Brian,

I did some work coevolving cooperative programs for symbolic regression:

"Heterogeneity and Specialization in Evolving Teams"
Proceedings of the Genetic and Evolutonary Computation Conference 2000
pages 778-785

and
"Using Genetic Algorithms to Evolve Cooperative Teams"
Proceedings of the Genetic and Evolutonary Computation Conference 2001
Page 783

If you don't have access to the proceedings I'd be happy to email you the
papers.

Terry Soule

On Fri, 10 Aug 2001, Brian Hargrave wrote:

> Can anyone refer some good papers on the benefits of coevolution, especially
for symbolic
> regression applications?
>
> Thanks,
>
> Brian K. Hargrave, B.S.M.E.
> Graduate Student
> Department of Mechanical and Aerospace Engineering
> Oklahoma State University
> Stillwater,OK
> (405)-744-5900
>
>
________________________________________________________________________________\
_____________________
>
> Get your FREE download of MSN Explorer at http://explorer.msn.com
>
> To unsubscribe from this group, send an email to:
> genetic_programming-unsubscribe@yahoogroups.com
>
>
>
> Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.
>

Hi Terry:

If you don't mind, I'd love to receive a copy of your papers.

Thanks in advance,
John

At 02:01 PM 8/10/2001 -0700, you wrote:

>Brian,
>
>I did some work coevolving cooperative programs for symbolic regression:
>
>"Heterogeneity and Specialization in Evolving Teams"
>Proceedings of the Genetic and Evolutonary Computation Conference 2000
>pages 778-785
>
>and
>"Using Genetic Algorithms to Evolve Cooperative Teams"
>Proceedings of the Genetic and Evolutonary Computation Conference 2001
>Page 783
>
>If you don't have access to the proceedings I'd be happy to email you the
>papers.
>
>Terry Soule
>
>On Fri, 10 Aug 2001, Brian Hargrave wrote:
>
> > Can anyone refer some good papers on the benefits of coevolution,
> especially for symbolic
> > regression applications?
> >
> > Thanks,
> >
> > Brian K. Hargrave, B.S.M.E.
> > Graduate Student
> > Department of Mechanical and Aerospace Engineering
> > Oklahoma State University
> > Stillwater,OK
> > (405)-744-5900
> >
> >
>
________________________________________________________________________________\
_____________________
> >
> > Get your FREE download of MSN Explorer at http://explorer.msn.com
> >
> > To unsubscribe from this group, send an email to:
> > genetic_programming-unsubscribe@yahoogroups.com
> >
> >
> >
> > Your use of Yahoo! Groups is subject to the Yahoo! Terms of Service.
> >
>
>
>
>To unsubscribe from this group, send an email to:
>genetic_programming-unsubscribe@yahoogroups.com
>
>
>
>Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

Sorry about my last spam of the group, when I only meant to reply to Terry.

This brings up a administrivia point -- should the "reply to" address be
the genetic programming mailing list, or the author?  Both options have
their positives, I'm just curious what the group things.

John

On Fri, 10 Aug 2001, John Allred wrote:

> [S]hould the "reply to" address be the genetic programming mailing
> list, or the author?

My two cents:

Personally, I prefer the author, since I'd rather my accidents be
sending public messages to one person, than private info to a whole
list.  But, reply to group seems to be the standard, so in the long run
it's probably best to stick with that.  I've gotten used to it without
too much of a problem.

- Martin

Interesting, we seem to have started an evolutionary tug of war to figure out
which mailing list to use. As I assume everyone is now subscribed to both
mailing lists, the stanford one and the yahoo one (didn't yahoo also start at
stanford?), we can now figure out where the list goes by checking which of
the lists captures most traffic.

As I've been voting for Yahoo, this message goes to the Yahoo list. We do
have to figure out which list to use before the two lists go out of sync. My
best guess is that we have a month of time.

BTW, my little poll elicited up till now two responses about lag and
mis-ordered messages. I am thus not the only one experiencing strange
problems with the majordomo at stanford.

Unfortunately I'm not sure how widespread this is as my question was not the
right one to ask on the stanford mailing list. It kind of reduced to the
question: "Whoever cannot hear me please answer!".

