gplab-users

Re: [gplab-users] Ant example in GPLab

[Sara S. <sa...@de...>]

Aurélien Boffy wrote:

> ...
> So, it is us who thank you for your wonderful library and long live GPLab!

Well, thank you!
I am glad GPLAB has been useful to you, not just a source of problems. :)

> Concerning the problem with the "eval" function:
> 
> Someone has already replied to your message in the MATLAB newsgroup and has
> suggested to pass strings for the function names.
> We had already tried this idea and that works.
> ...

Yes, it works for the artificial ant problem, but it ruins the 
evaluation of simple symbolic regression problems, or boolean problems, 
or any other problems that use MATLAB's built-in functions.

For example:
s1='plus(plus(2,2),times(3,1))'
s2='plus(''plus(''2'',''2'')'',''times(''3'',''1'')'')'

s2 is how s1 would have to be coded if using strings for the function 
names. While s1 evaluates nicely and fast, without the need to code 
anything else, s2 requires that plus and times are rewritten and 
overloaded so they can be evaluated. Otherwise it doesn't work.

 >> eval(s1)
ans = 7

 >> eval(s2)
??? Error: ")" expected, "numeric value" found.

So you understand why this is not a good solution either. The string 
must be used as it is built, with no further modifications. If that is 
not possible, I would rather perform the evaluation using the tree 
instead of the string, as you did. I think it may be possible to use it 
while avoiding any "hard-coding" of the function names (so that new 
functions can be added one day without causing problems).

I am going to try a few other things (I am not expecting any miracle 
answers from the MATLAB newsgroup) and if I conclude it is not possible 
to use the string without causing problems, I will code the evaluation 
using the tree (something using both your code and my code). This way 
only the ant evaluation will be afected, although I may need to add a 
few arguments to other related functions.

Boy, this has made me realize a really important feature of MATLAB that 
I was totally unaware of!

Thanks for all your help and suggestions. Do you want to become GPLAB 
developers? :-)

Sara

PS - Sorry to all the mailing list subscribers that are not interested 
in the artificial ant. But others may be interested in the developments 
of this problem.

Re: [gplab-users] Ant example in GPLab

[<bo...@el...>]

Hi again Sara,


This mail just to reassure you: you don't need to apologize since you cau=
sed
us absolutely no trouble. The initial step where we studied the ant
functions and then modified them help us to better understand Matlab
programming since it was our first experience with it. Then GPLab saved u=
s
an enormous amount of time as our goal was to review the ant problem and
think about Genetic Programming not re-program the ant problem.

With GPLab we were able to conduct many tests as you may have noticed in =
our
sources and try different scenario including some suggested by Langdon an=
d
Poli in "Better trained artificial ants" as well as one which we imagine.
Indeed we tried to modify the ant fitness so that it is diminished by 1
every time the ant doesn't eat a pellet.
It yields to a fitness which can be negative and therefore is not directl=
y
understandable. However we showed that it makes the ant stay on the trail
and it increases the average number of pellets eaten! Of course it would
need further experiments, moreover we only used 200 time steps due to a l=
ack
of time.

So, it is us who thank you for your wonderful library and long live GPLab=
!



Concerning the problem with the "eval" function:

Someone has already replied to your message in the MATLAB newsgroup and h=
as
suggested to pass strings for the function names.
We had already tried this idea and that works.

Here is an example of the string that we have to use with the eval functi=
on.
Then it works well.

antprogn3('antprogn3(''antprogn2(''''antprogn3(''''''''antmove'''''''',''=
'''
'''antprogn3(''''''''''''''''antmove'''''''''''''''',''''''''''''''''antm=
ove
'''''''''''''''',''''''''''''''''antright'''''''''''''''')'''''''',''''''=
''a
ntprogn3(''''''''''''''''antmove'''''''''''''''',''''''''''''''''antright=
'''
''''''''''''',''''''''''''''''antleft'''''''''''''''')'''''''')'''',''''a=
ntm
ove'''')'',''antmove'',''antleft2'')','antprogn2(''antmove'',''antmove'')=
','
antif(''antmove'',''antif(''''antprogn2(''''''''antright'''''''',''''''''=
ant
progn2(''''''''''''''''antmove'''''''''''''''',''''''''''''''''antmove'''=
'''
'''''''''')'''''''')'''',''''antmove'''')'')')



"tree2str" function that we use to get this string:

"
function str=3Dtree2str(tree,level)
str=3Dstrcat(tree.op,'2');
args=3D[];
for k=3D1:length(tree.kids)
   args{k}=3Dtree2str(tree.kids{k},level+1);
end
if ~isempty(args)
 str=3Dstrcat(str,'(',implode(args,',',level),')');
end
"

where the "implode" function is:

