MATLAB: Vectorize a for loop

optimize code for-loopvectorization

I am finding a optimal way of vectorizing the following piece of code (of course the real data is much greater than this):
start_peak_hour = [3 10 20 50];
end_peak_hour   = [5 14 27 70]; % always same size as start_peak_hour
n_peak = numel(start_peak_hour);
isPeak = zeros(1,80); 
for j=1:n_peak,
    isPeak(start_peak_hour(j):end_peak_hour(j)) = 1;
end
Your two cents are appreciated
-Sam

Best Answer

I highly doubt you'll find anything faster than that well-written for-loop. The JIT will pick that up and have wonders with it.
eval is evil and will be orders of magnitude slower and hideous/hard to follow.
start_peak_hour = [3 10 20 50];
end_peak_hour   = [5 14 27 70]; % always same size as start_peak_hour
n_peak = numel(start_peak_hour);
t1 = 0;
t2 = 0;
for tt = 1:50 %50x
tic
isPeak = zeros(1,80); 
for j=1:n_peak,
  isPeak(start_peak_hour(j):end_peak_hour(j)) = 1;
end
t1 = t1+toc;
tic
eval(sprintf('isPeak(start_peak_hour(%d):end_peak_hour(%d)) = 1,',[1:n_peak;1:n_peak]))
t2=t2+toc;
end
[t1 t2] % 0.0010    0.0681
t1/t2
On my system the FOR-loop is 70x faster (for this small problem), it'll get better with bigger ones. And it doesn't use eval or any other impossible to understand syntax.
My $0.05 - a full nickel :) .
Related Question