seeding random numbers

[(-Peter-)]

Hi..

I'm programming a Monte Carlo problem where i need to show that my
results are reproducible - that is I need to prove this by running by
program with seeded random numbers.

The program is build up of a main method and a lot of other sub-
routines. How can I create a random number generator, from which I can
get my seeded random numbers. It shall be possible to get one random
number at a time a lot times (some millions), and I need to call the
random number generator from any of the subroutines and get the same
sequence of numbers no matter from which routines i call them (also
shifting between the sub-routines for example 5 calls from routine 1
followed by 5 from the main routine should give the same numbers as if
I called the generator 4 times from the main routine followed by 5
times from routine number 10 followed by one call from the main
routine again)

I hope the description can be understood - it is a bit difficult to
explain..

Can anybody help me to solve this?

/Peter

 

[Eric Sosman]

Hi..

I'm programming a Monte Carlo problem where i need to show that my
results are reproducible - that is I need to prove this by running by
program with seeded random numbers.

The program is build up of a main method and a lot of other sub-
routines. How can I create a random number generator, from which I can
get my seeded random numbers. It shall be possible to get one random
number at a time a lot times (some millions), and I need to call the
random number generator from any of the subroutines and get the same
sequence of numbers no matter from which routines i call them (also
shifting between the sub-routines for example 5 calls from routine 1
followed by 5 from the main routine should give the same numbers as if
I called the generator 4 times from the main routine followed by 5
times from routine number 10 followed by one call from the main
routine again)

Create as many different java.util.Random objects as you
need, using the same seed value for each:

long seed = 123456789; // or whatever
Random rmain = new Random(seed);
Random rsub1 = new Random(seed);
...

Devote one Random object to each "subroutine" or group of
related subroutines, so each "subroutine" obtains values only
from its own Random object.

... but I may have misunderstood just what you mean by "the
same sequence." If it seems I have, please try to explain it
more clearly.

 

[(-Peter-)]

Create as many different java.util.Random objects as you
need, using the same seed value for each:

long seed = 123456789; // or whatever
Random rmain = new Random(seed);
Random rsub1 = new Random(seed);
...

Devote one Random object to each "subroutine" or group of
related subroutines, so each "subroutine" obtains values only
from its own Random object.

... but I may have misunderstood just what you mean by "the
same sequence." If it seems I have, please try to explain it
more clearly.

I think you have misunderstood it (because of the bad explanation)..

Generally I need ONE sequence (therefore I think I do only need one
generator) of random numbers for example (which are seeded, so I get
the same results every time I run the code)

0.21 0.23 0.1 0.9 0.5 0.7

I will when use the first number (in this example 0.21) to decide
which sub-routine should be run. This sub-routine (call this sub1)
uses the next random number (0.23) and when "decides" to go to another
sub-routine (sub2) which uses the random number (0.1)...

If I seeded the random number generator another way I would for
example get these numbers:

0.7 0.1 0.21 0.81 0.21 0.001

The first number (0.7) is now calling another subroutine(sub4) which
uses the number 0.1 to decide to go back to the main method which uses
the random number (0.21). This calls sub2 which uses 0.81....

That is no matter what the numbers are I will use them in them in the
same order - which numbers are used in which method is "random".

I hope this was a better explanation.. If not please tell me..

/Peter

 

[Eric Sosman]

Generally I need ONE sequence (therefore I think I do only need one
generator) of random numbers for example (which are seeded, so I get
the same results every time I run the code)

0.21 0.23 0.1 0.9 0.5 0.7

I will when use the first number (in this example 0.21) to decide
which sub-routine should be run. This sub-routine (call this sub1)
uses the next random number (0.23) and when "decides" to go to another
sub-routine (sub2) which uses the random number (0.1)...

If I seeded the random number generator another way I would for
example get these numbers:

0.7 0.1 0.21 0.81 0.21 0.001

The first number (0.7) is now calling another subroutine(sub4) which
uses the number 0.1 to decide to go back to the main method which uses
the random number (0.21). This calls sub2 which uses 0.81....

That is no matter what the numbers are I will use them in them in the
same order - which numbers are used in which method is "random".

I'm sorry, but I still don't understand your difficulty.
If you create two Random objects using the same seed value,
and if you make the same sequence of method calls on each,
they will deliver identical result streams. So if you run
your program using a Random whose seed is 123456789, and then
you run the program it a second time and again use the seed
value 123456789, you will get the same pseudo-random values
in both executions.

 

[Christian]

I think you have misunderstood it (because of the bad explanation)..

Generally I need ONE sequence (therefore I think I do only need one
generator) of random numbers for example (which are seeded, so I get
the same results every time I run the code)

0.21 0.23 0.1 0.9 0.5 0.7

I will when use the first number (in this example 0.21) to decide
which sub-routine should be run. This sub-routine (call this sub1)
uses the next random number (0.23) and when "decides" to go to another
sub-routine (sub2) which uses the random number (0.1)...

If I seeded the random number generator another way I would for
example get these numbers:

0.7 0.1 0.21 0.81 0.21 0.001

The first number (0.7) is now calling another subroutine(sub4) which
uses the number 0.1 to decide to go back to the main method which uses
the random number (0.21). This calls sub2 which uses 0.81....

