Array indexing for moving IQR

I'm calculating a moving iqr, similar to the movmean, movmad, and movstd functions, discarding the endpoints:

 A = [4 8 6 -1 -2 -3 -1 3 4 5];
 for i = 1:length(A)-k+1
    M(i) = iqr(A(i:i+k-1)
 end

 M = [3.0000    6.7500    6.0000    1.5000    1.5000    4.5000    3.7500    1.5000]

M is as expected. How can I do this more efficiently using array indexing?

Answer 1

[Jos (10584)]

When N (= number of elements in A) and K (= number of elements to use) are not too large you could create an intermediate (N-K+1)-by-K array. This could be (slightly) more efficient than a for-loop but also looks more cryptic ...

A = [4 8 6 -1 -2 -3 -1 3 4 5]
K = 3 

N = numel(A)-K+1 
I = cumsum([1:N ; ones(K-1,N)], 1)  
M = iqr(A(I), 1)

[Anna McCann]

Okay thanks, I'll try profiling it and see if it speeds it up.

Answer 2

[Walter Roberson]

R2016b and later,

iqr(A((1:k).'+(0:length(A)-k)))