Why do I receive this error when using cumsum: "Invalid data type. First input argument must be numeric or logical." ?

Hi,

I am a psychologist with very little experience in Matlab and am just trying to run a code (see attachment). The function is:

function [H]=dfaedit(file_name,plot_flag, outfile_flag, out_command_flag)

When calling the fucntion with the input arguments I always receive the error:

Error using cumsum

Invalid data type. First input argument must be numeric or logical.

Error in dfaedit (line 44)

Sum = cumsum(x);

I have tried using different file types (.txt, .dat, .csv), different imported variables (matrix, table, vectors, etc.) and extracting data from vectors but no matter the input I get the same cumsum error. The data itself is just a single column of integer values.

At first I thought that my data wasn't numerical but after having tried out all of the steps mentioned above I am out of ideas. Could anyone help me out?

Sorry if this seems obvious :)

Thanks!

[Torsten]

Could you also include your calling program and the Textfile ?

[lil brain]

Sure I have attached the text file. I guess will calling program you mean the matlab version?

[Torsten]

No, I mean what you wrote to fill the ... below:

function call
  file_name = ...;
  plot_flag = ...;
  outfile_flag = ...;
  out_command_flag = ...;
  H = dfaedit(file_name,plot_flag,outfile_flag,out_command_flag)
end

[lil brain]

I see. I gave them all a 1.

So I put into the command window:

dfaedit('sample_test.txt',1,1,1)

Answer 1

[Torsten]

Try this:

function main
  file_name = 'sample_test.txt';
  H = dfaedit(file_name,1,1,1)
end  

function  [H]=dfaedit(file_name,plot_flag, outfile_flag, out_command_flag)

%  DFA  Calculates and plots detrended fluctuations
%
%   For a 1-d time series in a one column text file named 'file_name',
%	DFA performs a detrended fluctuation analysis on this data. It first integrates
%   the data and then partitions it into different size subsets whose size  
%   varies from MinBox (default = 4) to MaxBox (default = 1/4 the data length). 
%   For the subsets of each size, the best fitting trend lines are found 
%   and their the root mean square residual (Q) is calculated.   The log10
%   of this fluctuation quantity (Q) is plotted against log10 of the points in a subset.
%   The slope of this regression estimates the Hurst exponent (H).  The fractal 
%   dimension can be calculated as 2-H.
%
%   If plot_flag = 1 then a regression plot is created.  If outfile_flag = 1, the file name, H
%   and r^2 are appended to a text file 'dfa_results.txt'.  If out_command_flag = 1, the data 
%   that are the basis for the regression plot are output to the command window.
%
%   Example:  dfa('noise-f.dat', 1, 1, 1)
%
%   I used the Peng's C code algorithm which is available at http://reylab.bidmc.harvard.edu.   
%   It is discussed in Peng et al. Quantification of scaling exponents and crossover  
%   phenomena in  nonstationaryheartbeat time series. Chaos 1995;5:82-87.  I vectorized the 
%   preliminary calculations for MatLab.  But the main loop I did not touch. Still not sure 
%   of the algorithm they are using to calculate the root mean square residual. 
%
%   Syntax:
%   dfa(file_name,plot_flag, outfile_flag, out_command_flag)
%
%   R. C. Schmidt 11-27-01
%-------------------------------------------------------------------------------------
format short g
%x = file_name;
fid = fopen(file_name);
data = textscan(fid,'%f');
fid = fclose(fid);
x = cell2mat(data);

numberpoints = length(x);

MinBox = 4;                 % minbox #of points in a box
MaxBox  = .25*numberpoints; % maxbox set 1/4 the data length
BoxSizeDensity = 4;
LogScaleFactor = power(2.0, 1.0/BoxSizeDensity);
index = 1:numberpoints; 
index = reshape(index, length(index), 1);  %change row to column vector

% Preliminary calculations
Sum = cumsum(x);
SumOfSumsSquared = cumsum(Sum.*Sum);
SumOfSums = cumsum(Sum);
SSc = cumsum(index.*Sum);

% now find best fit lines and find fluctuation about the line 
% loop for each box size, from MinBox to MaxBox
iteration = 1;
BoxSize = MinBox;
TempBoxSize = MinBox;
while BoxSize <= MaxBox
    s = 0.0;
    r1 = 1./(BoxSize + 1.0);
    Det = 12.0*r1*r1*r1/(1.0 - r1*r1);
    vv = BoxSize*(2.0*BoxSize + 1.0)/6.0;
    
    for j = 2:numberpoints - BoxSize
        dhh = SumOfSumsSquared(j + BoxSize) - SumOfSumsSquared(j-1);
        dhu = SumOfSums(j + BoxSize) - SumOfSums(j-1);
        dhv = SSc(j + BoxSize) - SSc(j-1) - dhu*j;
        
        s = s + dhh - (dhv*dhv +dhu*dhu*vv - dhv*dhu*BoxSize)*Det;
        j = j + 1;
    end
    
    den = (numberpoints - BoxSize)*(BoxSize + 1.0);
    Fluctuation = sqrt(s/den);
    
    log_points_in_box(iteration,1) = log10(BoxSize);  
    log_Q(iteration,1) = log10(Fluctuation);
    show_results(iteration,:) = [ iteration  BoxSize  Fluctuation  log_points_in_box(iteration,1)  log_Q(iteration,1) ];
    iteration = iteration + 1;
    