"
function string=3Dimplode(pieces,delimiter,level)
ap =3D '';
for i=3D1: 2^(level-1)
    ap =3D strcat(ap, '''');
end
if isempty(pieces) % no pieces to join, return empty string
   string=3D'';
else % no need for delimiters yet, so far there's only one piece
   string=3Dstrcat(ap,pieces{1},ap);
end
l=3Dlength(pieces);
p=3D1;
while p<l % more than one piece to join with the delimiter, the interesti=
ng
case
   p=3Dp+1;
 string=3Dstrcat(string,delimiter,ap,pieces{p},ap);
end
"



tree2str2(BEST.tree, 1) was used to get the string above.

Then, the other functions should be modified too:
For example:

"
function ntime=3Dantif2(actiontrue, actionfalse)
global trail;
global x;
global y;
global direction;
global npellets;
global maxtime;
global ntime;
if antfoodahead2
   ans=3Deval(actiontrue);             // AND NOT ONLY "ACTIONTRUE"
else
   ans=3Deval(actionfalse);            // AND NOT ONLY "ACTIONFALSE"
end
"


Kind regards,
Flavien Billard, Thomas De Soza & Aur=E9lien Boffy





----- Original Message -----
From: "Sara Silva" <sa...@de...>
To: "Aur=E9lien Boffy" <bo...@el...>
Cc: "Thomas DE SOZA" <de...@el...>;
<gpl...@li...>; "Flavien Billard"
<bil...@el...>; "Marc Schoenauer" <Mar...@lr...>
Sent: Tuesday, July 12, 2005 8:18 PM
Subject: Re: [gplab-users] Ant example in GPLab


