Chances of survival of individual chess pieces....

[Dr. Axel Schumacher]

Hi all,
I found this on the web, which I think is quite interesting:

http://www.quora.com/What-are-the-chanc ... rage-games



This result is based on the Million Base 2.2 (2.2m master-level tournament games) updated to January 2013 and available from http://www.top-5000.nl/pgn.htm.

In theory, could it be useful in game strategy? Or in programming a chess engine? What do you think?

White's d2 pawn is almost always busted, also Black's g8 Knight.

I asked the author and he made his code available here:

https://github.com/ojb500/SurvivingPieces

Would be interesting to calculate the survival or specific openings.

Cheers,
Axel

[Guenther]

The biggest difference is calculated for the c-pawn and then the d-pawn. I bet this is because bazillions of sicilians were played in the database.

Guenther

[Dr. Axel Schumacher]

The biggest difference is calculated for the c-pawn and then the d-pawn. I bet this is because bazillions of sicilians were played in the database.

Guenther

This is absolutely likely; at least (naturally) we do not have an equal distribution of openings. As such, as mentioned before, it would be good to have those tables for different openings.

Here the overall survival-difference in relation to the opposite piece:


Nevertheless interesting that White's central pawns are more easily lost.
The question is, would it be possible to use this information in strategic planning and for evaluation purposes. For example:
1.) Since at the beginning of the game, the likelihood of loosing White's d-pawn is high, should it not have an average pawn value of 1?
or
2.) Since White's f-Bishop has higher survival chances (and hence longer time to have an impact), shouldn't it be valued higher (e.g. 3.1) in comparison with the c-Bishop (value e.g. 2.9)?

A.

[Dr. Axel Schumacher]

The biggest difference is calculated for the c-pawn and then the d-pawn. I bet this is because bazillions of sicilians were played in the database.

Guenther

This is absolutely likely; at least (naturally) we do not have an equal distribution of openings. As such, as mentioned before, it would be good to have those tables for different openings.

Here the overall survival-difference in relation to the opposite piece:


Nevertheless interesting that White's central pawns are more easily lost.
The question is, would it be possible to use this information in strategic planning and for evaluation purposes. For example:
1.) Since at the beginning of the game, the likelihood of loosing White's d-pawn is high, should it not have an average pawn value of 1?
or
2.) Since White's f-Bishop has higher survival chances (and hence longer time to have an impact), shouldn't it be valued higher (e.g. 3.1) in comparison with the c-Bishop (value e.g. 2.9)?

A.

The asymmetry is even stronger if we do not compare bishops in the same rank, but opposing colored bishops:



Should we trade White's c-Bishop or Black's f-Bishop?

A.

[Vinvin]

Here the overall survival-difference in relation to the opposite piece:

The low score for the c7 pawn is likely because of all Sicilian openings c7-c5-cxd4. That probably the reason why the d2 pawn disappear often too ..

[carldaman]

The biggest difference is calculated for the c-pawn and then the d-pawn. I bet this is because bazillions of sicilians were played in the database.

Guenther

This is absolutely likely; at least (naturally) we do not have an equal distribution of openings. As such, as mentioned before, it would be good to have those tables for different openings.

Here the overall survival-difference in relation to the opposite piece:


Nevertheless interesting that White's central pawns are more easily lost.
The question is, would it be possible to use this information in strategic planning and for evaluation purposes. For example:
1.) Since at the beginning of the game, the likelihood of loosing White's d-pawn is high, should it not have an average pawn value of 1?
or
2.) Since White's f-Bishop has higher survival chances (and hence longer time to have an impact), shouldn't it be valued higher (e.g. 3.1) in comparison with the c-Bishop (value e.g. 2.9)?

A.

Hi,

Re: question #2, I think that generally the King's Bishop (or f-Bishop, if you prefer) is regarded as the more important Bishop, especially as an attacking piece for White, and a defensive one for Black. [There are exceptions, of course].


Should we trade White's c-Bishop or Black's f-Bishop?

The answer to your question from your other post should probably be 'Yes'. White's dark-squared 'c-' Bishop is often one of the pieces White seeks to trade off, either for the Nf6 or the opposing dark-squared Bishop.

