Winboard Forum

Hi all,
Readme of RomiChessP3b said engine use now learn.dat file as openingbook. I have the feeling this is not working or I have a wrong learn.dat file.
Please your comments,
Pablo

Hi all,
Readme of RomiChessP3b said engine use now learn.dat file as openingbook. I have the feeling this is not working or I have a wrong learn.dat file.
Please your comments,
Pablo

Well, it uses the learnfile as a 'kind' of opening book _after_
it has learned something of course
As long as you haven't played a lot of games you won't see
big effects. It also depends on what kind of tourneys you
run. In single matches over a big game number you should see
a bigger difference.

However I take the opportunity to ask Michael in this thread, if he plans
to release already _prepared_ learn files for Romi in the future?

I guess a lot of people reset learning after tournaments or matches
for fair condition, but if some learning would be already given in
the offered file, I would always take that one as a base, as for other
programs with similar features.

Regards,
Guenther

Guenther Simon wrote:Hi all,
Readme of RomiChessP3b said engine use now learn.dat file as openingbook. I have the feeling this is not working or I have a wrong learn.dat file.
Please your comments,
Pablo

Well, it uses the learnfile as a 'kind' of opening book _after_
it has learned something of course
As long as you haven't played a lot of games you won't see
big effects. It also depends on what kind of tourneys you
run. In single matches over a big game number you should see
a bigger difference.

However I take the opportunity to ask Michael in this thread, if he plans
to release already _prepared_ learn files for Romi in the future?

I guess a lot of people reset learning after tournaments or matches
for fair condition, but if some learning would be already given in
the offered file, I would always take that one as a base, as for other
programs with similar features.

Regards,
Guenther

Hello G?nther
Thanks for you reply. Now is clear.
Anyhow as a curiosity, if you open RomiChess with DOS GUI, the program automatically create a 23,4MB learn.dat good for nothing.
I think is a very good idea to try to obtain from Michael an "already prepared learn.dat".
Best wishes,
Pablo

Hello G?nther
Thanks for you reply. Now is clear.
Anyhow as a curiosity, if you open RomiChess with DOS GUI, the program automatically create a 23,4MB learn.dat good for nothing.

Hi Pablo,
May be a little missunderstanding? It creates this only the first
time the program is started and if no learn file is already there.
But as explained above it contains no learning in the beginning.

Regards,
Guenther

Hi Pablo,
Hi Guenther,

Romi will in the future have a book system. The learn file will then be an extention of Romis book. Books are to be primarily for personalities or for training against a particular opening or a general book if that is what a human opponent wants to play against. I had hoped that for computer competitions Romi would be able to create her own book from solely her own experiance. That way, when another engine meets Romi in battle it is Romi that is playing the opening moves and not a computers version of ECO. So what must be done to achive this if learning is not maintained from one event to the next? An already prepared file would not allow anyone to observe the progress that Romi can make in the designed manner. I guess as an engine author I must be flexible and provide differant enviroments to suit the purposses of the individuals using my program. Therefore I will provide two methods for having a book. One, the canned aproach and two, the experiential. In the canned approach a pgn file can be read and put into an opening book or into the learn file. In the experiential approach Romi can be allowed to gradually create her own book as is now the case, or a prepared file can be shiped around. Also a merging function may be included so people can combine thier efforts, if they wish to see faster progress. The first thing that I will do is write a function to read a pgn file, say gm2600.pgn, into the learn file and treat each game as though it was a game that Romi played herself. This seems as though it would please the most users of Romi the quickest and may demostrate where this methodology will arrive at if Romi were to play enough games. Any ideas are welcome.

Thanks,
Mike

Michael Sherwin wrote:Hi Pablo,
Hi Guenther,

Romi will in the future have a book system. The learn file will then be an extention of Romis book. Books are to be primarily for personalities or for training against a particular opening or a general book if that is what a human opponent wants to play against. I had hoped that for computer competitions Romi would be able to create her own book from solely her own experiance. That way, when another engine meets Romi in battle it is Romi that is playing the opening moves and not a computers version of ECO. So what must be done to achive this if learning is not maintained from one event to the next? An already prepared file would not allow anyone to observe the progress that Romi can make in the designed manner. I guess as an engine author I must be flexible and provide differant enviroments to suit the purposses of the individuals using my program. Therefore I will provide two methods for having a book. One, the canned aproach and two, the experiential. In the canned approach a pgn file can be read and put into an opening book or into the learn file. In the experiential approach Romi can be allowed to gradually create her own book as is now the case, or a prepared file can be shiped around. Also a merging function may be included so people can combine thier efforts, if they wish to see faster progress. The first thing that I will do is write a function to read a pgn file, say gm2600.pgn, into the learn file and treat each game as though it was a game that Romi played herself. This seems as though it would please the most users of Romi the quickest and may demostrate where this methodology will arrive at if Romi were to play enough games. Any ideas are welcome.

Thanks,
Mike

Hello Mike
Great idea,
Pablo