> Hi again,
>
> First of all, I must thank you and apologize once again for all the
> trouble you had. As I suspected, you are right in everything you say -
> doh!! :)
>
> Aur=E9lien Boffy wrote:
>
> > ...
> > - the main problem is in the "eval" function. We have coded an other
> > recursive function.
>
> Yes, the eval function is not behaving as I expected, at all. As soon a=
s
> it calls the antif function, *both* arguments are executed
> automatically, so the ant performs both actions. Then eval acts as if
> only the correct action was performed, but due to the usage of the
> global variables, the actions are not reversible and the correct state
> is not recovered! I am not sure if I missed this from the very
> beginning, or if the old version 5 of MATLAB (where most of GPLAB was
> developed) behaved differently.
>
> I have posted a help request to the MATLAB newsgroup to see if someone
> knows how to avoid this, because I must confess I don't like your
> solution either (but it is infinitely better than mine, of course - it
> works!). The problem is, one day you may want to change the problem
> slightly, add an additional function for the ant to perform, and you
> will have to recode both 'execution' and 'execution2' functions, which
> are already repeating a lot of code. Code reusability is a priority in
> GPLAB, so I am still trying to find a different solution based on the
> eval. But if this eval behavior cannot be avoided, you may be totally
> right in using the tree, not the string, in the evaluation.
>
> > - trail(y,x) instead of trail(x,y) twice:
> >         antmove : line 51, 52
> >         antfoodahead : line 30
>
> This one was entirely my fault (as if the other one wasn't...) and
> resulted from lack of attention. Of course it is a serious bug, but it
> should not affect learning and convergence - it is as if the ant is
> looking for food, not ahead, but on its side. And then it moves
> accordingly, so it's not so bad. I am saying this for the people who ma=
y
> be performing research experiments with the toolbox (like myself), and
> who may be concerned with the consequences this has on the dynamics of
> the evolution.
>
> The eval bug is much more serious. The ants have a much harder time
> learning (they do not benefit from the 'if') and do not perform as well
> as they should, but I always thought this was because I was using 400
> time steps intead of 600. In fact, the flawed ants have much less than
> 400 steps to look for food, because all the actions (even the ones that
> should not be performed) are using up time. I must review all my
> experiments on this! Curiously enough, in all my experiments with the
> flawed ants, the algorithm was able to reach an acceptable learning
> curve - each day I am more impressed with the learning power of GP...
>
> > - in the "antmove" function, line 52, we have to take care not to alw=
ays
> > remove the food pellet. Otherwise, the "antfoodahead" function always
> > returns that there is no food ahead...
>
> This was more or less my first and long solved bug... And I wish I coul=
d
> have spotted the other ones long ago, also. I guess I should have coded
> a visualization function for the ant behavior, as you did. I wasn't abl=
e
> to run your visualization function in version 6.1 of MATLAB, but I
> believe it can be easily changed so it works in older versions, too. Fo=
r
> now, I am going to post a warning message in GPLAB's webpage (and GP's
> mailing list), and later I will post new files with a *huge*
> acknowledgement to all you guys.
>
> Please let me know whatever else you discover on this problem. Once
> again, thank you.
> Sara
>
>
>
>
> > ----- Original Message -----
> > From: "Sara Silva" <sa...@de...>
> > To: "Thomas DE SOZA" <de...@el...>
> > Cc: <gpl...@li...>; "Flavien Billard"
> > <bil...@el...>; "Aur=E9lien Boffy" <bo...@el...>=
;
"Marc
> > Schoenauer" <Mar...@lr...>
> > Sent: Monday, July 11, 2005 1:18 PM
> > Subject: Re: [gplab-users] Ant example in GPLab
> >
> >
> >
> >>Hello there,
> >>
> >>My first reaction to your email is "yes, you are probably right". I h=
ave
> >>discovered bugs in the artificial ant before, I wouldn't be surprised=
 if
> >>there were others. I will check this as soon as I can and then report
> >>whatever findings I make, OK? For now, thanks for all your trouble, a=
nd
> >>I am so sorry if this has delayed your work.
> >>
> >>Another thing you may want to consider: In Koza's book the ant is
> >>supposed to be given 400 time steps to look for food, but Langdon and
> >>Poli (if I am not mistaken) have stated that the correct number of st=
eps
> >>was probably 600. GPLAB uses 400 as the default, but I guess if you
> >>don't change it to 600 you will not be able to reproduce Koza's resul=
ts.
> >>
> >>See you, and thanks once again,
> >>Sara
> >>
> >>
> >>
> >>Thomas DE SOZA wrote:
> >>
> >>>Hi Sara,
> >>>
> >>>We are 3 French students who have used your GPLab library during a
> >>>course in "Genetic Algorithms" lectured by Marc Schoenauer. Indeed w=
e
> >>>were supposed at the end of the course to make a small review of the
> >>>artificial ant problem and M. Schoenauer advised us to use Matlab an=
d
> >>>your library to make some tests.
> >>>
> >>>We are writing this mail to you because we think the ant example
> >>>included in GPLab may be flawed but as that was the first time we ha=
d
> >>>used Matlab and as we have only spent two weeks on it, we may be wro=
ng.
> >>>
> >>>Since the ant example in GPLab does not provide any graphical output
of
> >>>the ant walking on the Sante Fe Trail we coded a small function to
> >>>display the trail and the ant moving on it. That's how we discovered
> >>>that the artificial ant was not moving according to its tree. We
> >>>tracked down the problem and found 2 things :
> >>>
> >>>- first the Matlab eval function was not working as expected (we wer=
e
> >>>using Matlab 6.5 R13), that is the eval function did not interpret
well
> >>>the string returned by the tree2str function. For example, when
> >>>encountering "antif(antmove, antleft)", BOTH antmove and antleft wer=
e
> >>>executed.
> >>>
> >>>- second there is an inversion in the antmove function, when the ant
> >>>looks for food it looks at trail(newx, newy) but it should look at
> >>>trail(newy, newx) since the trail was acquired by the matlab load
> >>>function which acquires a matrix and therefore uses row and column c=
o-
> >>>ordinates.
> >>>
> >>>To correct this, we changed a few GPlab functions and added our own
> >>>eval function and it is working. In order to be sure we are not  wro=
ng,
> >>>can you confirm our discoveries ? Could it be because of our  Matlab
> >>>version ?
> >>>
> >>>Our work can be found at http://jc.desoza.free.fr/Fourmis.zip, but i=
t
> >>>is un-commented yet and it has some additionnal code to test improve=
d
> >>>artificial ants which might be confusing. If you could answer us on
the
> >>>points we mentioned earlier we could then comment our code and  make=
 it
> >>>easier to understand, then send it to you.
> >>>
> >>>Kind regards,
> >>>Flavien Billard, Aur=E9lien Boffy & Thomas De Soza
> >>>
> >>>-------------------------------------------------------
> >>>This SF.Net email is sponsored by the 'Do More With Dual!' webinar
> >>>happening
> >>>July 14 at 8am PDT/11am EDT. We invite you to explore the latest in
dual
> >>>core and dual graphics technology at this free one hour event hosted=
 by
> >>>HP, AMD, and NVIDIA.  To register visit
http://www.hp.com/go/dualwebinar
> >>>_______________________________________________
> >>>gplab-users mailing list
> >>>gpl...@li...
> >>>https://lists.sourceforge.net/lists/listinfo/gplab-users
> >>>
> >>>
> >>
> >
> >
>
>
>
> -------------------------------------------------------
> This SF.Net email is sponsored by the 'Do More With Dual!' webinar
happening
> July 14 at 8am PDT/11am EDT. We invite you to explore the latest in dua=
l
> core and dual graphics technology at this free one hour event hosted by
HP,
> AMD, and NVIDIA.  To register visit http://www.hp.com/go/dualwebinar
> _______________________________________________
> gplab-users mailing list
> gpl...@li...
> https://lists.sourceforge.net/lists/listinfo/gplab-users
>