Regards,
CL

[Dr. Axel Schumacher]

The biggest difference is calculated for the c-pawn and then the d-pawn. I bet this is because bazillions of sicilians were played in the database.

Guenther

This is absolutely likely; at least (naturally) we do not have an equal distribution of openings. As such, as mentioned before, it would be good to have those tables for different openings.

Here the overall survival-difference in relation to the opposite piece:


Nevertheless interesting that White's central pawns are more easily lost.
The question is, would it be possible to use this information in strategic planning and for evaluation purposes. For example:
1.) Since at the beginning of the game, the likelihood of loosing White's d-pawn is high, should it not have an average pawn value of 1?
or
2.) Since White's f-Bishop has higher survival chances (and hence longer time to have an impact), shouldn't it be valued higher (e.g. 3.1) in comparison with the c-Bishop (value e.g. 2.9)?

A.

Hi,

Re: question #2, I think that generally the King's Bishop (or f-Bishop, if you prefer) is regarded as the more important Bishop, especially as an attacking piece for White, and a defensive one for Black. [There are exceptions, of course].


Should we trade White's c-Bishop or Black's f-Bishop?

The answer to your question from your other post should probably be 'Yes'. White's dark-squared 'c-' Bishop is often one of the pieces White seeks to trade off, either for the Nf6 or the opposing dark-squared Bishop.

Regards,
CL

White usually should exchange the dark-squared bishop in the Dutch defense, thereby taking control of the dark squares, and it is often an ideal piece to sacrifice in a king-side attack. Also, an exchange of the Queen's bishop for the f6-knight in the English opening is not unusual.

The question is still if you should value it less; in the end you get something or exchanging the bishop.
Maybe a chess engine should be aware that some pieces have higher chances to be lost anyway. It could then decide to exchange it early in the game. On the other hand it may indicate weak spots to attack (or defend stronger) in certain openings.

Regards,
A.

[Ajedrecista]

Hello Axel:

I find this topic very interesting. I coded a clumsy parser more than a year ago (link here) in a try to search the frequency of promotions and underpromotions by file (a, b, ..., g, h). The results of that thread are a little off since my programming skills are very poor and that parser would not read correctly more than a promotion per line... but promotions are somewhat uncommon, so these results give a reasonable idea on what is happening... without expecting exact results, of course!

I posted the first version of the code at the original post but I later added more code until I got a programme that is easy to use (however, it is not public but I do not have problems in make it public). I analyzed some huge PGNs from CCRL and CEGT, reaching the conclusion of that wing pawns promoted more frequently than central pawns. I am aware that my parser only reads the squares of promotion: for example, a white pawn that promotes in a8 does not mean that started on a2. But knowing that, I found curious this V-shaped graphic of frequency of promotions by file. I obtained the result of more promotions on a1/a8 than h1/h8, probably due to the higher frequency of O-O instead of O-O-O.

I know that chances of survival are not promotions, but could be weakly correlated, and I wanted to note it. I have not download Million Base 2.2 from Ed's web because it is an .exe instead of a compressed PGN with ZIP, RAR or 7z. If not, I would use my tool for bring some results.

Last but not least, thank you very much to the programmer that did such task.

Regards from Spain.

Ajedrecista.

[Dr. Axel Schumacher]

Hello Axel:

I find this topic very interesting. I coded a clumsy parser more than a year ago (link here) in a try to search the frequency of promotions and underpromotions by file (a, b, ..., g, h). The results of that thread are a little off since my programming skills are very poor and that parser would not read correctly more than a promotion per line... but promotions are somewhat uncommon, so these results give a reasonable idea on what is happening... without expecting exact results, of course!

I posted the first version of the code at the original post but I later added more code until I got a programme that is easy to use (however, it is not public but I do not have problems in make it public). I analyzed some huge PGNs from CCRL and CEGT, reaching the conclusion of that wing pawns promoted more frequently than central pawns. I am aware that my parser only reads the squares of promotion: for example, a white pawn that promotes in a8 does not mean that started on a2. But knowing that, I found curious this V-shaped graphic of frequency of promotions by file. I obtained the result of more promotions on a1/a8 than h1/h8, probably due to the higher frequency of O-O instead of O-O-O.

