win-chance for each root move

[xr_a_y]

Why not run some engine testing on this with the first move forced by book ?

[xr_a_y]

ok I'm running something right now, which tool will I use to extract the required statistics (win/draw/loss by first move) from the generated big pgn ?

[xr_a_y]

the dirty way :

Code: Select all

../convert/pgn-extract/pgn-extract test.pgn -w40000 2>&1 | sed 's/^1\. \(...\).*}/\1/g' | grep -v "\[" | grep "1-0\|0-1\|1/2-1/2" | sort | uniq -c

For now I have this (with only like a thousand games ...)

Code: Select all

     34 a3  0-1
     25 a3  1-0
      8 a3  1/2-1/2
     27 a4  0-1
     30 a4  1-0
      9 a4  1/2-1/2
     34 b3  0-1
     24 b3  1-0
      9 b3  1/2-1/2
     26 b4  0-1
     29 b4  1-0
     11 b4  1/2-1/2
     22 c3  0-1
     30 c3  1-0
     14 c3  1/2-1/2
     33 c4  0-1
     20 c4  1-0
     13 c4  1/2-1/2
     33 d3  0-1
     20 d3  1-0
     13 d3  1/2-1/2
     28 d4  0-1
     27 d4  1-0
     11 d4  1/2-1/2
     22 e3  0-1
     36 e3  1-0
      8 e3  1/2-1/2
     31 e4  0-1
     29 e4  1-0
      6 e4  1/2-1/2
     23 f3  0-1
     33 f3  1-0
     10 f3  1/2-1/2
     24 f4  0-1
     33 f4  1-0
      9 f4  1/2-1/2
     25 g3  0-1
     27 g3  1-0
     14 g3  1/2-1/2
     38 g4  0-1
     25 g4  1-0
      3 g4  1/2-1/2
     40 h3  0-1
     15 h3  1-0
     11 h3  1/2-1/2
     37 h4  0-1
     21 h4  1-0
      8 h4  1/2-1/2
     19 Na3 0-1
     35 Na3 1-0
     12 Na3 1/2-1/2
     21 Nc3 0-1
     34 Nc3 1-0
     11 Nc3 1/2-1/2
     16 Nf3 0-1
     38 Nf3 1-0
     12 Nf3 1/2-1/2
     45 Nh3 0-1
     16 Nh3 1-0
      5 Nh3 1/2-1/2

[Ferdy]

Hi,

I'm looking for a list of win-chances per move in the starting position.
Like, historically playing b2-b4 gives you a 18% win-chance and e2-e4 51% (I made up those numbers).

Took the top10-60min.pgn at http://www.fastgm.de/downloads.html, add elo in the game and view it in HCE (Hiarcs Chess Explorer).

[xr_a_y]

Anything played by human is too much biased for the question. Nobody would play stupid move unless he is +400 elo, and thus will win. Only engine/engine games can answer the question.

[xr_a_y]

3700 games played now

Code: Select all

Rank Name                          Elo     +/-   Games   Score   Draws
   1 Fridolin310                    43      15    1851   56.1%   16.2%
   2 GreKo-Linux-64                 39      15    1851   55.6%   14.7%
   3 igel-x64_popcnt               -31      15    1850   45.6%   15.8%
   4 minic_0.55_linux_x64_avx2_bmi2     -52      14    1850   42.6%   17.9%

3701 of 60000 games finished.

Code: Select all

a3 white 33 % black 51 % draw 15 %
a4 white 50 % black 29 % draw 19 %
b3 white 35 % black 46 % draw 17 %
b4 white 39 % black 44 % draw 16 %
c3 white 43 % black 38 % draw 18 %
c4 white 35 % black 47 % draw 16 %
d3 white 29 % black 52 % draw 17 %
d4 white 31 % black 47 % draw 20 %
e3 white 59 % black 26 % draw 14 %
e4 white 47 % black 36 % draw 15 %
f3 white 48 % black 38 % draw 13 %
f4 white 46 % black 37 % draw 16 %
g3 white 36 % black 47 % draw 16 %
g4 white 33 % black 57 % draw 8 %
h3 white 23 % black 63 % draw 13 %
h4 white 32 % black 53 % draw 14 %
Na3 white 44 % black 36 % draw 19 %
Nc3 white 53 % black 31 % draw 15 %
Nf3 white 51 % black 30 % draw 18 %
Nh3 white 24 % black 59 % draw 16 %

Some results seems strange for now, this is of course not yet valid from a statistical point of view I guess ...

[xr_a_y]

Maybe polluted by Minic book ... I run it again from the beginning to ensure no book is used ...

[xr_a_y]

5000 games, no book

Code: Select all

a3 271 games, white 36 % black 50 % draw 12 %
a4 271 games, white 59 % black 20 % draw 20 %
b3 271 games, white 21 % black 57 % draw 21 %
b4 271 games, white 30 % black 50 % draw 18 %
c3 271 games, white 44 % black 39 % draw 15 %
c4 271 games, white 47 % black 36 % draw 15 %
d3 271 games, white 32 % black 46 % draw 20 %
d4 271 games, white 37 % black 41 % draw 20 %
e3 271 games, white 49 % black 37 % draw 13 %
e4 271 games, white 60 % black 24 % draw 15 %
f3 271 games, white 45 % black 39 % draw 15 %
f4 271 games, white 57 % black 22 % draw 19 %
g3 271 games, white 47 % black 37 % draw 15 %
g4 271 games, white 43 % black 43 % draw 12 %
h3 271 games, white 18 % black 60 % draw 21 %
h4 271 games, white 23 % black 61 % draw 15 %
Na3 271 games, white 43 % black 39 % draw 16 %
Nc3 271 games, white 67 % black 21 % draw 11 %
Nf3 270 games, white 52 % black 33 % draw 14 %
Nh3 270 games, white 20 % black 67 % draw 12 %
total games  5418
white wins   2272 41 %
black wins   2254 41 %
draws        892 16 %

[xr_a_y]

there is bias with engine also ... they are trained to play a specific way ...

[xr_a_y]

here's the full pgn for deeper analysis

https://serv.x-ray.fr/chess/test.pgn