Winboard Forum

Hi all!

It has been a whole year since TRACE 1.35 was released... wow, time flies!

DRUM ROLL

And now... the version of TRACE which played in the Australian Computer Chess Championships on Aug 20th can be downloaded from my home page.
http://members.optusnet.com.au/~john.boyd

What's new?

Here are the release notes....

TRACE 1.36 NC3 - Aug 21 2006 RELEASE

- This version came =2nd at the Australian Chess Championships 2006.
I had intended for TRACE 2.00 to be entered but was forced due to time constraints to make some improvements instead to 1.35...
- Extends moves that give check and also detects and extends single replies to check
- Transpositional opening book (book.bin). Untuned but surely better than the old book.txt
- Implemented Late Move Reductions popularised by always helpful Tord Romstad.... so now TRACE's average search depth is about a ply or two deeper.
- Improved backward pawn detection
- Passed pawn code reworked to take into account pieces and not just pawns
- Piece value tuning... Q v RR was giving an excessive bonus to the Q...
- Time management improved to use more time in the first half of the game.
- Pruning/reductions disallowed for moves that will probably alter alpha... so TRACE sees more tactics.
- Extensions are more strict... ie. only _safe_ passed pawn moves are extended
- Move legality is tested before makemove... so she is faster now.
- Nullmove reduction is now a mix of R=3 and R=2. Previous versions only used R=2.
- Search blindness due to lazy eval and delta pruning reduced.
- Quiescent move generator additionally generates 'interesting' pawn moves.
- Fixed some more analyze mode i/o lockups under Arena...
- Hard to give an exact figure but my tests show >40 Elo improvement over 1.35
- On an AMD64 Athlon 4000+ TRACE attains >1600000 nodes per second when running the built in benchmark. Type 'bench 13' at the prompt to test this yourself.

Cheers and enjoy!
Ross

Hi Ross,

It seems version 1.36 is not yet uploaded or perhaps
not into the intended 'download' folder. A directory
listing doesn't find it. Thanks anyhow

Best regards,
Guenther

Ross Boyd wrote:Hi all!

It has been a whole year since TRACE 1.35 was released... wow, time flies!

Hello Ross,

It is about time!

What do you think this is, a vacation?

Thanks much kind sir.

Hard to give an exact figure but my tests show >40 Elo improvement over 1.35

Yet another engine that I cannot play with. However I am so weak that the list was small anyways...




Thanks very much Ross.

Do you have a report on your engine's performance?

Later.

Hi Guenther,

Nothing ever works the first time...
Thanks for the 'heads up'
The link is fixed now!

Cheers!
Ross

Hi Roger,

On my web page is a very short summary of this year's NC3. Apparently, there will be a report on the ICGA site so I've been a bit lazy... again...

In a nutshell she performed very nicely till she came unstuck against Chompster on a Quad Opteron. Chompster won that game very deservedly - but clearly TRACE missed her chances.

I'm really curious to see how she goes in the rating lists...

And now its back to work on TRACE 2.... will it ever be finished???

Thanks for your kind comments!

Ross

Ross Boyd wrote:Hi all!

It has been a whole year since TRACE 1.35 was released... wow, time flies!

DRUM ROLL

And now... the version of TRACE which played in the Australian Computer Chess Championships on Aug 20th can be downloaded from my home page.
http://members.optusnet.com.au/~john.boyd

What's new?

Here are the release notes....

TRACE 1.36 NC3 - Aug 21 2006 RELEASE

- This version came =2nd at the Australian Chess Championships 2006.
I had intended for TRACE 2.00 to be entered but was forced due to time constraints to make some improvements instead to 1.35...
- Extends moves that give check and also detects and extends single replies to check
- Transpositional opening book (book.bin). Untuned but surely better than the old book.txt
- Implemented Late Move Reductions popularised by always helpful Tord Romstad.... so now TRACE's average search depth is about a ply or two deeper.
- Improved backward pawn detection
- Passed pawn code reworked to take into account pieces and not just pawns
- Piece value tuning... Q v RR was giving an excessive bonus to the Q...
- Time management improved to use more time in the first half of the game.
- Pruning/reductions disallowed for moves that will probably alter alpha... so TRACE sees more tactics.
- Extensions are more strict... ie. only _safe_ passed pawn moves are extended
- Move legality is tested before makemove... so she is faster now.
- Nullmove reduction is now a mix of R=3 and R=2. Previous versions only used R=2.
- Search blindness due to lazy eval and delta pruning reduced.
- Quiescent move generator additionally generates 'interesting' pawn moves.
- Fixed some more analyze mode i/o lockups under Arena...
- Hard to give an exact figure but my tests show >40 Elo improvement over 1.35
- On an AMD64 Athlon 4000+ TRACE attains >1600000 nodes per second when running the built in benchmark. Type 'bench 13' at the prompt to test this yourself.

Cheers and enjoy!
Ross

Hey Ross,

You mistimed this badly. Or maybe I was in a hurry starting the 3rd division to early...

Thanks anyway. This is not the last season...

Cheers,
G?bor

Hi G?bor,

Yes, not good timing... oh well... 1.35 will have to redeem herself.... after she got herself into all that mess....

Look forward to the next tourney.... regardless I'm enjoying the current one now...

Ross

Ross Boyd wrote:...
- Hard to give an exact figure but my tests show >40 Elo improvement over 1.35...

After all my test give a difference of +11 what is within the confidence intervall.

I let the two versions play a Nunn match under these conditions and got the following ratings:

Code: Select all

Rank Name                                      Elo    +    - games score oppo. draws
   1 Spike 1.2 Turin                          2838   48   45   320   88%  2351   11%
   2 Fruit-Toga 1.2.1                         2800   63   58   180   86%  2353   12%
   3 Fruit 2.1                                2735   47   45   300   82%  2328    9%
   4 Ruffian 2.1.0                            2694   51   49   200   74%  2392   14%
   5 Aristarch 4.50 (c) 2004 Stefan Zipproth  2589   41   40   320   67%  2367   10%
   6 TRACE 1.36 NC3                           2551   47   47   200   63%  2392   18%
   7 TRACE 1.35                               2540   48   48   200   61%  2392   12%
   8 Yace Paderborn                           2537   40   40   320   62%  2370   11%
   9 Pepito v1.59 (Conservador)               2530   40   40   320   61%  2371   13%
  10 The Baron 1.7.0                          2462   40   40   320   55%  2375   11%
  11 Scidlet 3.6                              2316   41   42   320   41%  2384    9%
  12 Natwarlal v0.12                          2176   45   47   320   30%  2393    6%
  13 Prophet-2.0-delta.1                      2134   57   60   200   27%  2392    6%
  14 Small Potato 0.6.1                       2019   48   49   320   21%  2403   10%
  15 Smash-1.0.1                              1968   62   66   200   18%  2392    6%
  16 Prophet 2.0.beta.4                       1858   67   72   200   13%  2392    8%
  17 Needle 0.53.1                            1713   64   73   320    5%  2422    5%


My test setting can be total rubbish of course. But it was a bit breath taking: after position 2 version 1.36 was 100 points ahead. After position 5 it were 38 points. After position 8 it was 10 points behind, to finish with the jutty it has now.

Maybe even 200 games say nothing after all.

Volker