Plumbing The Depths (OBender finishes strongly to win)

[Graham Banks]

PLUMBING THE DEPTHS

Intel Core 2 Duo E4400 @2GHz overclocked to 2.66GHz
2GB total RAM
Arena 1.1 GUI
HS-Mainbook.abk (v705)
256 MB hash for each
3-4-5 piece EGTBs
40/30 (adapted to CCRL 40/40)
6 cycles (54 rounds)
Run by Charles Smith


Final Standings

38.5 - OBender 2.11.5
34.5 - Adam 3.1
33.5 - Buzz 0.07
32.5 - Gibbon 2.01b
32.0 - Feuerstein 0.4.51
28.5 - BigLion 2.23w
20.5 - Clueless 1.4
20.0 - Micro-Max 4.8
16.5 - Marvin 1.3.0
13.5 - Smash 1.0.3


The complete tournament pgn (zipped) can be downloaded here:
http://kirr.homeunix.org/chess/discussi ... hp?id=6982


Notes
OBender would perform better with decent time management. It's time management is apalling - uses far too little of its time. This problem is still apparent in the latest version.
Buzz 0.07 has a bug that sees it throw away completely won games with a one move howler. This seems to have been fixed in Buzz 0.08.
Smash has a bug similar to that of Buzz 0.07.
No problems with the other engines - all reliable and stable.
Micro-Max is truly a marvellous achievement considering it has about 2000 lines of code!

[hgm]

Micro-Max is truly a marvellous achievement considering it has about 2000 lines of code!

That is 2000 characters of code. (1961, to be exact.)

Lines is only 120 or so (not counting empty lines or lines only containing comment). Of course lines do not really mean anything in a C program, as you can make a line arbitrarily long. I adopted the style of having each line contain only little code, so that the last 40% or so is available for commenting on the purpose of that code.

I had the source listing of uMax with me at last WCCC (two sheets of A4 paper). Stefan Mayer-Kahlen remarked that the length of the listing was just about the same size as the height of the stack of paper containing the printout of Shredder.


For the record: This character / line count does not include the Winboard interface, but pertains to the non-winboard-compatible stand-alone versions. I provide the Winboard interface only as a means for allowing people to try out how uMax plays, in a mor user-friendly manner. I simply copied the Winboard interface from Joker, and replaced the main program of (containing the I/O functions) by this.

I hope to replace this in the future by an adapter (like Polyglot) that would allow you to play the stand-alone versions themselves.

I like your logo for it, by the way.

[hgm]

OBender would perform better with decent time management. It's time management is apalling - uses far too little of its time. This problem is still apparent in the latest version.

Actually you are lucky. For most time controls OBender forfeits every game on time as it often uses 3 or 4 times as much as it is supposed to. At 40/2' it still continues to think 15 sec per move if its flag is already down for 2 min (if you don't play with "autocall flag").

About 30% of the engines I tested in the 2200 range seem to be unable to correctly manage the time in combination with an external book (using the 'force' command for the first few moves).

They fail to count the forced moves. So if you force 7 moves at 40/40', they aim to play upto 47 in 40'. The new 40 min they get at move 40 comes to them as a surprise, and they will use it up entirely for the next 7 moves.

That they then don't get another 40' new time at move 47 then comes as a (very unpleasant) surprise. They are convinced that the next chance for extra time will come only at move 87, so they will now try to play 40 moves in the time they left as a safety margin. Depending on how aggressively they try to use all their time, there might be less than 1 min left at that point (e.g. Lime). In engines that leave somewhat more margin (e.g. Averno) they might play the next 33 moves in ~7 min. So alternately they play 7 moves in 35 min and then 33 in 5 min. Only the first 40 min behave more or less normal (although they keep unreasonable spare time for moves 41-47).

With more forced moves it becomes less extreme, although the moves upto 40 suffer more from it. E.g. a long opening line that would force 25 opening moves would allow 40' for the next 15 moves, while the engine would think it has only 1'/move. This is still the normal time you would have per move without opening book, though. But then it will use its second time quota for move 41-65, and will have to play 66-80 in almost no time. But usually the game is decided by then, so it doesn't hurt so much.

[Richard Allbert]

Hi,

I'm pretty sure Lime does account for forced moves from an external book...

Code: Select all

if (!strncasecmp(command, "usermove", 8))
            {
              ptr += 9;
              readinmove(ptr);
              //in the case of a mps time control, decrement the moves to
              //go for the opponent - used for ponder time allocation
              if(s_param->movestogo[p->side^1] != -1)
              {
               s_param->movestogo[p->side^1]--;
               if(s_param->movestogo[p->side^1]==0) s_param->movestogo[p->side^1] = mps;
              }
              continue;
            }

It could of course be buggy

[Jim Ablett]

Obender 3.1.0 Intel build

Contains 2 compiles, 'Intel® Core™ Duo' and 'All processors'
Both a lot faster than original.

Also compiled with slightly different time increment code
(doesn't use windows 'gettickcount()' It may have better time
management now, though I haven't tested enough to be sure.

http://www.orbitfiles.com/download/id1975836609.html

Jim.

[Jim Ablett]

Mirrors:

http://www.mydatabus.com/public/deckard ... _win32.zip

or

http://w13.easy-share.com/6058241.html

Jim.

[Graham Banks]

Obender 3.1.0 Intel build

Contains 2 compiles, 'Intel® Core™ Duo' and 'All processors'
Both a lot faster than original.

Also compiled with slightly different time increment code
(doesn't use windows 'gettickcount()' It may have better time
management now, though I haven't tested enough to be sure.

http://www.orbitfiles.com/download/id1975836609.html

Jim.

Thanks Jim. I'll try it out soon!

Regards, Graham.

[Jim Ablett]

Hi Graham,

I noticed 'Obender' still having time problems in WBEC.

Here's a new version I've tinkered with.

Time management should be better and 39th move timeout
bug should be fixed. I also noticed 'Obender' doesn't send mate result
which is needed in Winboard to prevent a hanging game if the other engine
doesn't send it either. I've fixed this too.

New build 'Obender 3.1.0 JA' Intel compiler 10 PGO.

http://www.orbitfiles.com/download/id2006398132.html

Rgds,
Jim.

[Graham Banks]

Hi Graham,

I noticed 'Obender' still having time problems in WBEC.

Here's a new version I've tinkered with.

Time management should be better and 39th move timeout
bug should be fixed. I also noticed 'Obender' doesn't send mate result
which is needed in Winboard to prevent a hanging game if the other engine
doesn't send it either. I've fixed this too.

New build 'Obender 3.1.0 JA' Intel compiler 10 PGO.

http://www.orbitfiles.com/download/id2006398132.html

Rgds,
Jim.

Thanks Jim.

[Tony Thomas]

You are lucky, I couldnt use obender under my favorite time control. I have to try Jim's build after my CPU over heating problem is taken care of.