Myrddin 0.95 release

Discussion of anything and everything relating to chess playing software and machines.

Moderator: Ras

JVMerlino
Posts: 1413
Joined: Wed Mar 08, 2006 10:15 pm
Location: San Francisco, California

Myrddin 0.95 release

Post by JVMerlino »

This is to announce the release of Myrddin 0.95. Testing shows that it should be about 40 elo stronger than v0.94, with an estimated CCRL rating of 3210.

You can download the latest version here:
https://github.com/JVMerlino/Myrddin

Changelog:
-- Now using SEE for move ordering of captures, and raised castling moves higher in the move order
-- Now using IIR on pv nodes
-- More aggressive futility pruning
-- Added forced 16-byte alignment of transposition and eval hash structures (thanks to Jim Ablett!)
-- Added counter move heuristic
-- New larger network architecture of 768->1024->32->1
-- Fixed a bug such that fractional time increments, common in hyperbullet games or testing (e.g. 0.1 second), were ignored due to truncation. So an increment of 0.1 second per move would be set to 0.
-- Restored the resign feature, which somehow got deleted in v0.91 (thanks to Martin Sedlak for finding this bug!)

Many thanks to the community for its support and continued fun chats at CCRL tournaments.

jm
Frank Quisinsky
Posts: 7390
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Myrddin 0.95 release

Post by Frank Quisinsky »

Good morning John,

04:30 in Germany.
It's amazing how few quick games Mrddin loses against significantly stronger opponents. It also plays in a very human way. A really nice program, I like it.

Thank you!

Best
Frank
JVMerlino
Posts: 1413
Joined: Wed Mar 08, 2006 10:15 pm
Location: San Francisco, California

Re: Myrddin 0.95 release

Post by JVMerlino »

Thanks, Frank. I very much appreciate everything you do for our community. :D
Frank Quisinsky
Posts: 7390
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Myrddin 0.95 release

Post by Frank Quisinsky »

At first:
Means in my posting before (Myrddin, spelling mistake).

I always lack consistency.
I tend to admire people like Graham Banks. He’s tough as nails and tireless.

It's all because of my other hobbies!
But most spotters (people interested in commercial aviation) already own a chess computer.
Now I just have to get the computer geeks to the airports with their camera gear.

Back, your engine is really interesting.
I will looking more in detail on my notebook next months, on my list.
Also the xb mode runs perfect with the wb2uci adpater.

Such a program need playing strength levels, in special for games vs. humans.
Sure, will be complicated to do it in xb mode.
Frank Quisinsky
Posts: 7390
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Myrddin 0.95 release

Post by Frank Quisinsky »

Hi John,

in around 45 minutes the Junior 13.8.04 Yokohama HCE test-run is ready.
Better late as never :-)
A few years late, means Junior.

Contrary to all the rules...
I have to watch Myrddin 0.95 now. It starts in just under an hour.

Best
Frank

Go Myrddin go ...
Frank Quisinsky
Posts: 7390
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Myrddin 0.95 release

Post by Frank Quisinsky »

John,

if you like you can follow the next 20 hours ...

Results for Myrddin 0.94 NNCe xb in Elo

Code: Select all

   # Player                         :      Elo  Games  Score%   won  draw  lost  Points  Draw%  Error   OppAvg   OppE   OppD
  70 Myrddin 0.94 NNCe xb           :  3043.76   1300    30.7   170   459   671   399.5   35.3  14.64  3225.82   7.94   26.0
 
Much more interesting as boring Elo are the points (all engines have to play vs. the same 26 opponents 50 games = 1.300 games):

Code: Select all

Always based on 1.300 games played!
m-avg  = move-average
1/0<45 = 1:0 / 0:1 under 50 moves
rk     = ranks

64 out of 270 (23,71%), each programmer(s) / et. al. is allowed one rank!

