LISTSERV mailing list manager LISTSERV 16.0

Help for ECJ-INTEREST-L Archives


ECJ-INTEREST-L Archives

ECJ-INTEREST-L Archives


ECJ-INTEREST-L@LISTSERV.GMU.EDU


View:

Message:

[

First

|

Previous

|

Next

|

Last

]

By Topic:

[

First

|

Previous

|

Next

|

Last

]

By Author:

[

First

|

Previous

|

Next

|

Last

]

Font:

Monospaced Font

LISTSERV Archives

LISTSERV Archives

ECJ-INTEREST-L Home

ECJ-INTEREST-L Home

ECJ-INTEREST-L  January 2008

ECJ-INTEREST-L January 2008

Subject:

Re: random seeding

From:

Sean Luke <[log in to unmask]>

Reply-To:

ECJ Evolutionary Computation Toolkit <[log in to unmask]>

Date:

Fri, 18 Jan 2008 14:16:46 -0500

Content-Type:

text/plain

Parts/Attachments:

Parts/Attachments

text/plain (100 lines)

On Jan 17, 2008, at 5:13 AM, Michael Wilson wrote:

> If you create several instances in quick succession you get identical
> RNGs, and System.currentTimeMillis() doesn't have much entropy anyway.

I think this notion betrays a bit of cargo-cult programming, and it's
worth explaining why.


> private long defaultSeed() {
>
> final long xseed = ((iseed*ISEED_MULT)+ISEED_INC)&ISEED_MASK;
> final long nanos = System.nanoTime(); iseed = xseed;
> return (nanos>>>32)^(nanos<<32)^System.currentTimeMillis()^
> xseed^(((long)super.hashCode())<<32)^
> Runtime.getRuntime().freeMemory();
> }
>
> It combines entropy from the current real time, the number of clock
> ticks since last startup, the new object's memory address (via
> super.hashCode()), the amount of free memory in the JVM and the
> output of an embedded static RNG which uses the same algorithm as
> java.util.Random.

It's a cute function overall, but I have a few nits;

- nanoTime is not available on earlier Java versions, *and* it's not
guaranteed to work at all. It's probably not a good idea to use.
- hashCode is the wrong function. It should be
System.identityHashCode(...). It's also not as random as you think:
it's word aligned and sequentially packed on some systems.


> This produces a reasonable amount of entropy
> for experimental purposes (we have a couple of 'high quality' seeding
> methods too but they take longer to run).

I think this may be mistaken. It's true that if you used a crummy
generator with a small seed (such as c's rand()), providing a good
entropy seed would be important. But a crummy generator MT is
not. Generally speaking, providing seed entropy shouldn't have
almost any effect at all. It's waving a dead chicken.

Mersenne Twister has an internal state of 624 longs. If you fill
those longs with a poor-entropy value, or successive poor-entropy
values, after just a few pumps (probably just one pump), you will
*not* be able to notice from the MT output. It will be so random as
to pass every known statistical test. This is one reason why MT is
so well respected. (Eith one exception: if you have a *lot* of 0's
in your internal state, MT takes a while to warm up).

Even if you're still not willing to trust the math, we're not loading
the internal state with low-entropy values. We're loading it with
relatively high-entropy values generated as the stream output of a
carefully-selected linear congruential generator whose seed in turn
is set to the initial 32-bit number you provide. Ordinarily linear
congruential generators aren't fantastic to use as random number
generators for experiments, but they're just fine for providing
enough entropy to MT so that those cargo-cult programmers who don't
trust MT theorists can feel like you've waved the necessary dead
chickens.

This means that you should be able to use seeds like this:

2345, 2346, 2347, ...

and have pretty high-quality randomly different experimental
results. So you're likely not accomplishing anything with your super-
duper seeder above, other than feeling warm and fuzzy. If you have
statistical results that counter this, I (and I think a fair degree
of the RNG community) would really like to see them published! (no
really! I'm not trying to be sarcastic. What do you have?)

