Sender: |
|
Date: |
Tue, 24 Jan 2006 22:59:37 +0100 |
Reply-To: |
|
Content-Transfer-Encoding: |
7bit |
Content-Type: |
text/plain; charset=ISO-8859-1; format=flowed |
Subject: |
|
From: |
|
MIME-Version: |
1.0 |
In-Reply-To: |
|
Organization: |
CSTB |
Comments: |
|
Parts/Attachments: |
|
|
Thanks George for your quick reply,
> I am also an ECJ newbie, but it can't hurt to comment on your
> email. It always makes me feel better when people respond to me, even
> when they are wrong :).
I agree :-)
> I think that item 1/ is sensible. main() is not that difficult to
> re-implement, especialy if you are willing to do basically cut and paste
> from main() into the new webMain or whatever you call it.
It was my conclusion. In fact, I have put in the doPost what is more or
less done in Evolve.main (simplified a bit, since I don't need restart
from checkpoint)
> I do not
> understand very well what you mean in item 2/. By end-users input do you
> mean parameters such as those that appear in the various params files?
End-user parameters is the problem specific inputs which are entered in
form by the user, and describing the context in which he wants to
find a solution. To be a bit more precise, they are for instance the
dimensions of the piece to be processed by the system we are trying to
configure with the GA. These parameters are conveyed by the Web
container in the parameters of the Web request.
It would be possible to prgramatically store them in the ACJ parameter
database, but I don't see a real interest in doing this. My idea was to
store them in the EvolutionState.statics which seems to be there to make
app specific data available to the various objects involved in the process.
> As for finding the fitest individual, is you look at
> SimpleStatistics.finalStatistics(), for example, you will see that the
> best_of_run array is where the best individuals are contained. It should
> be fairly straightforward to access that member variable at the end of
> the run, since it is public. I have not looked at other implementations
> of Statistics, but I am pretty sure you will find it similarly easy to
> locate the best individuals such as each Statistics subclass considers
> them to be.
OK. That's the way I understood the philosophy of the system. Seems that
we are at least to share the same vision, so it should not be so wrong ;-)
Thanks a lot for your help.
Best regards
Eric Pascual
|
|
|