Re: [gplab-users] Ant example in GPLab

[Sara S. <sa...@de...>]

Hi again,

First of all, I must thank you and apologize once again for all the 
trouble you had. As I suspected, you are right in everything you say - 
doh!! :)

Aurélien Boffy wrote:

> ...
> - the main problem is in the "eval" function. We have coded an other
> recursive function.

Yes, the eval function is not behaving as I expected, at all. As soon as 
it calls the antif function, *both* arguments are executed 
automatically, so the ant performs both actions. Then eval acts as if 
only the correct action was performed, but due to the usage of the 
global variables, the actions are not reversible and the correct state 
is not recovered! I am not sure if I missed this from the very 
beginning, or if the old version 5 of MATLAB (where most of GPLAB was 
developed) behaved differently.

I have posted a help request to the MATLAB newsgroup to see if someone 
knows how to avoid this, because I must confess I don't like your 
solution either (but it is infinitely better than mine, of course - it 
works!). The problem is, one day you may want to change the problem 
slightly, add an additional function for the ant to perform, and you 
will have to recode both 'execution' and 'execution2' functions, which 
are already repeating a lot of code. Code reusability is a priority in 
GPLAB, so I am still trying to find a different solution based on the 
eval. But if this eval behavior cannot be avoided, you may be totally 
right in using the tree, not the string, in the evaluation.

> - trail(y,x) instead of trail(x,y) twice:
>         antmove : line 51, 52
>         antfoodahead : line 30

This one was entirely my fault (as if the other one wasn't...) and 
resulted from lack of attention. Of course it is a serious bug, but it 
should not affect learning and convergence - it is as if the ant is 
looking for food, not ahead, but on its side. And then it moves 
accordingly, so it's not so bad. I am saying this for the people who may 
be performing research experiments with the toolbox (like myself), and 
who may be concerned with the consequences this has on the dynamics of 
the evolution.

The eval bug is much more serious. The ants have a much harder time 
learning (they do not benefit from the 'if') and do not perform as well 
as they should, but I always thought this was because I was using 400 
time steps intead of 600. In fact, the flawed ants have much less than 
400 steps to look for food, because all the actions (even the ones that 
should not be performed) are using up time. I must review all my 
experiments on this! Curiously enough, in all my experiments with the 
flawed ants, the algorithm was able to reach an acceptable learning 
curve - each day I am more impressed with the learning power of GP...

> - in the "antmove" function, line 52, we have to take care not to always
> remove the food pellet. Otherwise, the "antfoodahead" function always
> returns that there is no food ahead...

This was more or less my first and long solved bug... And I wish I could 
have spotted the other ones long ago, also. I guess I should have coded 
a visualization function for the ant behavior, as you did. I wasn't able 
to run your visualization function in version 6.1 of MATLAB, but I 
believe it can be easily changed so it works in older versions, too. For 
now, I am going to post a warning message in GPLAB's webpage (and GP's 
mailing list), and later I will post new files with a *huge* 
acknowledgement to all you guys.

Please let me know whatever else you discover on this problem. Once 
again, thank you.
Sara




