normrnd vs.random

Dear, I would like to generate random nmbers with a normal distribution. From the documentation I see that there are two functioncs that seem to make the same things. normrnd(mu,sigma) and random('Normal',mu,sigma)

I wrote these two simpole for-loop in order to evaluate time consumption of these two functions and I found that normrnd is less time-consuming than random. I don't understand this difference. Could you help me?

tic for i=1:1000000 ciao=ciao+normrnd(0,0.01); end toc

tic for i=1:1000000 ciao=ciao+random('Normal',0,0.01); end toc

Kind regards

Andrea Giostri

Answer 1

[Titus Edelhofer]

Hi,

if you open random with an editor and search for "normrnd", you will see, that random in fact calls normrnd. That's why it takes a little more time. On the other hand: if you call normrnd and random one time with N=1000000 instead of N calls in a loop, it won't make a difference at all ...

Titus

[Andrew Newell]

There is also randn, which is slightly faster than either of the alternatives:

N = 10000;

mu = 0;

sigma = 0.01;

clear ciao

tic

ciao = sigma*sum(randn(N));

toc

clear ciao

tic

ciao = sum(normrnd(mu,sigma,N));

toc

clear ciao

tic

ciao = sum(random('Normal',mu,sigma,N));

toc

Elapsed time is 2.800175 seconds.

Elapsed time is 3.497521 seconds.

Elapsed time is 3.473260 seconds.

However, when I tried to increase N to 100000, my computer ran out of memory (and it has 12 GB).

[Titus Edelhofer]

Hi Andrew,

common mistake I often do as well: randn(N) is not randn(N,1) but randn(N,N). I'm not surprised this will blow your memory with N=100000 ;-).

Titus

[Andrew Newell]

Aaagh! Right at the top of http://www.mathworks.com/matlabcentral/answers/1759-dumb-mistakes-we-make-with-matlab! Thanks for reminding me.

[dany katamba mpoyi]

Normal random numbers - MATLAB normrnd - MathWorks Italia