But in any case I've seen I'm not unique in my troubles with the stanford
list, hence this message.

Cheers,

-Maarten-

now that you people have started using this new list my old filter fails to
plonk
gp stuff in my gpstuff mailbox. could you please continue to adopt the
convention
of preceding the subject by GP: ?????

paul coates


Maarten Keijzer wrote:
>
> Interesting, we seem to have started an evolutionary tug of war to figure out
> which mailing list to use. As I assume everyone is now subscribed to both
> mailing lists, the stanford one and the yahoo one (didn't yahoo also start at
> stanford?), we can now figure out where the list goes by checking which of
> the lists captures most traffic.
>
> As I've been voting for Yahoo, this message goes to the Yahoo list. We do
> have to figure out which list to use before the two lists go out of sync. My
> best guess is that we have a month of time.
>
> BTW, my little poll elicited up till now two responses about lag and
> mis-ordered messages. I am thus not the only one experiencing strange
> problems with the majordomo at stanford.
>
> Unfortunately I'm not sure how widespread this is as my question was not the
> right one to ask on the stanford mailing list. It kind of reduced to the
> question: "Whoever cannot hear me please answer!".
>
> But in any case I've seen I'm not unique in my troubles with the stanford
> list, hence this message.
>
> Cheers,
>
> -Maarten-
>
> To unsubscribe from this group, send an email to:
> genetic_programming-unsubscribe@yahoogroups.com
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

--
Paul S. Coates
MSc Tutor
University of East London School of Architecture
Holbrook rd
Stratford
London E15 3EA

To pile meta-mail upon meta-mail,

> Stanford pro:
>  1.  It's the status quo: no effort needed.
>  2.  The "stanford.edu" is prestigious.
>  3.  No adverts.

4. Default "reply to:" address is sender, not list

Someone pointed this out to me and I think it's a valid point. On the Yahoo
list you actually have to make an effort (manual editing of To: line) to
reply only to the author of the message.
Maybe someone knows how to achieve the above behaviour at Yahoo as well?


> Yahoo pro:
>  1.  It's already up and running - no effort needed.
>  2.  Much better services.
>  3.  Less work for the list manager.

2a.  No lag, no misordered posts, [no dropped mails?]

Sending this through the yahoo list to spread quickly, (the post I'm
replying to dropped into my mbox 4 hours after posting).

-Maarten-

ok ok enough already!

I will set up a new filter for myself, as if i didnt have enough to do
already, sorry to
have ruffled so many feathers.


meanwhile we are still evolving Lsystems over here, i'll post some stuff soon.

paul coates.





>now that you people have started using this new list my old filter
>fails to plonk
>gp stuff in my gpstuff mailbox. could you please continue to adopt
>the convention
>of preceding the subject by GP: ?????

Literally, no -- and without ill will. The Yahoo! list manager will
always include "[genetic_programming]" in the headers, typically in
front of all else. May I recommend a second filter?
--
--------------
William Tozier
bill@...

"It is better to have a permanent income than to be fascinating."
    -- Oscar Wilde


William Tozier wrote:
>
>    Part 1.1   Type: Plain Text (text/plain)

--
Paul S. Coates
MSc Tutor
University of East London School of Architecture
Holbrook rd
Stratford
London E15 3EA

>>now that you people have started using this new list my old filter
>>fails to plonk
>>gp stuff in my gpstuff mailbox. could you please continue to adopt
>>the convention
>>of preceding the subject by GP: ?????
>
> Literally, no -- and without ill will. The Yahoo! list manager will
> always include "[genetic_programming]" in the headers, typically in
> front of all else. May I recommend a second filter?

this is configurable.  the list owner can click the Settings button,
then near the upper right-hand corner of the page, there is a
"Subject tag" field.  click edit.

m.

(Just a lurker)


     Why is it that we blithely assume that Yahoo has excellent mailing
services?? While it may be true that, on average, they are better than the
Stanford mailers, I am in several groups of fantasy baseball players, some
of whom use a regular Yahoo E-mail account. Over the past year, there have
been several instances of Yahoo mail being delayed for many hours, or even a
*day* or two.