I know that chances of survival are not promotions, but could be weakly correlated, and I wanted to note it. I have not download Million Base 2.2 from Ed's web because it is an .exe instead of a compressed PGN with ZIP, RAR or 7z. If not, I would use my tool for bring some results.

Last but not least, thank you very much to the programmer that did such task.

Regards from Spain.

Ajedrecista.

Thanks or your interesting comment. It would be nice to correlate the data. Also, it seems you looked "only" at engine tournaments. Can you run the code over a human database?

Cheers,
Axel

[Ajedrecista]

Hello Axel:

Thanks or your interesting comment. It would be nice to correlate the data. Also, it seems you looked "only" at engine tournaments. Can you run the code over a human database?

Cheers,
Axel

I downloaded 28298 games of KID from this excellent site (file KIDOther7.pgn) because it was the PGN with more games. A good thing could be download all the PGNs of the openings of that site and merge them into a single PGN (for example from command prompt, using copy or type like copy *.pgn PGN_Mentor_openings_database.pgn). Anyway, here is the total output for KIDOther7.pgn:

Code: Select all

Search_promotions_and_underpromotions_by_crowning_squares, ® 2013.

------------------------------------------------------------
Tool that provides some info about promotions in a PGN file.
------------------------------------------------------------

Write down the full name of the PGN file (up to 64 characters), including .pgn:

KIDOther7.pgn

Write down the clock rate of the CPU (in GHz), only for timing the elapsed time of the calculations:

3

Searching for games and results...

Games:       28298      Estimated remaining time:    66.4 seconds.
  1-0:       10935      Estimated remaining time:    65.5 seconds.
  0-1:        7939      Estimated remaining time:    62.7 seconds.
  1/2-1/2:    9421      Estimated remaining time:    62.9 seconds.
  Unknown:       3
______________________________

White score:  55.29%
______________________________

White advantage:    36.93 Elo.
______________________________

Searching promotions and underpromotions...