Now as to why we have the defaultSeed(), setSeed(), and 32-bit seed
choices: because they are the standard for MT19937. See here:
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/MT2002/CODES/
mt19937ar.c
http://www.math.sci.hiroshima-u.ac.jp/~m-mat/MT/MT2002/emt19937ar.html

We chose to be 100% consistent with the MT standard rather than risk
wandering off the ranch in search of a seed twice as long (which only
benefits you if you need to do more than 2^32 experiments). If you'd
like something else, you should just go in and set all 624 initial
values in the MT array: there's a function for that. Otherwise,
there are *big* benefits to having a standard RNG.

32 bits, rather than a longg is not a big deal either. After all,
java.util.Random uses a only 48-bit seed rather than a long (and in
fact that's its entire internal state! java.util.Random is very poor).

Now: how to initialize RNGs in sequence? Assuming you let MT's
default seeder do its job, the answer then doesn't lie in seeding
entropy, but in simply guaranteeing that you have a different seed
each time. Just grab System.currentTimeMillis(), use it for your
first seed, then use the same value + C for your next seed, then the
next value + 2C for your next seed, then +3C, etc., where C is some
integer > 0. Just maintain that internal counter somewhere and you
should be fine.

Sean

Top of Message | Previous Page | Permalink

Advanced Options


Options

Log In

Log In

Get Password

Get Password


Search Archives

Search Archives


Subscribe or Unsubscribe

Subscribe or Unsubscribe


Archives

September 2019
August 2019
June 2019
April 2019
March 2019
January 2019
December 2018
November 2018
October 2018
July 2018
May 2018
January 2018
December 2017
November 2017
October 2017
September 2017
August 2017
July 2017
June 2017
May 2017
April 2017
March 2017
February 2017
January 2017
December 2016
November 2016
October 2016
September 2016
August 2016
July 2016
June 2016
May 2016
April 2016
March 2016
February 2016
January 2016
December 2015
November 2015
October 2015
September 2015
August 2015
July 2015
June 2015
May 2015
April 2015
March 2015
February 2015
January 2015
December 2014
November 2014
September 2014
August 2014
July 2014
June 2014
May 2014
April 2014
March 2014
February 2014
January 2014
December 2013
November 2013
October 2013
September 2013
August 2013
July 2013
June 2013
May 2013
April 2013
March 2013
February 2013
January 2013
December 2012
November 2012
October 2012
September 2012
July 2012
June 2012
May 2012
April 2012
March 2012
February 2012
January 2012
December 2011
November 2011
October 2011
September 2011
August 2011
July 2011
June 2011
May 2011
April 2011
March 2011
February 2011
January 2011
December 2010
November 2010
October 2010
September 2010
August 2010
July 2010
June 2010
May 2010
April 2010
March 2010
February 2010
January 2010
December 2009
November 2009
October 2009
September 2009
August 2009
July 2009
June 2009
May 2009
April 2009
March 2009
February 2009
January 2009
December 2008
November 2008
October 2008
September 2008
August 2008
July 2008
June 2008
May 2008
April 2008
March 2008
February 2008
January 2008
December 2007
November 2007
October 2007
September 2007
August 2007
July 2007
June 2007
May 2007
April 2007
March 2007
February 2007
January 2007
December 2006
November 2006
October 2006
September 2006
August 2006
July 2006
June 2006
May 2006
April 2006
March 2006
February 2006
January 2006
December 2005
November 2005
October 2005
September 2005
August 2005
July 2005
June 2005
May 2005
April 2005
March 2005
February 2005
January 2005
December 2004
November 2004
September 2004
August 2004
July 2004
June 2004
May 2004
April 2004
March 2004

ATOM RSS1 RSS2



LISTSERV.GMU.EDU

CataList Email List Search Powered by the LISTSERV Email List Manager