That is no matter what the numbers are I will use them in them in the
same order - which numbers are used in which method is "random".

I hope this was a better explanation.. If not please tell me..

/Peter

It seems to me that it would fit for you to create a single Random
object. Make it accessible to everywhere from the program.
i.e. using static or create a singleton to hold it.

Christian

 

[Patricia Shanahan]

(-Peter-) wrote:
....

Generally I need ONE sequence (therefore I think I do only need one
generator) of random numbers for example (which are seeded, so I get
the same results every time I run the code)

In that case, construct a single instance of java.util.Random,
specifying the seed on the constructor call, and use it throughout the
program.

Patricia

 

[Jason Cavett]

(-Peter-) wrote:

...


In that case, construct a single instance of java.util.Random,
specifying the seed on the constructor call, and use it throughout the
program.

Patricia

Patricia is right - a single instance across the program is the way to
go. HOWEVER...

The only downside to using java.util.Random is that it tends not to be
"random" enough. So, if you're interested in a "more random" answer,
check into the Mersenne Twister. (There's a good article on
Wikipedia.)

 

[Roedy Green]

I'm programming a Monte Carlo problem where i need to show that my
results are reproducible - that is I need to prove this by running by
program with seeded random numbers.

see http://mindprod.com/jgloss/pseudorandom.html

 

[Roedy Green]

In that case, construct a single instance of java.util.Random,
specifying the seed on the constructor call, and use it throughout the
program.

If you have threads this won't work. You would need one generator per
thread to get reproducible results. One generator will suffice so
long as you make calls on it in the same order each time for each
sample you want on a single thread.

 

[(-Peter-)]

Patricia is right - a single instance across the program is the way to
go.  HOWEVER...

The only downside to using java.util.Random is that it tends not to be
"random" enough.  So, if you're interested in a "more random" answer,
check into the Mersenne Twister.  (There's a good article on
Wikipedia.)

Okay I will look at that..

How do I create this object?

How would the object know which of the numbers it shall use?

I can not get a idea to construct it..

please help med

/Peter

 

[(-Peter-)]

seehttp://mindprod.com/jgloss/pseudorandom.html

thank you for the link - it looks interesting..

/peter

 

[(-Peter-)]

Okay I will look at that..

How do I create this object?

How would the object know which of the numbers it shall use?

I can not get a idea to construct it..

please help med

/Peter- Skjul tekst i anførselstegn -

- Vis tekst i anførselstegn -

I forgot one thing:

I got one idea: Creating some million numbers and saving it in a
array.. but I think that would be waste of memory!

/Peter

 

[Patricia Shanahan]

Okay I will look at that..

How do I create this object?

import java.util.Random;

....

Random randomNumberSource = new Random(someSeed);

[On the right hand side, you can substitute the constructor for any
class that extends java.util.Random, if you prefer a better quality
generator.]

How would the object know which of the numbers it shall use?

See the java.util.Random API documentation.

I can not get a idea to construct it..

Just like any other Java object. I'm not sure I'm seeing the difficulty
here.

Patricia

 

[(-Peter-)]

Okay I will look at that..

How do I create this object?

import java.util.Random;

...

Random randomNumberSource = new Random(someSeed);

[On the right hand side, you can substitute the constructor for any
class that extends java.util.Random, if you prefer a better quality
generator.]

How would the object know which of the numbers it shall use?

See the java.util.Random API documentation.

I can not get a idea to construct it..

Just like any other Java object. I'm not sure I'm seeing the difficulty
here.

Patricia- Skjul tekst i anførselstegn -

- Vis tekst i anførselstegn -

I'll try to work on it again..

thank you for your help.

 

[Jason Cavett]

I forgot one thing:

I got one idea: Creating some million numbers and saving it in a
array.. but I think that would be waste of memory!

/Peter

That would definitely not be a good solution to this problem.
(Besides, in a Monte Carlo simulation, it is quite possible that you
could need more than a million numbers. What do you do then?)

As Patricia said - making your class static would be the way to go.
Of course, Roedy correctly points out that if you have a multi-
threaded application, this solution is not guaranteed to work unless
you can guarantee the order that the threads are grabbing the random
numbers.

 

[Jason Cavett]

Okay I will look at that..

How do I create this object?

How would the object know which of the numbers it shall use?

I can not get a idea to construct it..

please help med

/Peter

You create the object like any normal Java object.

Blah var = new Blah();

You need to use the Mersenne Twister code (there is some found at
Wikipedia) along with your own code.

 

[Roedy Green]

why did the space after "see" disappear? I see this happening often.
My original has an ordinary hex 20 space.

 

[Roedy Green]

The only downside to using java.util.Random is that it tends not to be
"random" enough. So, if you're interested in a "more random" answer,
check into the Mersenne Twister. (There's a good article on
Wikipedia.)

there is also SecureRandom. See
http://mindprod.com/jgloss/pseudorandom.html for links to both.

 

[John W. Kennedy]

why did the space after "see" disappear? I see this happening often.
My original has an ordinary hex 20 space.

The editor for Google Groups is as buggy as OS/360 Release 1.

--
John W. Kennedy
"The whole modern world has divided itself into Conservatives and
Progressives. The business of Progressives is to go on making mistakes.
The business of the Conservatives is to prevent the mistakes from being
corrected."
-- G. K. Chesterton