Averaging Together Every N Columns

Hi!

I have a 16:10000 matrix (positive and negative decimal values) and want to reshape it to a 16:10 matrix by averaging each row every 1000 columns. What's the best way to do this?

Thanks very much! SE

Answer 1

[the cyclist]

This solution using the filter() command calculates interim results way beyond your need, but it's fast:

% Put in some pretend data
r = rand(16,10000);

% Define number of columns to average
AVG_COLS = 1000;

% Dimension over which to average
DIM = 2; % Columns

% Use filter to calculate the moving average across EVERY combination of columns
r_moving_avg = filter(ones(1,AVG_COLS)/AVG_COLS,1,r,[],DIM);

% Grab only the column averages that were actually wanted
r_avg = r_moving_avg(:,AVG_COLS:AVG_COLS:end)

Answer 2

[Jan]

r = rand(16, 10000);
s = reshape(r, 16, 1000, 10);
t = squeeze(mean(s, 2));

Some timings:

r = rand(16, 10000);
tic;
for k = 1:20
   AVG_COLS = 1000;
   DIM = 2; % Columns
   r_moving_avg = filter(ones(1,AVG_COLS)/AVG_COLS,1,r,[],DIM);
   r_avg = r_moving_avg(:,AVG_COLS:AVG_COLS:end);
end
toc

tic;
for k = 1:20
   N = 1000 ;
   szA = size(r) ;
   B = arrayfun(@(k) mean(r(:,k:min(szA(2),k+N-1)),2), 1:N:szA(2), 'un', 0) ;
   B = [B{:}] ;
end
toc

tic;
for k = 1:20
   s = reshape(r, 16, 1000, 10);
   t = squeeze(mean(s, 2));  end
toc

 Elapsed time is 1.671149 seconds.
 Elapsed time is 0.035495 seconds.
 Elapsed time is 0.016600 seconds.

Answer 3

[Sarah]

Thanks so much - works like a charm!

Answer 4

[Jos (10584)]

Another option:

A = rand(16,10000) ;
N = 1000 ;
szA = size(A) ;
B = arrayfun(@(k) mean(A(:,k:min(szA(2),k+N-1)),2), 1:N:szA(2), 'un', 0) ;
B = [B{:}] ;