Sigma Chess Hiarcs 12 MP?

Get your specific HIARCS/Junior support questions answered here as well as up-to-the-minute news!

Moderators: Watchman, Mark Uniacke, mrudolf

User avatar
Tord Romstad
Member
Posts: 60
Joined: Tue Jul 31, 2007 7:07 pm

Post by Tord Romstad »

Hello Martinus,

Thanks a lot for this very useful feedback!
martinus wrote:1. From 1000 to 1800 rating (the range I tested it playing some 1 minute + 8 second increment games) the program works at 7000 nodes/second. This produces a credible rating in the range 1500-1800 but its too much to play below 1500 rating.
What sort of Mac is this? 7000 nodes/second is slow. On my 2.4 GHz Core 2 Duo, Glaurung does about 60,000 nodes/second at ratings below 2200.
2. At 1000 rating 7000 nodes/second is too much and it looks like playing a 1500-1700 player.
Yes, if the nodes/second slowdown were the only way in which it was crippled, but it isn't. More about this below.
3. In the 1000 to 1200 rating some serious tactical blunders could be expected, kind of random moves, but I didn't saw any. I had to win based on positional knowledge and in tactical maneuvers that go beyond the horizon of the program.
Serious tactical blunders is what I have tried to achieve: Below 2200, the program starts pruning moves randomly. The probability that a move is pruned is not uniform, but depends on the strength (lower strength means that more moves are pruned), and on how hard the move would be to spot for a human player. For instance, the probability that a move is pruned is very big for moves which appear to lose material, and for underpromotions. Long diagonal moves are also pruned quite frequently. On the other end of the spectrum, obvious recaptures are almost never pruned.

Based on what you say, it seems I should increase the probability of blunders much more, especially at the weakest levels. Good to know; that's what I need this form of feedback for. Thanks again! :D
4. I would try to cripple more the engine in the 1000-1200 range to give a player to beginners. I made Glaurung play at 1000 versus Shredder Classic 3 at 1500 and they draw a two games match 1-1.
Testing against Shredder and Hiarcs at weaker levels to calibrate the strength is perhaps not a bad idea. I'll do some experiments when I have some CPU time to spare.
5. I couldn't make the program beep, or make any sound, after the move. There is a way?
Not really, unless you count the "tournament mode" which I have implemented for use when I operate Glaurung myself in tournaments. If you press Control+Option+Shift+T (a deliberately complex key combination, just to prevent users from typing it by accident), Glaurung will beep every time it makes a move, and also change the color of the board to an angry red, and print the text "Please adjust my time if necessary" in light green across the board. Needless to say, this feature isn't meant for end users.

An optional beep when the computer makes a move is trivial to implement, of course. I'll include it in the next version.
6. A bigger and more varied book would be better to play the engine. Small books are suited for engine vs engine matches.
Actually, there are two books included: An engine book, and a GUI book, intended for use with UCI engines without a book of their own. By default, the engine book is used, but the GUI book is bigger and more varied. You can switch to the GUI book in the engine configuration window.
Its nice to see things moving on in the mac chess scene. The Glaurung GUI development is also very good news, since we have few UCI interfaces for mac.
Yes, the current situation is quite sad. There are still no good free UCI compatible GUI for Mac OS X, and even the commercial options leave a lot to be desired. I hope the situation will improve.
Same for the arriving of another chess engine, now able to play games at our human level.
Well, based on your comments above, it seems I still haven't been able to cripple it quite enough. But thanks to your generous advice, I should be able to do better next time. :)

For the moderators: This subthread is beginning to look seriously off-topic for the "Hiarcs Customer Support & Announcements" forum. If possible, perhaps it should be moved to another forum? For instance, to the new "Macintosh chess forum" we would all like to see created? :wink:

Tord
User avatar
martinus
Member
Posts: 118
Joined: Sun Aug 12, 2007 7:49 pm
Location: Portugal

Post by martinus »

Hello Tord

I am using a mac mini intel core duo, 1.66 Ghz, 1.25 Giga Ram.
Later I saw that the guibook.bin is much bigger and varied than the book.bin. It is a nice book indeed.

Now going back to the subject of this thread, if I use Glaurung with Sigma Chess GUI I can make the program access the book.bin, but not the guibook.bin, changing its name didn't worked either.

In the instructions file, I saw that I had to configure the engine to make it use the 2 cores. So nodes/second improved from 7.000 to 12.000 with engine rated at 1000. I tested the progress of nodes/second with one core and it is like this:

1000 to 2100 rating --> 7.000 nodes/second
2200 rating --> 7.800 nodes/second
2300 rating --> 21.000 nodes/second
2400 rating --> 67.000 nodes/second
2500 rating --> 320.000 nodes/second

Now using two cores it start at 12.000 and goes up proportionally. But this mean that the software is now even stronger. I started a game rated at 1320 with time 1 minute + 13 sec. increment and it was like playing a 1700-1900 human with such time in the clock, the program even managed to play solid tactically and like a top amateur positionally.

