Thread Subject: Resetting the state of Rand randomly at start up

Hi,
I was using rand to simulate a price series, but rand
function returns the same rand no every time on start up,
hence I am not getting truly random price paths. To overcome
this I was thinking of changing the state of the rand series
generator to some random or non repeating variable, one such
variable could be system date and time but don't know how to
do this.
Can some one help with this.

For reference the manual on rand function reads :-

Quote

The sequence of numbers produced by rand is determined
by the internal state of the generator. Setting the
generator to the same fixed state enables you to repeat
computations. Setting the generator to different states
leads to unique computations. It does not, however, improve
statisticalproperties.
Because MATLAB resets the rand state at startup, rand
generates the same sequence of numbers in each session
unless you change the value of the state input.

UnQuote

"Sourav Roy" <souravroy1@rediffmail.com> wrote in message
<g1088r$lfr$1@fred.mathworks.com>...
> Hi,
> I was using rand to simulate a price series, but rand
> function returns the same rand no every time on start up,
> hence I am not getting truly random price paths. To
overcome
> this I was thinking of changing the state of the rand
series
> generator to some random or non repeating variable, one
such
> variable could be system date and time but don't know how
to
> do this.
> Can some one help with this.
>
> For reference the manual on rand function reads :-
>
> Quote
>
> The sequence of numbers produced by rand is determined
> by the internal state of the generator. Setting the
> generator to the same fixed state enables you to repeat
> computations. Setting the generator to different states
> leads to unique computations. It does not, however,
improve
> statisticalproperties.
> Because MATLAB resets the rand state at startup, rand
> generates the same sequence of numbers in each session
> unless you change the value of the state input.
>
> UnQuote
>
>
>

I think

rand('twister',sum(10*clock))

would randomize the seed (at least for the Mersenne twister
mode) for you.

Ken

I think this is what most people typically do:

rand('state',sum(100.*clock));

Ken

"Kenneth Eaton" <Kenneth.dot.Eaton@cchmc.dot.org> wrote in
message <g109i4$5t1$1@fred.mathworks.com>...
> I think this is what most people typically do:
>
> rand('state',sum(100.*clock));
>
> Ken

Hi,
Thanks I would try this.

"Sourav Roy" <souravroy1@rediffmail.com> wrote in message
<g1088r$lfr$1@fred.mathworks.com>...
> Hi,
> I was using rand to simulate a price series, but rand
> function returns the same rand no every time on start up,
> hence I am not getting truly random price paths. To
overcome
> this I was thinking of changing the state of the rand
series
> generator to some random or non repeating variable, one
such
> variable could be system date and time but don't know how
to
> do this.
> Can some one help with this.
>
> For reference the manual on rand function reads :-
>
> Quote
>
> The sequence of numbers produced by rand is determined
> by the internal state of the generator. Setting the
> generator to the same fixed state enables you to repeat
> computations. Setting the generator to different states
> leads to unique computations. It does not, however,
improve
> statisticalproperties.
> Because MATLAB resets the rand state at startup, rand
> generates the same sequence of numbers in each session
> unless you change the value of the state input.
>
> UnQuote
>
>
>

Here is the best method for avoiding this behavior. Create
a file called startup.m in the directory MATLAB starts up
in. Within startup.m enter the code:

%%%%%
lastState = load('randstate.mat','randnState','randState');
rand('state',lastState.randState)
randn('state',lastState.randnState)
%%%%%

Then, create a "finish.m" file in the same location with
the following code:

%%%%%
randnState = randn('state');
randState = rand('state');
save('randstate.mat','randnState','randState');
%%%%%%

What this does is automatically saves your random state
whenever you close MATLAB then automatically reloads that
state when your restart MATLAB. This will esentially "turn
off" MATLAB's behavior of resetting it's random state at
startup.

There is nothing significantly wrong with the "rand
('twister',sum(10*clock))" method, but the method I mention
is cleaner and more effectively ensures your random state
doesn't get "polluted".

Note: The startup.m file is automatically run by MATLAB on
startup. Also, the finish.m file is automatically run by
MATLAB on quitting. You can put any code in these files
you want to have automatically run.

Ooops. I've kept this code from old versions of MATLAB
(pre-twister). I guess what you really want to do is:

startup.m
%%%%%
lastState = load('randstate.mat','randnState','randState');
rand('twister',lastState.randState)
randn('state',lastState.randnState)
%%%%%

finish.m
%%%%%
randnState = randn('state');
randState = rand('twister');
save('randstate.mat','randnState','randState');
%%%%%%

On May 21, 12:57=A0am, "Sourav Roy" <souravr...@rediffmail.com> wrote:
> "Kenneth Eaton" <Kenneth.dot.Ea...@cchmc.dot.org> wrote in
> message <g109i4$5t...@fred.mathworks.com>...
>
> > I think this is what most people typically do:
>
> > rand('state',sum(100.*clock));
>
> > Ken
>
> Hi,
> Thanks I would try this.

Using 1e5 instead of 100 is better.

http://groups.google.com/group/comp.soft-sys.matlab/msg/a235d956cda15e4a

Hope this helps.

Greg