> ----- Original Message -----
> From: "Sara Silva" <sa...@de...>
> To: "Thomas DE SOZA" <de...@el...>
> Cc: <gpl...@li...>; "Flavien Billard"
> <bil...@el...>; "Aurélien Boffy" <bo...@el...>; "Marc
> Schoenauer" <Mar...@lr...>
> Sent: Monday, July 11, 2005 1:18 PM
> Subject: Re: [gplab-users] Ant example in GPLab
> 
> 
> 
>>Hello there,
>>
>>My first reaction to your email is "yes, you are probably right". I have
>>discovered bugs in the artificial ant before, I wouldn't be surprised if
>>there were others. I will check this as soon as I can and then report
>>whatever findings I make, OK? For now, thanks for all your trouble, and
>>I am so sorry if this has delayed your work.
>>
>>Another thing you may want to consider: In Koza's book the ant is
>>supposed to be given 400 time steps to look for food, but Langdon and
>>Poli (if I am not mistaken) have stated that the correct number of steps
>>was probably 600. GPLAB uses 400 as the default, but I guess if you
>>don't change it to 600 you will not be able to reproduce Koza's results.
>>
>>See you, and thanks once again,
>>Sara
>>
>>
>>
>>Thomas DE SOZA wrote:
>>
>>>Hi Sara,
>>>
>>>We are 3 French students who have used your GPLab library during a
>>>course in "Genetic Algorithms" lectured by Marc Schoenauer. Indeed we
>>>were supposed at the end of the course to make a small review of the
>>>artificial ant problem and M. Schoenauer advised us to use Matlab and
>>>your library to make some tests.
>>>
>>>We are writing this mail to you because we think the ant example
>>>included in GPLab may be flawed but as that was the first time we had
>>>used Matlab and as we have only spent two weeks on it, we may be wrong.
>>>
>>>Since the ant example in GPLab does not provide any graphical output  of
>>>the ant walking on the Sante Fe Trail we coded a small function to
>>>display the trail and the ant moving on it. That's how we discovered
>>>that the artificial ant was not moving according to its tree. We
>>>tracked down the problem and found 2 things :
>>>
>>>- first the Matlab eval function was not working as expected (we were
>>>using Matlab 6.5 R13), that is the eval function did not interpret  well
>>>the string returned by the tree2str function. For example, when
>>>encountering "antif(antmove, antleft)", BOTH antmove and antleft were
>>>executed.
>>>
>>>- second there is an inversion in the antmove function, when the ant
>>>looks for food it looks at trail(newx, newy) but it should look at
>>>trail(newy, newx) since the trail was acquired by the matlab load
>>>function which acquires a matrix and therefore uses row and column co-
>>>ordinates.
>>>
>>>To correct this, we changed a few GPlab functions and added our own
>>>eval function and it is working. In order to be sure we are not  wrong,
>>>can you confirm our discoveries ? Could it be because of our  Matlab
>>>version ?
>>>
>>>Our work can be found at http://jc.desoza.free.fr/Fourmis.zip, but it
>>>is un-commented yet and it has some additionnal code to test improved
>>>artificial ants which might be confusing. If you could answer us on  the
>>>points we mentioned earlier we could then comment our code and  make it
>>>easier to understand, then send it to you.
>>>
>>>Kind regards,
>>>Flavien Billard, Aurélien Boffy & Thomas De Soza
>>>
>>>-------------------------------------------------------
>>>This SF.Net email is sponsored by the 'Do More With Dual!' webinar
>>>happening
>>>July 14 at 8am PDT/11am EDT. We invite you to explore the latest in dual
>>>core and dual graphics technology at this free one hour event hosted by
>>>HP, AMD, and NVIDIA.  To register visit http://www.hp.com/go/dualwebinar
>>>_______________________________________________
>>>gplab-users mailing list
>>>gpl...@li...
>>>https://lists.sourceforge.net/lists/listinfo/gplab-users
>>>
>>>
>>
> 
> 

Re: [gplab-users] Ant example in GPLab

[<bo...@el...>]

Hi Sara,

Thank you for replying so fast.
We had noticed that there are a few differences between GPLAB and the boo=
k
of Langdon and Poli (the number of time steps, but also the sampling meth=
od
(tournament instead of roulette) and some other "genetic parameters").

However, the main problem of the program is that the artificial ant is no=
t
moving according to its tree. We have considered its path step by step an=
d
we have confronted each movement with the tree.
You can check our work or we can say to you more precisely where are the
main mistakes:

- the main problem is in the "eval" function. We have coded an other
recursive function.
- trail(y,x) instead of trail(x,y) twice:
        antmove : line 51, 52
        antfoodahead : line 30
- in the "antmove" function, line 52, we have to take care not to always
remove the food pellet. Otherwise, the "antfoodahead" function always
returns that there is no food ahead...

Best regards,

Thomas De Soza, Flavien Billard & Aur=E9lien Boffy




----- Original Message -----
From: "Sara Silva" <sa...@de...>
To: "Thomas DE SOZA" <de...@el...>
Cc: <gpl...@li...>; "Flavien Billard"
<bil...@el...>; "Aur=E9lien Boffy" <bo...@el...>; "M=
arc
Schoenauer" <Mar...@lr...>
Sent: Monday, July 11, 2005 1:18 PM
Subject: Re: [gplab-users] Ant example in GPLab


