How I can use rand('seed', ??) for neural network model?

Hi,

I'm using 'newfftd' which is neural network time delayed model, and I have no idea how I can use rand('seed', ??) for the neural network model? Here is my partial codes of rand('seed').

numDelays = 3;
rand('seed',1);

net = newfftd(P,T,0:numDelays,[20 8],{'logsig' 'logsig' 'purelin'}, 'traincgb', 'learngdm', 'msereg');
net = init(net)

However, I have no idea what the number of '1' means and how it can be scalable?

I found that I got different forecast performance when I changed the number by increasing 1 to 10.

Do I need to find the number while increase until I get a good forecasted result?

Thanks a lot for your reply in advance.

Answer 1

[Fangjun Jiang]

Every time you use rand(), it gives you a random number, that is what rand() is for. But what if you want to repeat that random number, for the purpose of duplicate your simulation results? That is when rand('seed',n) is useful. It gives you a way to generate rand numbers but repeatable. Try the following to see those rand numbers be repeated, also type "doc rand" for more info. rand('seed',n) where n is used to identify or specify a particular series.

clear all;
rand(1,3)
rand('seed',1);rand(1,3)
rand(1,3)
rand('seed',1);rand(1,3)

[Salma Hassan]

what is the different between using n=1 or n=0 , and what about this https://www.mathworks.com/matlabcentral/answers/72733-random-numbers-seed-setting#answer_82791

a=rng;

out=randn

rng(a)

out=randn

Answer 2

[Daniel Shub]

If you are using r2011a, then this link might help

http://blogs.mathworks.com/loren/2011/07/07/simpler-control-of-random-number-generation-in-matlab/

Answer 3

[Greg Heath]

Read the documentation in

 help rand

and

 doc rand

Hope this helps.

Thank you for formally accepting my answer

Greg