replace elements in columns of a matrix

Hi, I have columns block (which is a unique function output) A and B as below. I want to replace one element from column B in A each time (as in out) so that I ll do calculations on each column in out. Is there any way to do that without running out of memory possibly avoiding loops(A and B has millions of rows).The attachment demonstrates replacement up to block 2 but this continues all the way

Answer 1

[Thorsten]

This is the best I can think of

A = rand(1e5, 1);
B = rand(1e5, 1);
res = nan(size(A));

for i = 1:numel(A)
  temp = A(i);
  A(i) = B(i);
  res(i) = mean(A); % or whatever you want to compute
  A(i) = temp; 
end

[Image Analyst]

He says he needs out as a 2D array for some reason - that's why I said "No." He said " I ll do calculations on each column in out" so evidently he wants multiple columns in out. I suggested he not do that and come up with a more clever way, like perhaps just using and computing one column of out at a time.

[Tiina]

thats right, thanks. I guess the only way is to adjust Thorstens code to run by block!? store calculations per block then concatenate. Could you advise on efficient ways?

[Image Analyst]

If you're going to concatenate block by block (column by column) to get your 2-D array, then you won't have enough memory, like I explained in my answer. You're going to have to figure out a way to do what you need to do without having the 2D array all in memory at the same time. Why do you think the whole thing needs to be there? Why can't you deal with it column by column like Thorsten showed, and I also recommend? Just have a for loop over columns.

Answer 2

[Image Analyst]

I am pretty sure the answer is no. The problem is that the nth row of B goes into the nth column of out. Like in your example, the 5th row of B goes into the 5th column of "out". So with millions of rows, you will have millions of rows (of course) but you will have millions of columns. That is just far, far too big - you will run out of memory. So with, say, 3 million rows, you'll have a 9*10^12 element array at 8 bytes per element - that's 72,000 gigabytes GB of RAM. I have 32 GB of RAM (more than most people) and just after I fire up MATLAB it tells me the biggest array I can have is 6 GB. Six is a long way from 72 thousand.

Why do you think you want this "out" all in memory at one time? Can't you just do your process one row at a time without computing "out" in advance? I suggest you figure out how to utilize "out" in a more clever way so that you don't have to have it all in memory at one time.