Rolling max/min maximum and minimum

This may have already been asked and aswered, but I couldn't find anything on Matlab Answers or Google. Is there a slick vectorized way to calculate a rolling maximum/minimum for a vector. So the maximum over the previous n values (inclusive of the current value). For indices i < n the value should either be the max of the values with indices 1:i or NaN, I can live with either. So for example:

x = [1, 3, 2, 7, 4, 3]; n = 2;

should give

rollingMax = [NaN, 3, 3, 7, 7, 4] or [1, 3, 3, 7, 7, 4]

Answer 1

[Matt J]

If you have the Image Processing Toolbox, you can use IMDILATE

    >> imdilate( [1, 3, 2, 7, 4, 3] ,[1,1])

    ans =

         1     3     3     7     7     4

[Chris]

Don't have Imagine Processing Toobox unfortunately....

Answer 2

[Matt J]

There are also FEX contributions, like

http://www.mathworks.com/matlabcentral/fileexchange/24705-minmax-filter

[Arthur]

This one doesn't need mex: http://www.mathworks.com/matlabcentral/fileexchange/12550

[Matt J]

Yes, but it's accordingly not very fast :-(

Answer 3

[José-Luis]

Without a toolbox:

 x = randi(10,1,100);
 n = 2;
 maxVec = arrayfun(@(a,b) max(x(a:b)),1:numel(x)-n+1,n:numel(x));

[Matt J]

That's not really vectorized, since we know arrayfun underperforms speed-wise as compared to for-loops.