    % update the box size 
    TempBoxSize = TempBoxSize*LogScaleFactor;
    while round(TempBoxSize) <  BoxSize + 1.0
        TempBoxSize = TempBoxSize*LogScaleFactor;
    end
    BoxSize = round(TempBoxSize);
end

% got all boxes; now calculate H via trendline
r_trend = corrcoef(log_points_in_box, log_Q);
coefs = polyfit(log_points_in_box, log_Q,1);
r_line = polyval(coefs,log_points_in_box);

% calculate dimension and Hurst
H = coefs(1);
D = 2-H;

%display results in command window
if out_command_flag == 1
    disp('         iteration    points       Q       Log10(pnts)   Log10(Q)')
    disp(show_results)
end

% display results in a figure
if plot_flag == 1
    figure
    h = axes('Position', [0 0 1 1], 'Visible', 'off');
    axes('Position',[.1 .2 .75 .75])
    plot(log_points_in_box, log_Q,'b+')
    hold on
    plot(log_points_in_box,r_line, 'k')
    ylabel('log10(Q)');
    xlabel('log10(Points in Subset)');
    title(file_name);
    set(gcf, 'CurrentAxes', h);
    str(1) = {[sprintf('%2.5f',coefs(1)),'x + ', sprintf('%2.5f',coefs(2)), ' = y,   r^2 = ', sprintf('%2.2f',r_trend(1,2)^2), ',  n = ', sprintf('%d',length(log_Q)) ]};
    text(.1, .02, str, 'FontSize', 10, 'Color', 'r');
    str(1) = {['H = ', sprintf('%2.3f',H) '  D = ', sprintf('%2.3f',D) ]};
    text(.5, .02, str, 'FontSize', 10, 'Color', 'r');
end

% append results to a file rd_results.txt
if outfile_flag == 1
    fid = fopen('dfa_results.txt','a');
    fprintf(fid,'%s  %4.4f   %2.4f\n', file_name, H, r_trend(1,2)^2);
    status = fclose(fid);
end
end

[lil brain]

This worked! Great this is actually better as it allows me to use files and not just variables.

Answer 2

[Star Strider]

The code (that appears to have been written in 2001) doesn’t make sense!

The first argument to the function is ‘file_name’ that I assume would be a character array. It is subsequently assigned to ‘x’ and then the ‘Sum’ variable is assigned as cumsum(x). Doing any sort of calculation on a character array (without first converting to a numeric representation of the elements of the character array) isn’t possible.

That is throwing the error.

Please go back and review the documentation to determine what the arguments are supposed to be.

opts = weboptions('ContentType','text');
W = webread('https://www.mathworks.com/matlabcentral/answers/uploaded_files/857175/dfaedit.m', opts)

W = 
    'function  [H]=dfaedit(file_name,plot_flag, outfile_flag, out_command_flag)
     
     %  DFA  Calculates and plots detrended fluctuations
     %
     %   For a 1-d time series in a one column text file named 'file_name',
     %	DFA performs a detrended fluctuation analysis on this data. It first integrates
     %   the data and then partitions it into different size subsets whose size  
     %   varies from MinBox (default = 4) to MaxBox (default = 1/4 the data length). 
     %   For the subsets of each size, the best fitting trend lines are found 
     %   and their the root mean square residual (Q) is calculated.   The log10
     %   of this fluctuation quantity (Q) is plotted against log10 of the points in a subset.
     %   The slope of this regression estimates the Hurst exponent (H).  The fractal 
     %   dimension can be calculated as 2-H.
     %
     %   If plot_flag = 1 then a regression plot is created.  If outfile_flag = 1, the file name, H
     %   and r^2 are appended to a text file 'dfa_results.txt'.  If out_command_flag = 1, the data 
     %   that are the basis for the regression plot are output to the command window.
     %
     %   Example:  dfa('noise-f.dat', 1, 1, 1)
     %
     %   I used the Peng's C code algorithm which is available at http://reylab.bidmc.harvard.edu.   
     %   It is discussed in Peng et al. Quantification of scaling exponents and crossover  
     %   phenomena in  nonstationaryheartbeat time series. Chaos 1995;5:82-87.  I vectorized the 
     %   preliminary calculations for MatLab.  But the main loop I did not touch. Still not sure 
     %   of the algorithm they are using to calculate the root mean square residual. 
     %
     %   Syntax:
     %   dfa(file_name,plot_flag, outfile_flag, out_command_flag)
     %
     %   R. C. Schmidt 11-27-01
     %-------------------------------------------------------------------------------------
     format short g
     x = file_name;
     numberpoints = length(x);
     