> Hello there,
>
> My first reaction to your email is "yes, you are probably right". I hav=
e
> discovered bugs in the artificial ant before, I wouldn't be surprised i=
f
> there were others. I will check this as soon as I can and then report
> whatever findings I make, OK? For now, thanks for all your trouble, and
> I am so sorry if this has delayed your work.
>
> Another thing you may want to consider: In Koza's book the ant is
> supposed to be given 400 time steps to look for food, but Langdon and
> Poli (if I am not mistaken) have stated that the correct number of step=
s
> was probably 600. GPLAB uses 400 as the default, but I guess if you
> don't change it to 600 you will not be able to reproduce Koza's results.
>
> See you, and thanks once again,
> Sara
>
>
>
> Thomas DE SOZA wrote:
> > Hi Sara,
> >
> > We are 3 French students who have used your GPLab library during a
> > course in "Genetic Algorithms" lectured by Marc Schoenauer. Indeed we
> > were supposed at the end of the course to make a small review of the
> > artificial ant problem and M. Schoenauer advised us to use Matlab and
> > your library to make some tests.
> >
> > We are writing this mail to you because we think the ant example
> > included in GPLab may be flawed but as that was the first time we had
> > used Matlab and as we have only spent two weeks on it, we may be wron=
g.
> >
> > Since the ant example in GPLab does not provide any graphical output =
 of
> > the ant walking on the Sante Fe Trail we coded a small function to
> > display the trail and the ant moving on it. That's how we discovered
> > that the artificial ant was not moving according to its tree. We
> > tracked down the problem and found 2 things :
> >
> > - first the Matlab eval function was not working as expected (we were
> > using Matlab 6.5 R13), that is the eval function did not interpret  w=
ell
> > the string returned by the tree2str function. For example, when
> > encountering "antif(antmove, antleft)", BOTH antmove and antleft were
> > executed.
> >
> > - second there is an inversion in the antmove function, when the ant
> > looks for food it looks at trail(newx, newy) but it should look at
> > trail(newy, newx) since the trail was acquired by the matlab load
> > function which acquires a matrix and therefore uses row and column co=
-
> > ordinates.
> >
> > To correct this, we changed a few GPlab functions and added our own
> > eval function and it is working. In order to be sure we are not  wron=
g,
> > can you confirm our discoveries ? Could it be because of our  Matlab
> > version ?
> >
> > Our work can be found at http://jc.desoza.free.fr/Fourmis.zip, but it
> > is un-commented yet and it has some additionnal code to test improved
> > artificial ants which might be confusing. If you could answer us on  =
the
> > points we mentioned earlier we could then comment our code and  make =
it
> > easier to understand, then send it to you.
> >
> > Kind regards,
> > Flavien Billard, Aur=E9lien Boffy & Thomas De Soza
> >
> > -------------------------------------------------------
> > This SF.Net email is sponsored by the 'Do More With Dual!' webinar
> > happening
> > July 14 at 8am PDT/11am EDT. We invite you to explore the latest in d=
ual
> > core and dual graphics technology at this free one hour event hosted =
by
> > HP, AMD, and NVIDIA.  To register visit http://www.hp.com/go/dualwebi=
nar
> > _______________________________________________
> > gplab-users mailing list
> > gpl...@li...
> > https://lists.sourceforge.net/lists/listinfo/gplab-users
> >
> >
>

Re: [gplab-users] Ant example in GPLab

[Sara S. <sa...@de...>]

Hello there,

My first reaction to your email is "yes, you are probably right". I have 
discovered bugs in the artificial ant before, I wouldn't be surprised if 
there were others. I will check this as soon as I can and then report 
whatever findings I make, OK? For now, thanks for all your trouble, and 
I am so sorry if this has delayed your work.

Another thing you may want to consider: In Koza's book the ant is 
supposed to be given 400 time steps to look for food, but Langdon and 
Poli (if I am not mistaken) have stated that the correct number of steps 
was probably 600. GPLAB uses 400 as the default, but I guess if you 
don't change it to 600 you will not be able to reproduce Koza's results.