The GUI needs another important feature, to add to the sound beep after movement, and that is the ability to delete and add games from/to .pgn databases. With more features in the .pgn management it could easily go to a shareware version, there is no reason against it. It could develop in the Sigma Chess path with a freeware and a shareware version.

Another way to calibrate the elo strenght of the program would be using chess engine rating tests. I tend to have special confidence in the LCT II test http://pagesperso-orange.fr/lefouduroi/ ... _2_uci.htm. Problem is this test means a lot of time and work for humans, it is not doable, except if the engine runs the test in automatic mode. As far as I can remember some engines in the past were able to run tests in automatic mode; probably MacChess http://members.aol.com/Macchess/ was one of them. Usually mature tests like the LCTII are already very well calibrated even by humans.

Best Regards.
Stay well.
User avatar
Tord Romstad
Member
Posts: 60
Joined: Tue Jul 31, 2007 7:07 pm

Post by Tord Romstad »

martinus wrote:I am using a mac mini intel core duo, 1.66 Ghz, 1.25 Giga Ram.
Later I saw that the guibook.bin is much bigger and varied than the book.bin. It is a nice book indeed.

Now going back to the subject of this thread, if I use Glaurung with Sigma Chess GUI I can make the program access the book.bin, but not the guibook.bin, changing its name didn't worked either.
That's correct. The book format is different, the Glaurung engine cannot use the GUI book. The Glaurung engine uses the PolyGlot book format. I think it is possible to find lots of different PolyGlot books around on the Internet, or use PolyGlot to build your own books.
In the instructions file, I saw that I had to configure the engine to make it use the 2 cores. So nodes/second improved from 7.000 to 12.000 with engine rated at 1000. I tested the progress of nodes/second with one core and it is like this:

1000 to 2100 rating --> 7.000 nodes/second
2200 rating --> 7.800 nodes/second
2300 rating --> 21.000 nodes/second
2400 rating --> 67.000 nodes/second
2500 rating --> 320.000 nodes/second

Now using two cores it start at 12.000 and goes up proportionally.
Still quite slow -- but since you had to manually set the number of threads to two, I guess you are using Tiger? Unfortunately, the Tiger binaries are a lot slower than the Leopard binaries, especially in 32-bit mode. The 32-bit Leopard binary is about 60% faster than the 32-bit Tiger binary on my computer.

The reason for the big difference is that the Leopard binaries are compiled with the Intel compiler, and the Tiger binaries with Apple's GCC. Unfortunately, I haven't been able to compile a Tiger compatible binary with the Intel compiler. :(
this mean that the software is now even stronger.
True, perhaps I should make an attempt to detect the speed of the CPU and adjust the Elo scale accordingly.
I started a game rated at 1320 with time 1 minute + 13 sec. increment and it was like playing a 1700-1900 human with such time in the clock, the program even managed to play solid tactically and like a top amateur positionally.
OK, I'll try to increase the frequency of blunders, and perhaps also to let the quality of the positional evaluation decrease at lower strength level.
The GUI needs another important feature, to add to the sound beep after movement,
I've added the beep now, but I haven't updated the program on my web site. It's a bit excessive to release new versions every day. I'll keep adding new improvements until I have a more substantial upgrade to offer.
and that is the ability to delete and add games from/to .pgn databases.
You can add games to PGN databases already: Choose "Add Game to File..." from the File menu. Deleting games isn't supported yet, but of course it is planned for a future version. There is a "Delete Game" button in the game list window, but it doesn't work yet.
With more features in the .pgn management it could easily go to a shareware version, there is no reason against it.
There are reasons against it, but they are ideological and practical rather than technical. Glaurung will always remain free, at least the version for desktop computers.
Another way to calibrate the elo strenght of the program would be using chess engine rating tests. I tend to have special confidence in the LCT II test http://pagesperso-orange.fr/lefouduroi/ ... _2_uci.htm. Problem is this test means a lot of time and work for humans, it is not doable, except if the engine runs the test in automatic mode. As far as I can remember some engines in the past were able to run tests in automatic mode; probably MacChess http://members.aol.com/Macchess/ was one of them. Usually mature tests like the LCTII are already very well calibrated even by humans.
No problem, I have plenty of tools for running EPD tests automatically.

Tord
User avatar
martinus
Member
Posts: 118
Joined: Sun Aug 12, 2007 7:49 pm
Location: Portugal

Post by martinus »

Yes, I forgot that: running Tiger 10.4.11
Until now I still couldn't managed the GUI to add games to databases, and I tested that command. Lets see in a next version.

Regards.
User avatar
Dark Horse
Hiarcs Team Captain
Posts: 3673
Joined: Sun May 11, 2008 4:25 pm
Location: from a cubicle in Bangalore

Post by Dark Horse »

Tord Romstad wrote:
I think it is possible to find lots of different PolyGlot books around on the Internet, or use PolyGlot to build your own books.
Hi, I am interested to know how to make an Opening Book, any tips for a noob ? thanks
Post Reply