     So, by all means, consider the switch, but let's assume that mailing
problems will happen occasionally wherever we go.

     I''m of the opinion that the peope managing the list should just make a
choice, and end this Tower of Babble [sic]. The only reason I can see for
keeping the Stanford address is that it is the one that newbies will find
published in several places. Perhaps the Stanford list can reply to new
subscribers with a pointer to the Yahoo list??? Or perhaps this "look on
Yahoo" reply should be sent to *all* Stanford traffic, after a period of
time, if we turn off the Stanford list?


     Probably the worst solution is to continue having two lists ;<).


         Thanks.

                 Chris
--
----- Original Message -----
From: Maarten Keijzer <mak@...>
To: Norman Paterson <norman@...>
Cc: <genetic_programming@yahoogroups.com>
Sent: Monday, August 13, 2001 6:13 AM
Subject: [genetic_programming] Re: GP: GP mailing list fixed


> To pile meta-mail upon meta-mail,
>
> > Stanford pro:
> > 1.  It's the status quo: no effort needed.
> > 2.  The "stanford.edu" is prestigious.
> > 3.  No adverts.
>
> 4. Default "reply to:" address is sender, not list
>
> Someone pointed this out to me and I think it's a valid point. On the
Yahoo
> list you actually have to make an effort (manual editing of To: line) to
> reply only to the author of the message.
> Maybe someone knows how to achieve the above behaviour at Yahoo as well?
>
>
> > Yahoo pro:
> > 1.  It's already up and running - no effort needed.
> > 2.  Much better services.
> > 3.  Less work for the list manager.
>
> 2a.  No lag, no misordered posts, [no dropped mails?]
>
> Sending this through the yahoo list to spread quickly, (the post I'm
> replying to dropped into my mbox 4 hours after posting).
>
> -Maarten-
>
> To unsubscribe from this group, send an email to:
> genetic_programming-unsubscribe@yahoogroups.com
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/
>
>
>

I thought I'd christian genetic_programming@yahoogroups.com with some
controversy: in fact, I'm digging for early EC programming stuff, but got
this one stuck in my craw.

So anyway, I was reading along in the EvoNet Online GP Tutorial, and came
across the interesting statement.  "GP uses programs like individuals, but
the division of opinions appears with its implemnetation. Cramer uses
strings of characters of constant length to generate programs with a
certain structure (Cramer, 1985)."

Think different, EvoNet!  I read this as a continuation of the canard that
Cramer's first paper just used constant strings.  Here's the original
Cramer paper in HTML if you're interested in GP prehistory:
http://www.sover.net/~nichael/nlc-publications/icga85/

In the paper Cramer first proposes lisp s-expressions as the basic
algorithmic language for a "sequential programming language that is
suitable for manipulation by GAs".  He then shows how the s-expressions
can be encoded as arbitrary-length strings of numbers in a preorder
fashion (in the "JB" language).  Then, arguing that this isn't a great way
to go about it, he turns around and proposes manipulating them directly as
trees, using subtree crossover (in the "TB" language).

Along the way, it appears that Cramer racked up quite an impressive number
of firsts, as best as I've been able to tell:

1. Proposed an arbitrary length, string-based stack-oriented GP language,
a precursor to Grammatical Evolution, HiGP, AIMGP, etc.

2. Argued that such an encoding could make crossover precariously close to
randomization (that had always been my concern too).

3. Proposed using arbitrary-length s-expressions as the representational
form of a GP program (the "PL" language), including a memory, progn, and
iterative statements (and goto!).  Used the language minus goto in the
actual representations.

4. Argued for the Turing completeness of the PL language.

5. Argued for the need for GP to be closed under genetic operators.

6. Proposed implementing the PL language using arbitrary-sized GP tree
structures (the "TB" language).

7. Proposed subtree crossover.

8. Proposed "inversion", that is, crossover of two subtrees within the
same individual.

9. Proposed two kinds of tree mutation:

	 A. mutation of terminals into other terminals.

	 B. mutation of nonterminals -- whose children are all terminals --
	 into other nonterminal functions, and adding or deleting child
	 terminals to fix up the arity.