See you, and thanks once again,
Sara



Thomas DE SOZA wrote:
> Hi Sara,
> 
> We are 3 French students who have used your GPLab library during a  
> course in "Genetic Algorithms" lectured by Marc Schoenauer. Indeed we  
> were supposed at the end of the course to make a small review of the  
> artificial ant problem and M. Schoenauer advised us to use Matlab and  
> your library to make some tests.
> 
> We are writing this mail to you because we think the ant example  
> included in GPLab may be flawed but as that was the first time we had  
> used Matlab and as we have only spent two weeks on it, we may be wrong.
> 
> Since the ant example in GPLab does not provide any graphical output  of 
> the ant walking on the Sante Fe Trail we coded a small function to  
> display the trail and the ant moving on it. That's how we discovered  
> that the artificial ant was not moving according to its tree. We  
> tracked down the problem and found 2 things :
> 
> - first the Matlab eval function was not working as expected (we were  
> using Matlab 6.5 R13), that is the eval function did not interpret  well 
> the string returned by the tree2str function. For example, when  
> encountering "antif(antmove, antleft)", BOTH antmove and antleft were  
> executed.
> 
> - second there is an inversion in the antmove function, when the ant  
> looks for food it looks at trail(newx, newy) but it should look at  
> trail(newy, newx) since the trail was acquired by the matlab load  
> function which acquires a matrix and therefore uses row and column co- 
> ordinates.
> 
> To correct this, we changed a few GPlab functions and added our own  
> eval function and it is working. In order to be sure we are not  wrong, 
> can you confirm our discoveries ? Could it be because of our  Matlab 
> version ?
> 
> Our work can be found at http://jc.desoza.free.fr/Fourmis.zip, but it  
> is un-commented yet and it has some additionnal code to test improved  
> artificial ants which might be confusing. If you could answer us on  the 
> points we mentioned earlier we could then comment our code and  make it 
> easier to understand, then send it to you.
> 
> Kind regards,
> Flavien Billard, Aurélien Boffy & Thomas De Soza
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by the 'Do More With Dual!' webinar 
> happening
> July 14 at 8am PDT/11am EDT. We invite you to explore the latest in dual
> core and dual graphics technology at this free one hour event hosted by 
> HP, AMD, and NVIDIA.  To register visit http://www.hp.com/go/dualwebinar
> _______________________________________________
> gplab-users mailing list
> gpl...@li...
> https://lists.sourceforge.net/lists/listinfo/gplab-users
> 
> 

[gplab-users] Ant example in GPLab

[Thomas DE S. <de...@el...>]

Hi Sara,

We are 3 French students who have used your GPLab library during a =20
course in "Genetic Algorithms" lectured by Marc Schoenauer. Indeed we =20=

were supposed at the end of the course to make a small review of the =20
artificial ant problem and M. Schoenauer advised us to use Matlab and =20=

your library to make some tests.

We are writing this mail to you because we think the ant example =20
included in GPLab may be flawed but as that was the first time we had =20=

used Matlab and as we have only spent two weeks on it, we may be wrong.

Since the ant example in GPLab does not provide any graphical output =20
of the ant walking on the Sante Fe Trail we coded a small function to =20=

display the trail and the ant moving on it. That's how we discovered =20
that the artificial ant was not moving according to its tree. We =20
tracked down the problem and found 2 things :

- first the Matlab eval function was not working as expected (we were =20=

using Matlab 6.5 R13), that is the eval function did not interpret =20
well the string returned by the tree2str function. For example, when =20
encountering "antif(antmove, antleft)", BOTH antmove and antleft were =20=

executed.

- second there is an inversion in the antmove function, when the ant =20
looks for food it looks at trail(newx, newy) but it should look at =20
trail(newy, newx) since the trail was acquired by the matlab load =20
function which acquires a matrix and therefore uses row and column co-=20=

ordinates.

To correct this, we changed a few GPlab functions and added our own =20
eval function and it is working. In order to be sure we are not =20
wrong, can you confirm our discoveries ? Could it be because of our =20
Matlab version ?

Our work can be found at http://jc.desoza.free.fr/Fourmis.zip, but it =20=

is un-commented yet and it has some additionnal code to test improved =20=

artificial ants which might be confusing. If you could answer us on =20
the points we mentioned earlier we could then comment our code and =20
make it easier to understand, then send it to you.

Kind regards,
Flavien Billard, Aur=E9lien Boffy & Thomas De Soza=

[gplab-users] New files available for GPLAB Version 2

[Sara S. <sa...@de...>]

Dear GPLAB users,

New files are available for GPLAB Version 2. They make the toolbox run 
faster. You can find them here:
http://gplab.sourceforge.net/download.html

As always, comments and suggestions are welcome.

Cheers,
Sara

Re: [gplab-users] Parsing Software

[Sara S. <sa...@de...>]

Hi Paul,

I don't know of any parser, but I wonder if a simple search and replace 
would do the trick? If not, maybe building a function similar to 
tree2str.m so that it translates a GPLAB tree into Excel format, instead 
of MATLAB format. I hope this helps somehow.