 1/64  a1=Q      83     Estimated remaining time:    61.3 seconds.
 2/64  a1=R       0     Estimated remaining time:    60.1 seconds.
 3/64  a1=B       0     Estimated remaining time:    58.9 seconds.
 4/64  a1=N       0     Estimated remaining time:    57.8 seconds.
 5/64  b1=Q      67     Estimated remaining time:    56.9 seconds.
 6/64  b1=R       0     Estimated remaining time:    56.1 seconds.
 7/64  b1=B       0     Estimated remaining time:    55.1 seconds.
 8/64  b1=N       0     Estimated remaining time:    54.1 seconds.
 9/64  c1=Q      69     Estimated remaining time:    53.2 seconds.
10/64  c1=R       1     Estimated remaining time:    52.4 seconds.
11/64  c1=B       0     Estimated remaining time:    51.5 seconds.
12/64  c1=N       0     Estimated remaining time:    50.7 seconds.
13/64  d1=Q      79     Estimated remaining time:    50.0 seconds.
14/64  d1=R       0     Estimated remaining time:    49.3 seconds.
15/64  d1=B       1     Estimated remaining time:    48.5 seconds.
16/64  d1=N       2     Estimated remaining time:    47.7 seconds.
17/64  e1=Q      72     Estimated remaining time:    46.9 seconds.
18/64  e1=R       1     Estimated remaining time:    46.2 seconds.
19/64  e1=B       0     Estimated remaining time:    45.4 seconds.
20/64  e1=N       2     Estimated remaining time:    44.8 seconds.
21/64  f1=Q      95     Estimated remaining time:    43.9 seconds.
22/64  f1=R       2     Estimated remaining time:    43.1 seconds.
23/64  f1=B       0     Estimated remaining time:    42.4 seconds.
24/64  f1=N       1     Estimated remaining time:    41.6 seconds.
25/64  g1=Q      69     Estimated remaining time:    40.8 seconds.
26/64  g1=R       0     Estimated remaining time:    40.0 seconds.
27/64  g1=B       0     Estimated remaining time:    39.2 seconds.
28/64  g1=N       1     Estimated remaining time:    38.5 seconds.
29/64  h1=Q      70     Estimated remaining time:    37.7 seconds.
30/64  h1=R       1     Estimated remaining time:    36.9 seconds.
31/64  h1=B       1     Estimated remaining time:    36.2 seconds.
32/64  h1=N       2     Estimated remaining time:    35.4 seconds.
33/64  a8=Q     125     Estimated remaining time:    34.6 seconds.
34/64  a8=R       1     Estimated remaining time:    33.9 seconds.
35/64  a8=B       0     Estimated remaining time:    33.1 seconds.
36/64  a8=N       1     Estimated remaining time:    32.4 seconds.
37/64  b8=Q     122     Estimated remaining time:    31.6 seconds.
38/64  b8=R       0     Estimated remaining time:    30.9 seconds.
39/64  b8=B       0     Estimated remaining time:    30.1 seconds.
40/64  b8=N       1     Estimated remaining time:    29.3 seconds.
41/64  c8=Q     188     Estimated remaining time:    28.6 seconds.
42/64  c8=R       1     Estimated remaining time:    27.9 seconds.
43/64  c8=B       0     Estimated remaining time:    27.1 seconds.
44/64  c8=N       1     Estimated remaining time:    26.4 seconds.
45/64  d8=Q     185     Estimated remaining time:    25.7 seconds.
46/64  d8=R       2     Estimated remaining time:    24.9 seconds.
47/64  d8=B       0     Estimated remaining time:    24.2 seconds.
48/64  d8=N       4     Estimated remaining time:    23.5 seconds.
49/64  e8=Q      78     Estimated remaining time:    22.7 seconds.
50/64  e8=R       0     Estimated remaining time:    22.0 seconds.
51/64  e8=B       0     Estimated remaining time:    21.3 seconds.
52/64  e8=N       3     Estimated remaining time:    20.5 seconds.
53/64  f8=Q      59     Estimated remaining time:    19.8 seconds.
54/64  f8=R       1     Estimated remaining time:    19.0 seconds.
55/64  f8=B       0     Estimated remaining time:    18.3 seconds.
56/64  f8=N       1     Estimated remaining time:    17.6 seconds.
57/64  g8=Q      56     Estimated remaining time:    16.8 seconds.
58/64  g8=R       0     Estimated remaining time:    16.1 seconds.
59/64  g8=B       0     Estimated remaining time:    15.3 seconds.
60/64  g8=N       1     Estimated remaining time:    14.6 seconds.
61/64  h8=Q      53     Estimated remaining time:    13.9 seconds.
62/64  h8=R       0     Estimated remaining time:    13.1 seconds.
63/64  h8=B       0     Estimated remaining time:    12.4 seconds.
64/64  h8=N       0     Estimated remaining time:    11.7 seconds.

Promotions/underpromotions to:
  Q:   1470
  R:     10
  B:      2
  N:     20

Searching promotions/underpromotions that give check...

1/8   *1=Q+     192     Estimated remaining time:    11.0 seconds.
2/8   *1=R+       1     Estimated remaining time:    10.2 seconds.
3/8   *1=B+       2     Estimated remaining time:     9.5 seconds.
4/8   *1=N+       4     Estimated remaining time:     8.8 seconds.
5/8   *8=Q+     245     Estimated remaining time:     8.0 seconds.
6/8   *8=R+       3     Estimated remaining time:     7.3 seconds.
7/8   *8=B+       0     Estimated remaining time:     6.6 seconds.
8/8   *8=N+       8     Estimated remaining time:     5.8 seconds.

Searching promotions/underpromotions that give checkmate...

1/8   *1=Q#       0     Estimated remaining time:     5.1 seconds.
2/8   *1=R#       0     Estimated remaining time:     4.4 seconds.
3/8   *1=B#       0     Estimated remaining time:     3.7 seconds.
4/8   *1=N#       0     Estimated remaining time:     2.9 seconds.
5/8   *8=Q#       0     Estimated remaining time:     2.2 seconds.
6/8   *8=R#       0     Estimated remaining time:     1.5 seconds.
7/8   *8=B#       0     Estimated remaining time:     0.7 seconds.
8/8   *8=N#       0

Number of lines of this PGN file:     513535

Approximated elapsed time:    61.5 seconds.