Holy cow!  I know that some earlier items might be attributable to
Smith; does anyone else know of any earlier work which might also be
ascribable as GP, at least as the term is used nowadays?

Sean, on the prowl for early GP systems...

**********************************************************************
			    CALL FOR PAPERS

	    Genetic and Evolutionary Computation Conference
			       GECCO-2002
**********************************************************************

       A recombination of the Seventh Annual Genetic Programming
        Conference (GP-2002) and the International Conference on
		     Genetic Algorithms (ICGA-2002)

  Co-Sponsored by the American Association for Artificial Intelligence

		  July 9-13, 2002 (Tuesday - Saturday)
		      New York City, New York, USA

See the latest in YOUR favorite branch of Evolutionary Computation,
AND explore developments in other, related tracks. The Genetic and
Evolutionary Computation Conference (GECCO-2002) will continue the
tradition of presenting the latest high-quality results in the growing
field of genetic and evolutionary computation.  Topics include, but
are not limited to, genetic algorithms (GA); genetic programming (GP);
evolution strategies (ES); evolutionary programming (EP); evolvable
hardware (EH); evolutionary robotics (ER); real-world applications
(RWA); classifier systems (CS); DNA and molecular computing (DNA);
artificial life, adaptive behavior, agents, and ant colony optimization
(AAAA); optimal design of engineered structures; methodology,
pedagogy, and philosophy (MPP); genetic scheduling and routing (GS);
and other areas to be announced.

Papers are now invited (see further information below), with
manuscripts to be received for review no later than WEDNESDAY January
23, 2002.  Each paper submitted to GECCO will be rigorously reviewed,
in a blind review process, by one of at least seven separate and
independent program committees specializing in various aspects of
genetic and evolutionary computation.  These committees make their own
final decisions on submitted papers for their areas, subject only to
conference-wide space limitations and procedures.

STUDENT TRAVEL GRANTS
Travel grants to assist graduate students in meeting the expenses of
attending GECCO to present a paper will be available. See the web site
for details.

WORKSHOPS
Proposals for Birds-of-a-Feather Workshops for GECCO-2002 are now
being solicited.  In addition to a Graduate Student Workshop, many
other workshops on a variety of EC-related topics will be held during
GECCO-2002, on Tuesday, July 9.  See the web pages
(www.isgec.org/GECCO-2002/workshops) or contact the Workshops Chair,
Alwyn Barry, alwyn.barry@..., for the latest list of topics or
for information on submitting a proposal to organize a workshop.

In addition to the regular workshops, there will be a full-day
Industrial Applications track featuring fielded applications of
genetic and evolutionary computation.

TUTORIALS
Thirty tutorials by recognized experts on diverse areas of GEC are
being planned for presentation on Wednesday, July 10, 2002 (see
listing and updates on www.isgec.org/GECCO-2002).

CONFERENCE CHAIR:  Erick Cantu-Paz (cantupaz@...)
PROCEEDINGS EDITOR-IN-CHIEF:  Bill Langdon
CHAIRS OF CORE PROGRAM POLICY COMMITTEES:
    GA -- Keith Mathias            GP - Riccardo Poli
    ES & EP - Guenter Rudolph      Real-World Apps - L. "Dave" Davis
CHAIRS OF SPECIAL PROGRAM COMMITTEES:
(Additional Special Program Committees are still being organized)
    Alife, Adaptive Behavior, Agents, and Ant Colonies - Vasant Honavar
    Evolutionary Robotics  - Alan Schultz
    Evolutionary Scheduling and Routing - Edmund Burke
    Evolvable Hardware  - Julian Miller
    Industrial Roundtable - L. "Dave" Davis
    Search-Based Software Engineering - Joachim Wegener

BUSINESS COMMITTEE:  David E. Goldberg and John Koza

ADDITIONAL INFORMATION
For information concerning hotel reservations, travel discounts,
student housing, student travel grants, graduate student workshop,
proposals for workshops, proposals for additional tutorials,
late-breaking papers, and other matters, visit
www.isgec.org/GECCO-2002.  For technical matters, email the General
Chair, Erick Cantu-Paz at cantupaz@....  For administrative
matters, email gecco@....  Conference administered by the
American Association for Artificial Intelligence.  Conference operated
by the International Society for Genetic and Evolutionary Computation,
Inc., a not-for-profit corporation.