     MinBox = 4;                 % minbox #of points in a box
     MaxBox  = .25*numberpoints; % maxbox set 1/4 the data length
     BoxSizeDensity = 4;
     LogScaleFactor = power(2.0, 1.0/BoxSizeDensity);
     index = 1:numberpoints; 
     index = reshape(index, length(index), 1);  %change row to column vector
     
     % Preliminary calculations
     Sum = cumsum(x);
     SumOfSumsSquared = cumsum(Sum.*Sum);
     SumOfSums = cumsum(Sum);
     SSc = cumsum(index.*Sum);
     
     % now find best fit lines and find fluctuation about the line 
     % loop for each box size, from MinBox to MaxBox
     iteration = 1;
     BoxSize = MinBox;
     TempBoxSize = MinBox;
     while BoxSize <= MaxBox
     	  s = 0.0;
     	  r1 = 1./(BoxSize + 1.0);
     	  Det = 12.0*r1*r1*r1/(1.0 - r1*r1);
     	  vv = BoxSize*(2.0*BoxSize + 1.0)/6.0;
     
     	  for j = 2:numberpoints - BoxSize
     	    dhh = SumOfSumsSquared(j + BoxSize) - SumOfSumsSquared(j-1);
     	    dhu = SumOfSums(j + BoxSize) - SumOfSums(j-1);
     	    dhv = SSc(j + BoxSize) - SSc(j-1) - dhu*j;
     
     	    s = s + dhh - (dhv*dhv +dhu*dhu*vv - dhv*dhu*BoxSize)*Det;
     		j = j + 1;
     	   end
     	
     	  den = (numberpoints - BoxSize)*(BoxSize + 1.0);
     	  Fluctuation = sqrt(s/den);
     	  
           log_points_in_box(iteration,1) = log10(BoxSize);  
     	  log_Q(iteration,1) = log10(Fluctuation);
     	  show_results(iteration,:) = [ iteration  BoxSize  Fluctuation  log_points_in_box(iteration,1)  log_Q(iteration,1) ];
     	  iteration = iteration + 1;
     
     % update the box size 
     	  TempBoxSize = TempBoxSize*LogScaleFactor;
     	  while round(TempBoxSize) <  BoxSize + 1.0
             TempBoxSize = TempBoxSize*LogScaleFactor;
           end
     	  BoxSize = round(TempBoxSize);
     end
     
     % got all boxes; now calculate H via trendline
     r_trend = corrcoef(log_points_in_box, log_Q);
     coefs = polyfit(log_points_in_box, log_Q,1);
     r_line = polyval(coefs,log_points_in_box);
     
     % calculate dimension and Hurst
     H = coefs(1);
     D = 2-H;
     
     %display results in command window
     if out_command_flag == 1
     disp('         iteration    points       Q       Log10(pnts)   Log10(Q)')
     disp(show_results)
     end
     
     % display results in a figure
     if plot_flag == 1
     	figure
     	h = axes('Position', [0 0 1 1], 'Visible', 'off');
     	axes('Position',[.1 .2 .75 .75])
     	plot(log_points_in_box, log_Q,'b+')
     	hold on
     	plot(log_points_in_box,r_line, 'k')
     	ylabel('log10(Q)');
     	xlabel('log10(Points in Subset)');
     	title(file_name);
     	set(gcf, 'CurrentAxes', h);
     	str(1) = {[sprintf('%2.5f',coefs(1)),'x + ', sprintf('%2.5f',coefs(2)), ' = y,   r^2 = ', sprintf('%2.2f',r_trend(1,2)^2), ',  n = ', sprintf('%d',length(log_Q)) ]};
     	text(.1, .02, str, 'FontSize', 10, 'Color', 'r');
     	str(1) = {['H = ', sprintf('%2.3f',H) '  D = ', sprintf('%2.3f',D) ]};
     	text(.5, .02, str, 'FontSize', 10, 'Color', 'r');
     end
     
     % append results to a file rd_results.txt
     if outfile_flag == 1
     	fid = fopen('dfa_results.txt','a');
     	fprintf(fid,'%s  %4.4f   %2.4f\n', file_name, H, r_trend(1,2)^2);
     	status = fclose(fid);
     end
     
     '

.

[lil brain]

This seems to be what what was wrong. Thank you!

[Star Strider]

My pleasure!

.