Calculate mean of a matrix with different indices

Hi,

I have a loop as below.

RCAr = []
for k = 1:length(RCA);  
...
...
afstand = thisData(1:a,1);
lengte = size(dx);
lengte = round(lengte);
for i = 1:lengte-1
    P1 = [dx(i), dy(i), dz(i)];    
    P2 = [dx(i+1), dy(i+1), dz(i+1)];
    angleradian(i) = acos((P2(2)-P1(2)) / norm(P2-P1));
    distance(i) = norm(P1 - P2);
     
end
angleradian2(:,1) = angleradian;
distance2(:,1) = distance;
    RCAr(:,k) = angleradian2; %this point is not working
    plot(afstand,angleradian2, '-')
    hold on

end

I want to make a matrix of k columns with all the angleradians. But the indices of angleradian does not have the same length. How can I fix this?

I want to calculate the mean of each row and stdev of each row of that matrix.

Answer 1

[Bjorn Gustavsson]

So you effectively have different number of dx for each k? If so I'd do this something like this:

angleradian2 = nan([MaxNrRows,MaxNrCols]); % pre-allocate a nan-array with correct size
RCAr = angleradia2; % just make place for this one too
distance = RCAr;    % and this one,
afstand = RCAr;     % and this one...
for i1 = 1:length(RCA);  % Changed k to i1 to get identifying label on loop-variable
  ...
  ...
  afstand(i1,1:a) = thisData(1:a,1);
  lengte = size(dx,1); % or 2 instead of 1, size returns a 2 (at least) element array with
                       % [sz1, sz2, ... szN] since you use this as an upper
                       % bound of your iteration it should be a scalar!
  lengte = round(lengte);
  for i2 = 1:lengte-1  % changed i to i2 for same reason as above AND keep i free for complex unit
    P1 = [dx(i2), dy(i2), dz(i2)];    
    P2 = [dx(i2+1), dy(i2+1), dz(i2+1)];
    angleradian2(i1,i2) = acos((P2(2)-P1(2)) / norm(P2-P1));
    distance(i1,i2) = norm(P1 - P2);
    
  end
  RCAr(i1,:) = angleradian2(i1,:); %this point is not working
  plot(afstand,angleradian2, '-')
  hold on
  
end

HTH

[Dion Theunissen]

Thanks, I only need to know the maxnrows and columns before looping. Is this also possible during the loop? Or do I have to check this before and add this in this row.

[Bjorn Gustavsson]

Preferably you should figure out how big your variables need to be beforehand - that reduces the number of times matlab needs to reallocate space for the arrays (this is the terrible time-waster in many "lazy" algorithms that relies on the automatic array-growing capabilities of matlab (it costs to be lazy (unfortunately...))). The general idea to use nan for your data is that you can use the 'omitnan' flag to the statistics functions afterwards to only take the average of the numerical values in the arrays, if you grow stuff automagically there will be zero-padding done implicitly.

Another way around this problem is to use another array, lets say nrSamples, to store how many data-points you get for each i2.

Another way of dealing with this is to store the results for each i2 in cell-arrays and then calculate the statistics cell-by-cell.

[Dion Theunissen]

Allright, thanks for your help! It works for me.

[Bjorn Gustavsson]

Great!