Create a matrix that contains a specific total value, made by the maximum values, of another matrix

So, I have an NxM matrix of doubles that is asymmetrically arranged. What I'd like to do is create a second matrix that contains a specific percentage of the total of the first matrix at indexes that match the location of the points in the first matrix, and have the second matrix be zero everywhere else.

I have written code that can do this, but its painfully slow. I'm hoping someone might have some ideas on how to make this more efficient. The code I have written is below:

temp_array=original_array;                                    % copy original array
ratio=0.6;                                                              % desired percentage total of original array
len=length(original_array); hgt=height(original_array);                 % Create empty array
dummy=zeros(len, hgt);
total=ratio*sum(sum(original_array));
running_tot=0;
while running_tot<total
    temp_val = max(max(temp_array));                  % Find max value of input
    [y_val, x_val]=find(temp_array==temp_val);     % Pull indicies
    dummy(y_val, x_val)=temp_val;                        % Put max value in dummy array
    temp_array(y_val, x_val)=0;                             % Delete max value
    running_tot=running_tot+temp_val;
end

This preserves the original array and contains the relevant data in the array titled "dummy." But it's egregiously slow - I'd like to make it faster. I've considered reversing the way the loop works and deleting the minimum values rather than keeping the maximum values, but there's no way to tell if the min(min(temp_matrix)>0, which would get the loop stuck.

Any thoughts on how I might go about increasing the efficency of this code? Maybe a matrix-based method that I haven't thought of?

Thanks!

Answer 1

[Matt J]

temp_array=original_array;                                    % copy original array
ratio=0.6;                                                              % desired percentage total of original array
total=ratio*sum(original_array(:));

u=unique(original_array);

idx=find(cumsum(u,'reverse')<total,1);
thresh=u(idx);

temp_array(temp_array>=thresh)=0;

dummy=original_array-temp_array;

[Matt]

That works perfectly! It took me a little while to work out

idx=find(cumsum(u,'reverse')<total,1);
thresh=u(idx);

but I think I've got it figured out. It probably would have taken me quite a while to work out using cumsum. Thanks for your help!