Hi,
On Wednesday, April 11, 2012 8:27:26 PM UTC+2, Goldritter wrote:
Another thing about the posted code here.
When the alogrithm is stopped, the start-stop-agent became the state
stopped. Because this agent has not the state running anymore, there
is no send for any other agent anymore.
Ahh ok.
I had read following text on http://clojure.org/agents
Note that use of Agents starts a pool of non-daemon background threads
that will prevent shutdown of the JVM. Use
shutdown-agentshttp://clojure.github.com/clojure/clojure.core-api.html#clojure.core/send-off
to
terminate these threads
Thanks.
I stumbled across these programs too, when I searched for evolution
algorithm in clojure ;).
Might be, that I can use these at work, if the need arise...
Another thing about the posted code here.
When the alogrithm is stopped, the start-stop-agent became the state
stopped. Because
On Apr 11, 2012, at 2:27 PM, Marcus Lindner wrote:
Another thing about the posted code here.
When the alogrithm is stopped, the start-stop-agent became the state
stopped. Because this agent has not the state running anymore, there is
no send for any other agent anymore. But these agents are
Sorry. I meant the code I posted at the beginning. Not your code [1] and
[2].
Sorry for causing a misunderstanding :(.
Am 12.04.2012 01:36, schrieb Lee Spector:
On Apr 11, 2012, at 2:27 PM, Marcus Lindner wrote:
Another thing about the posted code here.
When the alogrithm is stopped, the
Hi,
Am Freitag, 6. April 2012 11:33:25 UTC+2 schrieb Goldritter:
I am a littlebit puzzeled about the chosen structure.
First we have the functions for the algorithm and the agents which
performs the function. There are the creator agents in the domicilems,
which adds new tested
I don't know if this will help with the issues that are really motivating this
thread, but for what it's worth I've written a couple of genetic programming
systems (genetic algorithms in which the genomes are programs that are run as
part of the fitness tests) in Clojure and I generally use
I am a littlebit puzzeled about the chosen structure.
First we have the functions for the algorithm and the agents which performs
the function. There are the creator agents in the domicilems, which adds
new tested individuals into the populatin. Then the grimreaper which
removes individuals from
Maybe the reason is the STM.
If I rmember it correctly, then agents and refs are controlled by
Clojures STM mechanic. Eventuelly the us of map increase the work for
the STM and so it needs more time to schedule all the agent calls and
this result in a higher memory usage.
When you use pmap
Hmmminteresting thoughts...
I was under the impression that that agents are asynchronous and thus
do not depend on the STM , which is only for co-ordinated change! There
is a dedicated Thread pool where agents are assigned separate threads
for their work.You seem to suggest otherwise...I
Might be.
I have spent the last week to write some profiling code for the
algorithm, which also calculates the memory usage.
I wanted to try it out with the original code with different number of
dimiciles and with the pmap code.
But with the STM it is only a guess. I must reread the chapter
A short side note.
I looked into The Joy of Clojure and there is a nice diagram on page
251 about the difference if you use send and send-off to send actions to
an agent.
The intersting on send is, that the agent queues the send actions if you
use send. If you use send-off the agent does not
The intersting on send is, that the agent queues the send actions if
you use send. If you use send-off the agent does not stores the action
into a queue.
Yet the code uses send predominantly...
On 03/04/12 18:39, Marcus Lindner wrote:
Could be, that in the case of pmap a similiar effect
Yep. Your're right.
Am 03.04.2012 20:04, schrieb Jim - FooBar();:
The intersting on send is, that the agent queues the send actions if
you use send. If you use send-off the agent does not stores the
action into a queue.
Yet the code uses send predominantly...
On 03/04/12 18:39, Marcus
Hi,
sorry to be late to the party...
Please let me try to answer some of your questions in one post.
1. The main purpose for the code was to show how many threads could work on
one shared data-structure and how the introduction of concurrent
programming can change the algorithms: in this case
Wow that is indeed very clear and informative...Thanks a lot!
I could not agree more! I was struggling to understand the 'meaning' of
the algorithm and what it was trying to demonstrate...as Stefan said,
one would expect the most expensive bit in any evolutionary computation
to be the
Thanks for the answer.
I wondered too why the GA took so long to solve this small problem.
I tried send for the grimreaper, but this does not solve the population
increase problem. On a test run the population even increses much
faster, wehen I used only send.
But I must say I liked the idea
Is it possible to explain briefly what this genetic algorithm tries to
accomplish? I mean what problem is it producing solution for? It never
seems to end regardless of number-of-domiciles!
I think i got a glimpse of a ThursdayGNEXT string but i'm not sure
what it means...
Jim
On 26/03/12
First there is an error in my code :(
These are the allowed symbols from which the creator function can
choose to create or mutate a gen:
def allowed-symbols (str abcdefghijklmnopqrstuvwxyzäöü
ABCDEFGHIJKLMNOPQRSTUVWXYZÄÖÜ))
This is the targetd which should be fulfilled:
aa ok i see, that makes perfect sense... thanks a lot!
Jim
On 02/04/12 19:06, Goldritter wrote:
First there is an error in my code :(
These are the allowed symbols from which the creator function can
choose to create or mutate a gen:
def allowed-symbols (str abcdefghijklmnopqrstuvwxyzäöü
Shouldn't the line (inside start-evolution) :
*(map #(send % a-day-in-the-life-agent-fn) domiciles)
*
be:
*(pmap #(send % a-day-in-the-life-agent-fn) domiciles)* ???
why does it need to happen serially?
Jim
On 02/04/12 19:06, Goldritter wrote:
First there is an error in my code :(
These
The onle setup that works for my machine (the one that finds the
solution in reasonable time) is with number-of-domiciles = 3, the
missing space and using pmap instead of map when evolution starts...the
population size moves from around 1180 to 1230 but i can see all 4
cores of mine firing up
I think the reason here is, that the used agents perform the day-of-life
function.
Another reason is also, that the individual, which is created by the
agent, is put again into teh ref population. Could be that this might
prefent the use of pmap, or increase the time for the scheduling.
When
Without the pmap and only 2 domiciles I got the solution in 2 minutes on
my 4 core machine.
Might be interesting which effect the pmap might have in combination
with agents.
Am 02.04.2012 20:41, schrieb Jim - FooBar();:
The onle setup that works for my machine (the one that finds the
Ok i understand what you mean...i had a closer look at the code and it
seems that the hard work is done by agents...However, I did some further
testing out of curiosity and got back some weird behaviour...On my
machine it doesn't really matter whether i have 2 or 3 domiciles - i
will still get
Hi,
For my little project I wanted to reprogramm the gentic algorithm from
the book Clojure written by Stefan Kamphausen and Tim Oliver Kaiser.
quickly glancing at the code, it looks like a valid translation into
English.
Did you try to tune number-of-domiciles? Depending on your machine
Thanks for the answer.
Yes, the number of domicles was the problem.
When I use one or two domiciles the population remains relativ stable. If I
use 3 or more, then the population is increasing.
Is it a good idea to add some helpers for the grimreaper if the population
grows too fast?
Marcus
For my little project I wanted to reprogramm the gentic algorithm from
the book Clojure written by Stefan Kamphausen and Tim Oliver Kaiser.
All the basicfunction like creating the population, selecting
individuals, adding individuals, removing individuals and
recombination works fine.
During the
28 matches
Mail list logo