HOW TO SUBMIT A PAPER TO THE GECCO CONFERENCE

The deadline for ARRIVAL at the physical address below of the eight
(8) paper copies of each submitted paper is WEDNESDAY, JANUARY 23,
2002.  The address is GECCO-2002, c/o AAAI, 445 Burgess Drive, Menlo
Park, CA 94025, USA.  Phone 650-328-3123.

Submitted papers are to be in single-spaced, 10-point type on 8-1/2" x
11" OR A4 paper with 1" margins at top and 3/4" margin at left, right,
and bottom, and are a maximum of eight (8) pages.  Papers may not be
submitted by email or fax.  Details regarding the formatting and
content of the papers are contained at the web site,
www.isgec.org/GECCO-2002.

Review criteria will include significance of the work, novelty,
clarity, writing quality, and sufficiency of information to permit
replication (if applicable).

The first-named author (or other corresponding author designated by
the authors) will be notified of acceptance or rejection on
approximately the first week of March, 2002.  It is preferred (but
not required) that the format of submitted papers roughly follow the
required format for final camera-ready papers, which is posted on the
GECCO web page and is similar to that of GECCO-2001.

FOR THE LATEST INFORMATION, see the web site, www.isgec.org/GECCO-2002

MARK YOUR CALENDARS NOW for an exciting meeting with the latest
high-quality results in Genetic and Evolutionary Computation. New
York, NY, July 9-13, 2002.

--
Erick Cantu-Paz, General Chair, GECCO-2002
Center for Applied Scientific Computing
Lawrence Livermore National Laboratory

It looks like i joined the wrong group.  Seems like all of you are
college grads.
Anyway, im a senior comp sci student and i was looking for a thesis
topic.  I thought a topic in the field of AI would be nice.
Could anyone suggest a topic to me?? or redirect me to a related
discussion group?? tnx.

I read somewhere that GAs could be used to adjust weights in ANNs and
give near optimal solutions to NP complete problems like knapsack and
TSP.  My thesis adviser prefers marketable ideas.  Im squeezing my
brains out for ideas.
   Are there any applications of GAs to WAP?  Data compression?
Network security? Expert systems?
    BTW what are the latest trends in genetic programming?

>I read somewhere that GAs could be used to adjust weights in ANNs and
>give near optimal solutions to NP complete problems like knapsack and
>TSP.  My thesis adviser prefers marketable ideas.  Im squeezing my
>brains out for ideas.
>  Are there any applications of GAs to WAP?  Data compression?
>Network security? Expert systems?

Many of these questions can be answered via keyword searches at:

http://liinwww.ira.uka.de/bibliography/Ai/index.html

(or for genetic programming specifically:

http://liinwww.ira.uka.de/bibliography/Ai/genetic.programming.html)

For example I just tried "compression", "knapsack", and "security" and got
many interesting hits for each.


>   BTW what are the latest trends in genetic programming?

You can probably get a good feel for this by browsing the GECCO proceedings
for the last couple of years (self-plug alert: I'm an editor). The most
recent is:

Spector, L., E. Goodman, A. Wu, W.B. Langdon, H.-M. Voigt, M. Gen, S. Sen,
M. Dorigo, S. Pezeshk, M. Garzon, and E. Burke, editors. 2001. Proceedings
of the Genetic and Evolutionary Computation Conference, GECCO-2001. San
Francisco, CA: Morgan Kaufmann Publishers.
http://www.mkp.com/books_catalog/catalog.asp?ISBN=1-55860-774-9


  -Lee
--
Lee Spector
Associate Professor of Computer Science    lspector@...
Cognitive Science, Hampshire College       http://hampshire.edu/lspector/
Amherst, MA 01002                          413-559-5352, Fax: 413-559-5438

Hi genetic
Yahoo is somehow not allowing me to post to this list.