                                         points   1:0 1/2   0:1     %   m-avg     1/0<45       rk
-------------------------------------------------------------------------------------------------
45. Myrddin 0.94 NNCe xb                  399,5   170 459   671   30,73   82    1/ 13  - 12   19#
Myrddin is at the moment: Private First Class ... rank 19
But with 40 Elo more ... Corporal is possible!

Code: Select all

20 ranks:
---------
01# General Field Marshal *****         1.300,0 (100,00%) - 1.144,0 ( 88,00%) ... represents a goal!
02# General of Engines ****             1.143,5 ( 87,97%) - 1.105,0 ( 85,00%) ... perhaps possible in 2026?
03# Lieutenant General ***              1.104,5 ( 84,97%) - 1.066,0 ( 82,00%)
04# Major General **                    1.065,5 ( 81,97%) - 1.027,0 ( 79,00%)
05# Brigadier General *                 1.026,5 ( 78,97%) -   988,0 ( 76,00%)
00# General (Ret.)                                                            ... Honorary title
---
06# Colonel                               987,5 ( 75,97%) -   936,0 ( 72,00%)
07# Lieutenant Colonel                    935,5 ( 71,97%) -   884,0 ( 68,00%)
08# Major                                 883,5 ( 67,97%) -   832,0 ( 64,00%)
09# Captain                               831,5 ( 63,97%) -   780,0 ( 60,00%)
10# First Lieutenant                      779,5 ( 59,97%) -   728,0 ( 56,00%)
11# Second Lieutenant                     727,5 ( 55,97%) -   676,0 ( 52,00%)
---
12# Sergeant Major                        675,5 ( 51,97%) -   637,0 ( 49,00%)
13# Master Sergeant                       636,5 ( 48,97%) -   598,0 ( 46,00%)
14# Sergeant First Class                  597,5 ( 45,97%) -   559,0 ( 43,00%)
15# Staff Sergeant                        558,5 ( 42,97%) -   520,0 ( 40,00%)
16# Sergeant                              519,5 ( 39,97%) -   481,0 ( 37,00%)
17# Corporal                              480,5 ( 36,97%) -   442,0 ( 34,00%)
18# Specialist                            441,5 ( 33,97%) -   403,0 ( 31,00%)
19# Private First Class                   402,5 ( 30,97%) -   364,0 ( 28,00%)
---
20# Private                               363,5 ( 27,97%) -     0,0 (  0,00%)
Here are the live links:

1.
https://www.amateurschach.de/fling/etoc-g_t071.html ...
for Corporal you need 34%, results from 0.94 = 30,73%

2.
https://www.amateurschach.de/fling/etoc-g_t071.sto ...
Shredder's *.sto tournament org. You can see the results and which games are still running

Now ...
Myrddin is about to get a promotion!

I am so crazy!
But don't tell the others!

:-)

On my main page you can find all the other results I have:
https://www.amateurschach.de/main/_engines.htm

And here you can download the games:
https://www.amateurschach.de/main/_etoc-g.htm

Enough, I have to follow again ...
Time for Myrddin 0.95 NNCe xb

Best
Frank
Frank Quisinsky
Posts: 7390
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Myrddin 0.95 release

Post by Frank Quisinsky »

Hi John,

1. The move-average is clearly lesser!
Thats perfect, what I saw = no boring and often computer chess typical endgame chains.
Again, different points you engine is more a human-like playing style.

v0.95

Code: Select all

11. Myrddin 0.95 NNCe xb           (191+,   473=,   636-)  32.9%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         77     1300
           white won         82      115
           black won         90       76
          white draw         85      282
          black draw         85      191
          white loss         72      253
          black loss         66      383
            all wins         85      191     1300      14.7%
           all draws         85      473     1300      36.4%
           all loses         69      636     1300      48.9%
v0.94

Code: Select all

11. Myrddin 0.94 NNCe xb           (170+,   459=,   671-)  30.7%

          CATEGORIES   MOVE_AVE    GAMES    TOTAL  GAME_RATE
           all games         82     1300
           white won         87      103
           black won         93       67
          white draw         87      273
          black draw         93      186
          white loss         77      274
          black loss         74      397
            all wins         89      170     1300      13.1%
           all draws         90      459     1300      35.3%
           all loses         75      671     1300      51.6%
2. More fast won games, but unfortunately, more fast lost games!
Can be a bit random, more games as 1.300 needed.

