The organizers of the Gifu Challenge have announced that they will use a new server-based protocol named SBCP instead of the Modem Protocol in future tournaments. A description of the protocol can be found here: http://igo.bz/cgf/CGF2004-Protocol.txt
I am not happy that they did not choose GTP, it would have been a good choice for all the requirements listed in the their specification.
Programmers that already use GTP for regression testing and interfacing the engine to a user interface will have to support another protocol or use a program translating the protocols.
I know that at least 4 Go programs which participated in tournaments already support GTP: GnuGo, Explorer, NeuroGo and Golois (anyone knows more?).
- Markus
On Mon, 4 Aug 2003, Markus Enzenberger wrote:
The organizers of the Gifu Challenge have announced that they will use a new server-based protocol named SBCP instead of the Modem Protocol in future tournaments. A description of the protocol can be found here: http://igo.bz/cgf/CGF2004-Protocol.txt
I am not happy that they did not choose GTP, it would have been a good choice for all the requirements listed in the their specification.
Programmers that already use GTP for regression testing and interfacing the engine to a user interface will have to support another protocol or use a program translating the protocols.
I know that at least 4 Go programs which participated in tournaments already support GTP: GnuGo, Explorer, NeuroGo and Golois (anyone knows more?).
Let me add that bridges between GTP and GMP already exist, and would certainly be ready for the next tournament. So all other programs would just have to setup this bridge, and would not have to implement a new protocol, either.
Regards, Arend Bayer
On Monday, August 4, 2003, at 06:08 PM, Markus Enzenberger wrote:
The organizers of the Gifu Challenge have announced that they will use a new server-based protocol named SBCP instead of the Modem Protocol in future tournaments. A description of the protocol can be found here: http://igo.bz/cgf/CGF2004-Protocol.txt
I am not happy that they did not choose GTP, it would have been a good choice for all the requirements listed in the their specification.
I agree that defining yet another protocol is cumbersome, but one thing that the Gifu Challenge wants to do is to use a server (WING, a NNGS derivative I believe) to connect the two competing programs, which is a good thing IMHO. To do this with GTP would still require some kind of GTP - WING interface, isn't it?
Of course, some programs also already routinely play on servers (GNU Go, Many Faces, ...), but I think that in addition to the ability to play you would need something to setup the game (or maybe the game proposal can already be sent by the server instead of one of the programs?)
Marco Scheurer Sen:te, Lausanne, Switzerland http://www.sente.ch
IMHO. To do this with GTP would still require some kind of GTP - WING interface, isn't it?
when using GTP the referee program would connect to the 2 Go programs (actually you would have to call the programs server). The referee program would control the programs decide, when a new game is started, moves are played and generated.
You can already use GoGui (http://gogui.sf.net) for that. Version 0.2.2 contains several tools (TwoGtp, NetGtp, GtpNet, GmpToGtp) for playing games between programs which may be somewhere else on the network or connected to a computer by serial line. GoGui lacks some features that might be required in a tournament, but it is a working example how you could use a referee program and GTP for automatically playing games.
- Markus
Markus wrote:
The organizers of the Gifu Challenge have announced that they will use a new server-based protocol named SBCP instead of the Modem Protocol in future tournaments. A description of the protocol can be found here: http://igo.bz/cgf/CGF2004-Protocol.txt
I am not happy that they did not choose GTP, it would have been a good choice for all the requirements listed in the their specification.
Features (d), (e), and (l) in the list wouldn't be directly supported by using GTP, but it shouldn't be very difficult to write a tournament program which mirrors the games to a go server. Features (j) and (k) have very little to do with computer go tournaments. Otherwise I agree that GTP fits the bill perfectly.
Programmers that already use GTP for regression testing and interfacing the engine to a user interface will have to support another protocol
That protocol looks nice as go server protocols go and is clearly superior to the IGS/NNGS protocol for use by programs. Still it doesn't look very attractive to implement in an engine compared to GTP.
or use a program translating the protocols.
This is certainly the approach GNU Go would take. We already have a NNGS-GTP bridge (gnugoclient) and this would be similar (but cleaner due to the nicer protocol). Absolutely doable but preferrable if it wasn't needed (i.e. if the tournament software talked GTP directly with the go programs).
[quoted from a later message:]
when using GTP the referee program would connect to the 2 Go programs (actually you would have to call the programs server).
It doesn't really matter whether the referee connects to the engines or the other way round. Both variants can be used.
/Gunnar