final testing for McBrain 2.2 underway...

[MikeB]

McBrain is a Stockfish engine and solely exists because of the Stockfish team and its contributors, without them, there would be no McBrain...

as before it will run Thomas Zipproth's Cerebellum book ( fantastic book, it's worth about 30 ELO by itself ) and will feature play by ELO and other goodies, such as a UCI option to turn on most of Joe Ellis' Matefiner code , as well as a "brute force" options ( tunrs off all reductions and null move) and naturally will be based on the latest SF code and some very nice patches which the SF team will commit to in some form in the near future I am sure
and one little bit of code that will skinny down the tree incrementally as you increase the number of threads for your multi-core machine - my testing shows 10 or more ELO gain when using 8 cores or more than standard Stockfish.

Code: Select all

6892 run scripts were prepared.
04/28 01:07
found 1 hosts to use
configuration used:
hostname                     #cpus
local                          10
60 of 6000 games completed...
time control: 3000+50
Date: 04/28/17 : 01:24:10
125 game(s) loaded
Rank Name                     Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 McBrain v2.2l 64 POPCNT   3104   0.0   25   25    84   43.0  51.2   12   10   62  14.3  73.8  3098 
   2 McBrain v2.2m 64 POPCNT   3100   3.9   26   26    83   41.5  50.0   14   14   55  16.9  66.3  3100 
   3 Stockfish 8 64 POPCNT     3095   5.3   26   26    83   40.5  48.8   12   14   57  14.5  68.7  3102 
---------------------------------------------------------------------------------------------------------
125 game(s) loaded

the code is locked down, this will only decide if a UCI option is true or false by default...

my guess is that McBrain should test about 20 ELO or more higher than Stockfish 8 ( likewise current development SF would also test about 20 ELO or more higher than SF 8 I believe)

follow the results here live ( updates about every 120 games or so , about 20 minutes) : https://www.dropbox.com/s/6qd33ywphjyec ... a.txt?dl=1

pgn file: https://www.dropbox.com/s/an1vp8zrxwrrnab/all.pgn?dl=1

more results as I was typing this:

(note: it looks like I have a counter issue, must have fat fingered something in the script )

Code: Select all

246 game(s) loaded
Rank Name                     Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 McBrain v2.2l 64 POPCNT   3115   0.0   16   16   164   88.0  53.7   27   15  122  16.5  74.4  3092 
   2 McBrain v2.2m 64 POPCNT   3104  11.4   17   17   163   82.5  50.6   25   23  115  15.3  70.6  3098 
   3 Stockfish 8 64 POPCNT     3081  22.7   16   16   165   75.5  45.8   18   32  115  10.9  69.7  3110 
---------------------------------------------------------------------------------------------------------

v2.2m has the Mate Finder code set to true and typically that will test out about 10 or more ELO lower with it set to true.

I have also reduced and simplified the UCI options, eliminating some of the more technical ones relating to controlling search (lmr depth, pruning, probcut etc).

Should be released on or about May 1st with an announcement here.

[maxdeg]

Thanks - looking forward to it

[MikeB]

due to a fat finger, ran more games than desired...

Code: Select all

7602 game(s) loaded
Rank Name                     Rating   Δ     +    -     #     Σ    Σ%     W    L    D   W%    =%   OppR 
---------------------------------------------------------------------------------------------------------
   1 McBrain v2.2l 64 POPCNT   3110   0.0    3    3  5068 2648.0  52.2  905  677 3486  17.9  68.8  3095 
   2 McBrain v2.2m 64 POPCNT   3105   4.3    3    3  5068 2596.5  51.2  871  746 3451  17.2  68.1  3097 
   3 Stockfish 8 64 POPCNT     3085  20.7    3    3  5068 2357.5  46.5  641  994 3433  12.6  67.7  3108 
---------------------------------------------------------------------------------------------------------

so one can see that the SF development team has made good progress since the release of SF 8.

Just putting some final ( non functional ) touches on the code now to prepare for a release ..

I will make the matefinder code active ( true) as the default, since most people prefer the analysis with matefinder code than worry about 4 ELO, but if you do want the strongest engine in game play, the recommendation is to uncheck the Matefinder option in the UCI options ( I am calling it Matefinder code since it is just about 100% Matefinder code by Joe Ellis, I'm just making it available an an option as opposed to a second engine to load).