3. Elo goes higher, in my testing: +22,78

Code: Select all

  70 Myrddin 0.95 NNCe xb           :  3066.75   1300    32.9   191   473   636   427.5   36.4  13.54  3225.92   7.96   26.0
  71 Myrddin 0.94 NNCe xb           :  3043.97   1300    30.7   170   459   671   399.5   35.3  14.97  3225.92   7.96   26.0
4. Now rank 18 = Specialist ... that's a promotion!
Not a Corporal but a Specialist ... a WB Specialist ... again no technical problems.
I think the time-management is also a little bit better with the time-control I am using.

:-)

I think v.095 a bit more speculative after the opening book moves, can be good for more wins, plays more aggressively but my feeling is more ... plays more speculative.

eval is better ... detects early on whether there is a loss or a win, thats good!

I think a strong update you do here!

5. Thats for me interesting for fast looking ...

Code: Select all

                                         points   1:0 1/2   0:1     %   m-avg     1/0<45       rk
-------------------------------------------------------------------------------------------------
45. Myrddin 0.95 NNCe xb                  427,5   191 473   636   32,88   77    5/ 21  - 16   18#
--. Myrddin 0.94 NNCe xb                  399,5   170 459   671   30,73   82    1/ 13  - 12   19#
Best
Frank

PS: I like that so much, looking more in the details.
There are occasional misjudgments, but most of the time the impressions turn out to be accurate.
If not ... I can say ... I am so crazy and to old for all this.
That's good, the perfect excuse, and everyone believes it.
At the moment I am working on new stats to the most impotant pawn structures.
The secrets for get "playing-strength".
What I saw here is the weak point in Myrddin. In the late mid-game your engine exchange bishops / knight without to look on the pawn structures. Often I missed more forced pawn moves, pawn-chain is a small problem, I think more with black pieces as with white pieces. But again, I have not enough to make a clear stat to the very very important pawn-structures.
JVMerlino
Posts: 1413
Joined: Wed Mar 08, 2006 10:15 pm
Location: San Francisco, California

Re: Myrddin 0.95 release

Post by JVMerlino »

Thanks so much for these games and your exciting and unique rating system. :D I really appreciate your efforts.

Many thanks again,
jm
Frank Quisinsky
Posts: 7390
Joined: Wed Nov 18, 2009 7:16 pm
Location: Gutweiler, Germany
Full name: Frank Quisinsky

Re: Myrddin 0.95 release

Post by Frank Quisinsky »

Hi John,

thank you.
The unique rating system starts with a bang if all engines are in my list.
I hope ready end of 7/27.

After I will see tourneys between ...
01. First 41 in a tourney
02. Place 42-83 in a tourney
03. Place 84-125 in a tourney ... 41 engines = 40 x 50 games = 2.000 games per engine, my older FCP-Tourneys.
and so on

So, I have enough time to work in the meantime on my new opening book for the second part of ETOC-G.

And the third and most important part?
The people in white coats can come pick me up then.


I have to prepare that, too. I already have a few chess computers in my emergency kit.
With a Sony Air-7 (spotter, my hobby) and a Casio-DC checkcard as database.

Keep up your good work on Myrddin!

Frank from Gutweiler, Sommerau in the near, 1km.
Sommerau is a country 60 people life with a castle-ruin from 1059 and a lady of the castle who haunts the place.
Maybe later I am married the lady of the castle and go with my chess computers ...

But without a chess computer by Ed Schröder!!
He doesn't like me anymore.
I've somehow managed to annoy him my whole life.

Best
Frank