Cheers,
Sara


(s) Paul Downs wrote:

> Hi
> 
> I am using GPLAB to carry out riverflow modelling. However I need to import the generated equations into Excel. Do you know of any Parsing software that wil turn the generated equation into a format Excel will understand
> 
> 
> Thanks 
> 
> 
> Paul
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by:
> Sybase ASE Linux Express Edition - download now for FREE
> LinuxWorld Reader's Choice Award Winner for best database on Linux.
> http://ads.osdn.com/?ad_idU88&alloc_id065&opÌk
> _______________________________________________
> gplab-users mailing list
> gpl...@li...
> https://lists.sourceforge.net/lists/listinfo/gplab-users
> 

[gplab-users] Parsing Software

[\(s\) P. D. <pau...@st...>]

Hi

I am using GPLAB to carry out riverflow modelling. However I need to =
import the generated equations into Excel. Do you know of any Parsing =
software that wil turn the generated equation into a format Excel will =
understand


Thanks=20


Paul

[gplab-users] GPLAB 2.0 new bug fix available

[Sara S. <sa...@de...>]

Dear GPLAB users,

A new bug fix is available for GPLAB 2.0:
http://gplab.sourceforge.net/download.html

Cheers,
Sara

[gplab-users] Bug fixes available for GPLAB Version 2

[Sara S. <sa...@de...>]

Dear GPLAB users,

Bug fixes are now available for GPLAB Version 2:
http://gplab.sourceforge.net/download.html

Cheers,
Sara

Re: [gplab-users] system identification

[Sara S. <sa...@de...>]

Hi there,

No, you cannot add variables to the terminal set, but could it be that 
the time delay you mentioned is a function, not a terminal? Sorry for my 
ignorance in this field...!

If anyone can provide any help, it will much appreciated! Thanks.

Sara


she...@ma... wrote:

> hi
> 
> i'm trying to set a time delay (z^-1 or u(t-n)) as a terminal in my terminal
> list but having difficulty.  Is it possible to add variables to the terminal
> list or can i only add constants and 'rand'? if i can how do i get gplab
> to recognise the time delay.
> I want my tree to be evaluated as a transfer function eg. (3z^-1)/(1+z^-1).
> 
> Thanks
> Sheena
> 
> 
> 
> -------------------------------------------------------
> SF.Net email is sponsored by Shop4tech.com-Lowest price on Blank Media
> 100pk Sonic DVD-R 4x for only $29 -100pk Sonic DVD+R for only $33
> Save 50% off Retail on Ink & Toner - Free Shipping and Free Gift.
> http://www.shop4tech.com/z/Inkjet_Cartridges/9_108_r285
> _______________________________________________
> gplab-users mailing list
> gpl...@li...
> https://lists.sourceforge.net/lists/listinfo/gplab-users
> 

[gplab-users] system identification

[<she...@ma...>]

hi

i'm trying to set a time delay (z^-1 or u(t-n)) as a terminal in my termi=
nal
list but having difficulty.  Is it possible to add variables to the termi=
nal
list or can i only add constants and 'rand'? if i can how do i get gplab
to recognise the time delay.
I want my tree to be evaluated as a transfer function eg. (3z^-1)/(1+z^-1=
).

Thanks
Sheena

Re: [gplab-users] demos

[Sara S. <sa...@de...>]

Hi Sheena,

I'm not sure I understand. Are you actually trying to compile the 
toolbox? Do the demos *run* at all? What version of MATLAB are you using?

Cheers,
Sara


she...@ma... wrote:

> Hi Sara, 
> 
> I've just started using your gplab but the demos for version2 dont compile
> with toolbox version2. they will however compile with toolbox version1!
> 
> Regards, 
> Sheena
> 
> 
> 
> -------------------------------------------------------
> This SF.Net email is sponsored by OSTG. Have you noticed the changes on
> Linux.com, ITManagersJournal and NewsForge in the past few weeks? Now,
> one more big change to announce. We are now OSTG- Open Source Technology
> Group. Come see the changes on the new OSTG site. www.ostg.com
> _______________________________________________
> gplab-users mailing list
> gpl...@li...
> https://lists.sourceforge.net/lists/listinfo/gplab-users
> 

[gplab-users] demos

[<she...@ma...>]

Hi Sara, 

I've just started using your gplab but the demos for version2 dont compil=
e
with toolbox version2. they will however compile with toolbox version1!

Regards, 
Sheena

[gplab-users] GPLAB version 2

[Sara S. <sa...@de...>]

Dear GPLAB users,

GPLAB version 2 is finally ready and available at its new site:
http://gplab.sourceforge.net

Here are the changes from the previous version:
http://gplab.sourceforge.net/features.html#changes

Users who wish to be informed whenever new releases (and bug fixes) 
become available should subscribe the recently created gplab-users 
mailing list at:

https://lists.sourceforge.net/lists/listinfo/gplab-users

In the future I will not send the announcing e-mails directly to the 
users, but only to the mailing list (I will also keep posting them in 
the GP mailing list, though).

Thank you all!
Sara Silva
sa...@de...
http://eden.dei.uc.pt/~sara/
(new homepage)