;
warning('off','MATLAB:singularMatrix')
error=0; %no error
% Maximum number of iterations
Nmax = 50;
if nargin==4
    runmode=1; %output warnings!
end
if( length( x ) ~= length( y ))
    fprintf( 'x and y should be of equal length\n\r' );
    exit;
end

n = length( x );
x = reshape( x, n, 1 );
y = reshape( y, n, 1 );

%sort according to x
X = [x,y];
X = sortrows( X );
x = X(:,1);
y = X(:,2);

%Checking if the data is normalized
dx = diff( x );
dy = 0.5*(y(1:length(y)-1) + y(2:length(y)));
s = sum( dx .* dy );
if( s > 1.5 | s < 0.5 )
    fprintf( 'Data is not normalized! The pdf sums to: %f. Normalizing...\n\r', s );
    error=1;
    y = y ./ s;
end

X = zeros( n, 3 );
X(:,1) = 1;
X(:,2) = x;
X(:,3) = (x.*x);


% try to estimate mean mu from the location of the maximum
[ymax,index]=max(y);
mu = x(index);

% estimate sigma
sigma = 1/(sqrt(2*pi)*ymax);

if( nargin == 3 )
    sigma = sigma0;
end

if( nargin == 4 )
    mu = mu0;
end

%xp = linspace( min(x), max(x) );

% iterations
ii=0;
while ii<Nmax
    ii=ii+1;
%    yp = 1/(sqrt(2*pi)*sigma) * exp( -(xp - mu).^2 / (2*sigma^2));
%    plot( x, y, 'o', xp, yp, '-' );

    dfdsigma = -1/(sqrt(2*pi)*sigma^2)*exp(-((x-mu).^2) / (2*sigma^2));
    dfdsigma = dfdsigma + 1/(sqrt(2*pi)*sigma).*exp(-((x-mu).^2) / (2*sigma^2)).*((x-mu).^2/sigma^3);

    dfdmu = 1/(sqrt(2*pi)*sigma)*exp(-((x-mu).^2)/(2*sigma^2)).*(x-mu)/(sigma^2);

    F = [ dfdsigma dfdmu ];
    a0 = [sigma;mu];
    f0 = 1/(sqrt(2*pi)*sigma).*exp( -(x-mu).^2 /(2*sigma^2));
    if any(isnan(F(:)))
        error=1;
        ii=Nmax;
    else
        F2=F'*F;
        if rank(F2)<min(size(F))
            error=1;
            ii=Nmax;
        else
            a = (F2)^(-1)*F'*(y-f0) + a0;
            sigma = a(1);
            mu = a(2);
            if( sigma < 0 )
                sigma = abs( sigma );
                if runmode==1
                    fprintf( 'Instability detected! Rerun with initial values sigma0 and mu0! \r' );
                    fprintf( 'Check if your data is properly scaled! p.d.f should approx. sum up to 1 \r' );
                end
                error=1;
                ii=Nmax;
            end
        end
    end
    
end

Answer 1

[the cyclist]

Two thoughts:

First: Are you certain you have the correct warning? After you run your code that gives the warning, what is the output of

w = warning('query','last')

(Your warning does look like it matches the one you are turning off, so I'm guessing that's not the issue.)

Second: I believe that turning off warnings only persists for the session. Does this warning happen during the same session in which you turned it off?

[Jakob Sievers]

The above solution worked for my needs. Thanks for your attention though :)

Answer 2

[Real User]

warning('off','MATLAB:nearlySingularMatrix')

But the cyclist's answer works for any error (that's how I found this).