Improve performance on finding max values of matrix

Hi everyone,

I have a performance issue with finding the maximum values of a matrix:

I have a matrix of 3D coordinates of points in the form

data=[X1 Y1 Z1,  X2 Y2 Z2, ..., Xn Yn Zn] 

With n over about 4.10^6

The values of X and Y are a grid, so basically I have about 3000 times the same Y values (Y1 to Y3000 are equal, Y3001-Y6000 are equal…). I want to find for each Y coordinate the maximum X coordinate. What I currently have is:

while i<length(data)           %find for each Y Xmax
    xmax(j)=max(data(data(:,2)==data(i,2),1));
    ymax(j)= data(i,2);
    i=find(data(:,2)== data(i+1,2), 1, 'last');
    j=j+1;
end

This takes about 3 min to execute. Is there a better way to do this?

Answer 1

[Daniel M]

clearvars;

% create fake data
[y,x] = meshgrid(1:5);
data = [rand(size(x(:))).*x(:),y(:)].*2; % different values of x (1st col), for grouped y (2nd col)

% get unique values of y
[ymax,~,locMembersY] = unique(data(:,2));

% get max x for each y
xmax = splitapply(@(v) max(v), data(:,1), locMembersY);

[Jsqn]

Works great. I went from 180s with my code to 0.7566s...

Thank you very much Daniel

[Daniel M]

My pleasure, what a speedup!