http://kirill-kryukov.com/chess/discuss ... p?id=37193
Schooner Version 1.5
Moderator: Ras
-
D Sceviour
- Posts: 570
- Joined: Mon Jul 20, 2015 5:06 pm
Re: Schooner Version 1.5
Thank you Graham for the shared link on CCRL.
Reminder: The Schooner files will have to be renamed *.exe.
http://kirill-kryukov.com/chess/discuss ... p?id=37193
http://kirill-kryukov.com/chess/discuss ... p?id=37193
-
Sylwy
- Posts: 5049
- Joined: Fri Apr 21, 2006 4:19 pm
- Location: IAȘI - the historical capital of MOLDOVA
- Full name: Silvian Rucsandescu
-
D Sceviour
- Posts: 570
- Joined: Mon Jul 20, 2015 5:06 pm
Re: My test
Ruxy Sylwyka,
Thank you for the Nemeton results. In the first game in the CCRL 40/4 against Alex 2.14a, Schooner1.5 played 68. R-f4?? and lost a near adjudicated win:
[d]r7/P7/1K2p3/5p1k/R7/8/8/8 w - - 5 69
It is difficult to figure out what went wrong. If you paste the position and analyze it using Schooner, you will see it plays the best move Kb7! There is something wrong in the hash table between game moves and it errors. This type of error (bad game moves that cannot be duplicated in console mode) is very difficult to trace because one cannot simply set up the position and test it. The only way to fix this at the current moment is to erase the hash table between game moves, but then it plays weaker overall. Let me know if this is preferred and I will issue an update.
There are still plenty of things left to do in the engine, but there comes a time when one has to stop working on the thing and post the results. Graham Banks has been looking for an update for some time. I have lots of ideas to make it play 10 to 20 elo stronger, but I have no idea how to make it play 800 elo stronger. Perhaps it is time to abandon the engine and put my talents towards an existing stronger engine. Schooner still has lots of original ideas which do make good computer chess moves.
That dumb link in the OP might make viewing this article difficult. The moderators can delete the first link in the OP if they want since it doesn't work. In fact, bad links should be deleted anyway. Otherwise viewing in thread mode makes it easier.
Thank you for the Nemeton results. In the first game in the CCRL 40/4 against Alex 2.14a, Schooner1.5 played 68. R-f4?? and lost a near adjudicated win:
[d]r7/P7/1K2p3/5p1k/R7/8/8/8 w - - 5 69
It is difficult to figure out what went wrong. If you paste the position and analyze it using Schooner, you will see it plays the best move Kb7! There is something wrong in the hash table between game moves and it errors. This type of error (bad game moves that cannot be duplicated in console mode) is very difficult to trace because one cannot simply set up the position and test it. The only way to fix this at the current moment is to erase the hash table between game moves, but then it plays weaker overall. Let me know if this is preferred and I will issue an update.
There are still plenty of things left to do in the engine, but there comes a time when one has to stop working on the thing and post the results. Graham Banks has been looking for an update for some time. I have lots of ideas to make it play 10 to 20 elo stronger, but I have no idea how to make it play 800 elo stronger. Perhaps it is time to abandon the engine and put my talents towards an existing stronger engine. Schooner still has lots of original ideas which do make good computer chess moves.
That dumb link in the OP might make viewing this article difficult. The moderators can delete the first link in the OP if they want since it doesn't work. In fact, bad links should be deleted anyway. Otherwise viewing in thread mode makes it easier.
-
carldaman
- Posts: 2287
- Joined: Sat Jun 02, 2012 2:13 am
Re: My test
Other engines, even stronger ones, make such hash-related blunders occasionally. Does Schooner do this too frequently in your view?D Sceviour wrote:Ruxy Sylwyka,
Thank you for the Nemeton results. In the first game in the CCRL 40/4 against Alex 2.14a, Schooner1.5 played 68. R-f4?? and lost a near adjudicated win:
[d]r7/P7/1K2p3/5p1k/R7/8/8/8 w - - 5 69
It is difficult to figure out what went wrong. If you paste the position and analyze it using Schooner, you will see it plays the best move Kb7! There is something wrong in the hash table between game moves and it errors. This type of error (bad game moves that cannot be duplicated in console mode) is very difficult to trace because one cannot simply set up the position and test it. The only way to fix this at the current moment is to erase the hash table between game moves, but then it plays weaker overall. Let me know if this is preferred and I will issue an update.
There are still plenty of things left to do in the engine, but there comes a time when one has to stop working on the thing and post the results. Graham Banks has been looking for an update for some time. I have lots of ideas to make it play 10 to 20 elo stronger, but I have no idea how to make it play 800 elo stronger. Perhaps it is time to abandon the engine and put my talents towards an existing stronger engine. Schooner still has lots of original ideas which do make good computer chess moves.
That dumb link in the OP might make viewing this article difficult. The moderators can delete the first link in the OP if they want since it doesn't work. In fact, bad links should be deleted anyway. Otherwise viewing in thread mode makes it easier.
CL
-
D Sceviour
- Posts: 570
- Joined: Mon Jul 20, 2015 5:06 pm
Re: My test
Once is too frequent for me, but yes it does occur often. It is irritating that I do not know where to look for this type of error. Is it memory leak, bad PV restoration? Perhaps this problem should be posted in the Programming and Technical discussions.carldaman wrote:Other engines, even stronger ones, make such hash-related blunders occasionally. Does Schooner do this too frequently in your view?
-
cdani
- Posts: 2204
- Joined: Sat Jan 18, 2014 10:24 am
- Location: Andorra
Re: My test
Maybe a bug in legal move verification? Also if not doing it already, maybe you can avoid returning immediately on tt hit if pv node.D Sceviour wrote: There is something wrong in the hash table between game moves and it errors.
I know the only way existing, persistenceD Sceviour wrote: I have lots of ideas to make it play 10 to 20 elo stronger, but I have no idea how to make it play 800 elo stronger.
Working on own engine I think is funnier and one feels freer. I think is more frustrating but also more rewarding.D Sceviour wrote:Perhaps it is time to abandon the engine and put my talents towards an existing stronger engine.
Daniel José -
http://www.andscacs.com