This is a test..


Have a good day,
Deepak 			    <URL:http://www.glue.umd.edu/~deego>
--
    You know you've spent too much time on the emacs when you spill
    milk and the first thing you think is "C-_"

Hi genetic

I hope  this one gets posted..

Have a good day,
Deepak 			    <URL:http://www.glue.umd.edu/~deego>
--
    You know you've spent too much time on the emacs when you spill
    milk and the first thing you think is "C-_"

Hi!

I have resently resumed the work on my master paper. The paper is
supposed to be a survey of GA and GP, with the mayor focus on GP. I
originaly began my work early in the spring of 99. For a number of
reasons I didn't finnish my paper then.

My qestion now is what has happaned since then?
What ground braking work has been presented?

I've made some seaches on the web but most of the sites(those i knew
from spring 99) out there were either gone or out dated.

Any pointers to good web resources and books/papaers would be much
appreciated.

Christofer Jönsson
Christofer.Jonsson@...

Hi Christofer,
	 I'm not sure that I can answer your question, but I am interested in what
you find.  I'm writing a sort of GP suite that I hope will include many
types of good/standard selection regime, etc.  A skeleton of the program is
working already, so this is one of the major remaining tasks.  So, if you're
interested in this, or you come across cool stuff, please feel free to let
me know!

Best,
Ben


-----Original Message-----
From: Christofer Jönsson [mailto:christofer.jonsson@...]
Sent: Thursday, September 06, 2001 9:27 AM
To: genetic_programming@yahoogroups.com
Subject: [genetic_programming] Master paper on GA/GP question


Hi!

I have resently resumed the work on my master paper. The paper is
supposed to be a survey of GA and GP, with the mayor focus on GP. I
originaly began my work early in the spring of 99. For a number of
reasons I didn't finnish my paper then.

My qestion now is what has happaned since then?
What ground braking work has been presented?

I've made some seaches on the web but most of the sites(those i knew
from spring 99) out there were either gone or out dated.

Any pointers to good web resources and books/papaers would be much
appreciated.

Christofer Jönsson
Christofer.Jonsson@...



To unsubscribe from this group, send an email to:
genetic_programming-unsubscribe@yahoogroups.com



Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

Benjamin Ashpole wrote:
>
> Hi Christofer,
>         I'm not sure that I can answer your question, but I am interested in
what
> you find.  I'm writing a sort of GP suite that I hope will include many
> types of good/standard selection regime, etc.  A skeleton of the program is
> working already, so this is one of the major remaining tasks.  So, if you're
> interested in this, or you come across cool stuff, please feel free to let
> me know!
>

There's such a thing now, (a couple, actually), it's called EO, at
http://eodev.sourceforge.net ; it includes GP along with any selection
regime you might think of, and then some... The reference at the bottom
is also a good one, if you work in Java, and want to do distribution of
tasks among different machines.

J
--
Tutorial Perl => http://granavenida.com/perl
PPSN2002 => http://ppsn2002.ugr.es
Dream => http://dr-ea-m.sourceforge.net

Dear colleagues,
I would like to know if someone have a filter that convert GP lisp output
format ( like (* 4.55 (+ 3.0 y)) ) to a more clear presentation that enable
show the results to an end user that don't have computing knowledge.
After all, show good accuracy isn't enought! We have to show what GP learned
from the inputs and that it is not a non-sense results or magician.
Thank you for all,
James.

_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp

Now that we've (partially) moved to yahoo, how do new members join the
list?


Thanks,
Terry Soule

Dear James

2 years ago I was looking for the same thing.
An american professor helped me with a LISP one-liner
(very elegant expression) that converts expressions
in prefix notation into infix. I am at work now but I
can mail this expression when I get back home.
(You will need a LISP interpretor/compiler, eg. clisp if
you run Linux)

-----Original Message-----
From: James Werner [mailto:jamwer2000@...]
Sent: Friday, September 07, 2001 2:24 PM
To: genetic_programming@yahoogroups.com
Subject: [genetic_programming] GP output filter


Dear colleagues,
I would like to know if someone have a filter that convert GP lisp output
format ( like (* 4.55 (+ 3.0 y)) ) to a more clear presentation that enable
show the results to an end user that don't have computing knowledge.
After all, show good accuracy isn't enought! We have to show what GP learned

from the inputs and that it is not a non-sense results or magician.
Thank you for all,
James.

