Octochess r4632

[Codesquid]

A new version of Octochess has been released.

Probably the most important change is that Octochess now follows the limits set by the GUI on the size of the transposition table and the number of threads to use, making it much easier to test Octochess against other engines under equal conditions.

• CECP and UCI: Number of threads can be set
• CECP and UCI: Size of transposition table can be set
• CECP: Support fixed move time (st command)
• Improved playing strength
• Specialized endgame evaluation
• Many small performance optimizations

You can download the new version from http://octochess.org/octochess-r4632.7z

[lucasart]

A new version of Octochess has been released.

Probably the most important change is that Octochess now follows the limits set by the GUI on the size of the transposition table and the number of threads to use, making it much easier to test Octochess against other engines under equal conditions.

• CECP and UCI: Number of threads can be set
• CECP and UCI: Size of transposition table can be set
• CECP: Support fixed move time (st command)
• Improved playing strength
• Specialized endgame evaluation
• Many small performance optimizations

You can download the new version from http://octochess.org/octochess-r4632.7z

Thank you Tim for the new release. But unfortunately, it really doesn't work for me. Sample session

Code: Select all

uci
id name Octochess
id author Tim Kosse
option name Hash type spin default 1314 min 4 max 1048576
option name Threads type spin default 2 min 1 max 2
uciok
ucinewgame
position startpos moves e2e4
go movetime 10000
bestmove e7e6
go depth 10
bestmove g1f3
quit

* bestmove e7e6 was returned immediately, and no search seems to have been performed (or perhaps a very very fast one with no display at all). I tried go wtime winc btime binc and same thing happenned.
* go depth 10 returned bestmove g1f3 ? who said that e7e6 should have been played ? the whole point of UCI is to not be path dependant and you specify the position and search conditions and treat each search as a new one.
* quit is irresponsive
* my task manager is not seeing 1.3 GB of memory consumption, so perhaps there was something wrong allocating the memory perhaps ? besides, 1314 MB is really not a good choice for a default value. Use something small, like a common like 16 or 32 and think of it as a common denominator for everyone (thing Octochess should be useable on a cell phone). even better is to alloc the TT at the last moment when a setoptin name Hash is issued or a go command is issued and the TT is still not allocated. this would allow it to run on small hardware and not try to alloc big memory before it is told.

[Codesquid]

bestmove e7e6 was returned immediately, and no search seems to have been performed (or perhaps a very very fast one with no display at all). I tried go wtime winc btime binc and same thing happenned.

Octochess brings its own opening book. What you see is a book move. Either start Octochess with the --nobook argument or delete opening_book.db

UCI's OwnBook option isn't understood yet.

go depth 10 returned bestmove g1f3 ? who said that e7e6 should have been played ? the whole point of UCI is to not be path dependant and you specify the position and search conditions and treat each search as a new one.

Interesting. Looks like UCI differs from CECP here. I'll have this fixed in the next version.

quit is irresponsive

Thanks for spotting this. Will be fixed soon.

my task manager is not seeing 1.3 GB of memory consumption, so perhaps there was something wrong allocating the memory perhaps ?

Memory will only be allocated when an actual search starts, which doesn't happen for book moves.

besides, 1314 MB is really not a good choice for a default value. Use something small, like a common like 16 or 32 and think of it as a common denominator for everyone (thing Octochess should be useable on a cell phone).

It's the default for your machine, because your machine has 4GiB of RAM. The default is always a third of total available RAM.

even better is to alloc the TT at the last moment when a setoptin name Hash is issued or a go command is issued and the TT is still not allocated. this would allow it to run on small hardware and not try to alloc big memory before it is told.

Octochess allocates the table at the last moment already.

[Codesquid]

Update: r4638 has been released which addresses the UCI issues found by Lucas.

• UCI: Fix quit command not actually quitting
• UCI: After returning bestmove after a go, the move is not actually applied
• UCI: Implement support for the OwnBook option
• UCI: Basic support for the movetime argument to the go command

You can download the new version from http://octochess.org/

@Moderators: Can you please adjust the topic title and the link in the original posting? I can't seem to edit my old posts.

[El Gringo]

Hi,

Try to unzip it with Winzip or WinRar > errors see below

! C:\share\octochess-r4638.7z: Unknown method in octochess-linux-generic-r4638
! C:\share\octochess-r4638.7z: Unknown method in octochess-linux-sse4-r4638
! C:\share\octochess-r4638.7z: Unknown method in authors.txt
! C:\share\octochess-r4638.7z: Unknown method in copying.txt
! C:\share\octochess-r4638.7z: Unknown method in news.txt
! C:\share\octochess-r4638.7z: Unknown method in readme.txt
! C:\share\octochess-r4638.7z: Unknown method in opening_book.db
! C:\share\octochess-r4638.7z: Unknown method in octochess-windows-generic-r4638.exe
! C:\share\octochess-r4638.7z: Unknown method in octochess-windows-sse4-r4638.exe
! C:\share\octochess-r4638.7z: operation failed

???

Johan

[mar]

Hi,

Try to unzip it with Winzip or WinRar > errors see below

You should try to download the latest 7zip and unzip with it.

[Codesquid]

Try to unzip it with Winzip or WinRar > errors see below

Try using 7-zip, it's totally free: http://www.7-zip.org/

[lucasart]

Update: r4638 has been released which addresses the UCI issues found by Lucas.

• UCI: Fix quit command not actually quitting
• UCI: After returning bestmove after a go, the move is not actually applied
• UCI: Implement support for the OwnBook option
• UCI: Basic support for the movetime argument to the go command

You can download the new version from http://octochess.org/

@Moderators: Can you please adjust the topic title and the link in the original posting? I can't seem to edit my old posts.

Thank you Tim. It's working now. I'm using cutechess-cli under linux. Sample session:

Code: Select all

lucas@lucas-desktop:~/Chess$ ./cutechess-cli.sh -each proto=uci tc=6+0.1 timemargin=200 book=./Book.bin bookdepth=8 option.Hash=64 option.OwnBook=false -engine cmd=./Engines/octochess_r4638 -engine cmd=./Engines/doublecheck_3.5.1 option.Verbose=false -concurrency 2 -games 100 -repeat -resign 3 700 -draw 150 100 -pgnout ./octo_dc.pgn

it scores quite decently well (about 40% against DiscoCheck 3.5.1 in 6"+0.1").
running now for my Open Source Bullet rating list. results should be availavlble tomorrow in the tournament foruim.