The obtained results will be saved into Promotions_stats_KIDOther7.pgn.txt file.

The results were successfully saved into Promotions_stats_KIDOther7.pgn.txt file.

Thanks for using this tool. Press enter to exit.

Code: Select all

KIDOther7.pgn
 
a1=Q:     83
a1=R:      0
a1=B:      0
a1=N:      0
 
b1=Q:     67
b1=R:      0
b1=B:      0
b1=N:      0
 
c1=Q:     69
c1=R:      1
c1=B:      0
c1=N:      0
 
d1=Q:     79
d1=R:      0
d1=B:      1
d1=N:      2
 
e1=Q:     72
e1=R:      1
e1=B:      0
e1=N:      2
 
f1=Q:     95
f1=R:      2
f1=B:      0
f1=N:      1
 
g1=Q:     69
g1=R:      0
g1=B:      0
g1=N:      1
 
h1=Q:     70
h1=R:      1
h1=B:      1
h1=N:      2
 
a8=Q:    125
a8=R:      1
a8=B:      0
a8=N:      1
 
b8=Q:    122
b8=R:      0
b8=B:      0
b8=N:      1
 
c8=Q:    188
c8=R:      1
c8=B:      0
c8=N:      1
 
d8=Q:    185
d8=R:      2
d8=B:      0
d8=N:      4
 
e8=Q:     78
e8=R:      0
e8=B:      0
e8=N:      3
 
f8=Q:     59
f8=R:      1
f8=B:      0
f8=N:      1
 
g8=Q:     56
g8=R:      0
g8=B:      0
g8=N:      1
 
h8=Q:     53
h8=R:      0
h8=B:      0
h8=N:      0
 
=======================
 
a1=*:      83 ( 13.41%)
b1=*:      67 ( 10.82%)
c1=*:      70 ( 11.31%)
d1=*:      82 ( 13.25%)
e1=*:      75 ( 12.12%)
f1=*:      98 ( 15.83%)
g1=*:      70 ( 11.31%)
h1=*:      74 ( 11.95%)
 SUM:     619
 
-----------------------
 
a8=*:     127 ( 14.38%)
b8=*:     123 ( 13.93%)
c8=*:     190 ( 21.52%)
d8=*:     191 ( 21.63%)
e1=*:      81 (  9.17%)
f8=*:      61 (  6.91%)
g8=*:      57 (  6.46%)
h8=*:      53 (  6.00%)
 SUM:     883
 
-----------------------
 
a*=*:     210 ( 13.98%)
b*=*:     190 ( 12.65%)
c*=*:     260 ( 17.31%)
d*=*:     273 ( 18.18%)
e*=*:     156 ( 10.39%)
f*=*:     159 ( 10.59%)
g*=*:     127 (  8.46%)
h*=*:     127 (  8.46%)
 SUM:    1502
 
=======================
 
*1=*:     619 ( 41.21%)
*8=*:     883 ( 58.79%)
 SUM:    1502
 
=======================
 
*1=Q:     604 ( 97.58%)
*1=R:       5 (  0.81%)
*1=B:       2 (  0.32%)
*1=N:       8 (  1.29%)
 SUM:     619
 
-----------------------
 
*8=Q:     866 ( 98.07%)
*8=R:       5 (  0.57%)
*8=B:       0 (  0.00%)
*8=N:      12 (  1.36%)
 SUM:     883
 
-----------------------
 
 *=Q:    1470 ( 97.87%)
 *=R:      10 (  0.67%)
 *=B:       2 (  0.13%)
 *=N:      20 (  1.33%)
 SUM:    1502
 
=======================
 
(   1502 promotions and underpromotions in   28298 games) ~ 0.0531 (promotions and underpromotions)/game.

I am surprised with these results. I sincerely expected the V-shaped graphic of other times... Maybe people resign before clear promotions in a and h files while let to promote in central squares (specially in d file in this case) because they can capture the promoted pawn inmediately. My tool is not so smart to perform this analysis.

I guess that this PGN is from human games. I see more underpromotions to knight than to rook, just the opposite than in computer-computer games.

Regards from Spain.

Ajedrecista.