_________________________________________________________________
Get your FREE download of MSN Explorer at http://explorer.msn.com/intl.asp



To unsubscribe from this group, send an email to:
genetic_programming-unsubscribe@yahoogroups.com



Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

Hi!

I've been having a bit of a headache with my beowulf cluster code - we
currently have several machines of different speeds, so while one machine
might perform 5 generations, another might only perform 3 generations.

Would you generally let the fast machine and the slower machine exchange
genetic material? And if one machine restarted it's evolution, started
from zero, if it still allowed others to supply genetic material, it'd be
like it never restarted within only a handful of generations.

I'm thinking, if they're within 10 generations of each other, I'll allow
the exchange, but that's such a neat solution. Another solution might be
to make the population on the faster machine larger, and on the slower
machine make it smaller, so that they keep about the same pace...

Any ideas?

thanks,
mattias fagerlund

Our plan is to have one computer hold the population, and farm out
evaluations to all computers.  So, the faster computers will simply get
a higher percentage of the population to evaluate.  This has a lot of
nice properties, including that we can repeat runs by simply using the
same seed, assuming evaluation is deterministic.

I think the "only exchange info if within 10 generations" thing is
particularly dangerous; in one run, populations may be completely
separate from almost the start, whereas in another, they'll be sharing
info most of the way through.  That's a very strange dynamic.

- Martin

Mattias Fagerlund wrote:
>
> Hi!
>
> I've been having a bit of a headache with my beowulf cluster code - we
> currently have several machines of different speeds, so while one machine
> might perform 5 generations, another might only perform 3 generations.
>
> Would you generally let the fast machine and the slower machine exchange
> genetic material? And if one machine restarted it's evolution, started
> from zero, if it still allowed others to supply genetic material, it'd be
> like it never restarted within only a handful of generations.
>
> I'm thinking, if they're within 10 generations of each other, I'll allow
> the exchange, but that's such a neat solution. Another solution might be
> to make the population on the faster machine larger, and on the slower
> machine make it smaller, so that they keep about the same pace...
>
> Any ideas?
>
> thanks,
> mattias fagerlund
>
>
> To unsubscribe from this group, send an email to:
> genetic_programming-unsubscribe@yahoogroups.com
>
>
>
> Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

Mattias,

All of the CPUs in our cluster are the same speed so I haven't had to deal
with this, but your idea of scaling population sizes with clock speed seems
like a good approach to me. Restarting a node and then importing
individuals from late in other runs seems like a bad idea (as you seem to
think as well).

  -Lee


>Hi!
>
>I've been having a bit of a headache with my beowulf cluster code - we
>currently have several machines of different speeds, so while one machine
>might perform 5 generations, another might only perform 3 generations.
>
>Would you generally let the fast machine and the slower machine exchange
>genetic material? And if one machine restarted it's evolution, started
>from zero, if it still allowed others to supply genetic material, it'd be
>like it never restarted within only a handful of generations.
>
>I'm thinking, if they're within 10 generations of each other, I'll allow
>the exchange, but that's such a neat solution. Another solution might be
>to make the population on the faster machine larger, and on the slower
>machine make it smaller, so that they keep about the same pace...
>
>Any ideas?
>
>thanks,
>mattias fagerlund
>
>
>
>To unsubscribe from this group, send an email to:
>genetic_programming-unsubscribe@yahoogroups.com
>
>
>
>Your use of Yahoo! Groups is subject to http://docs.yahoo.com/info/terms/

--
Lee Spector
Associate Professor of Computer Science    lspector@...
Cognitive Science, Hampshire College       http://hampshire.edu/lspector/
Amherst, MA 01002                          413-559-5352, Fax: 413-559-5438