A complete 2000 lines of code engine

[xr_a_y]

Minic 1.35 is out. This release has a chance to get over 3000 in the CCRL Blitz 4CPU list

Indeed !

http://ccrl.chessdom.com/ccrl/404/cgi/e ... 4-bit_4CPU

So cool

Thanks for testing

[xr_a_y]

Recent results

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 Defenchess_2.2                313      26     870   85.9%   19.8%
   2 demolito                       86      20     869   62.1%   28.8%
   3 minic_dev                     -20      19     870   47.1%   35.7%
   4 Winter0.7                     -25      19     870   46.4%   30.0%
   5 Hakkapeliitta                 -36      20     870   44.9%   27.2%
   6 rodentIII                     -39      19     870   44.4%   32.1%
   7 minic_1.37                    -45      18     870   43.6%   37.8%
   8 amoeba                        -45      19     870   43.6%   29.3%
   9 minic_1.35                    -45      18     869   43.5%   36.8%
  10 PeSTO_bmi2                    -81      20     870   38.5%   24.6%

4349 of 45000 games finished.

[xr_a_y]

Not much progress recently

40/20sec TC, 256 Mb TT

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 rodentIII                      28       9    4000   54.1%   34.4%
   2 minic_1.38                     10       8    4000   51.4%   38.5%
   3 Winter0.7                       2       9    4000   50.3%   28.9%
   4 minic_1.35                     -2       8    4000   49.8%   39.9%
   5 PeSTO_bmi2                    -39       9    4000   44.4%   26.6%

10000 of 10000 games finished.

[CMCanavessi]

Not much progress is still progress

[xr_a_y]

Minic 1.39 is available with at probably +20 Elo versus 1.35
thanks to
- pruning tuning based on king danger eval
- "initiative" eval (formula from Xiphos)

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 igel-last                      66      16    1135   59.4%   38.9%
   2 Winter0.7                      36      16    1135   55.1%   38.0%
   3 minic_1.39                     10      15    1134   51.4%   47.1%
   4 PeSTO_bmi2                      7      16    1134   51.0%   36.4%
   5 rodentIII                       1      16    1134   50.1%   40.7%
   6 amoeba                        -10      16    1134   48.5%   38.7%
   7 minic_1.35                    -27      15    1135   46.1%   44.2%
   8 minic_1.19                    -38      15    1135   44.5%   44.1%
   9 Topple_master                 -43      16    1134   43.8%   38.5%

5105 of 36000 games finished.

[voffka]

Oh my, I see Igel in your benchmarks, I need to hurry up and throw 200 more elo ...

[xr_a_y]

Oh my, I see Igel in your benchmarks, I need to hurry up and throw 200 more elo ...

Well, as usual, this is short TC, and Igel has made a very big step recently in this domain too. Don't hurry too much, let me fill the gap

[xr_a_y]

Minic 1.42 is available.

This is a time management bug fix release for incremental TC like 2min+1sec. In this case, this is a +30Elo patch. I don't know if Minic was already tested at new CCRL blitz TC, but this can't be good, this release is fixing that.

In other cases, like 40/25, this release is not much better (so absolutely no need to stop 1.39 test currently running).

[xr_a_y]

Minic 1.44 is available with probably +15Elo more.

And much better than 1.35 at sudden death TC

Code: Select all

Score of minic_1.44 vs minic_1.35: 70 - 34 - 151 [0.571]
Elo difference: 49.4 +/- 27.1, LOS: 100.0 %, DrawRatio: 59.2 %

[Alayan]

- pruning tuning based on king danger eval

Could you expand on what you are doing there ?

Your code source is available for a look, but an explanation of your thought process, what you tried, etc. is a useful complement to raw code.