Where I can find 5 piece tablebase statistics?

[Jouni]

I mean: white win xx %
black win yy %
draw zz %.

I only found frequency statistics so far.

thanks Jouni

[hgm]

I think I saw those at the ICGA website. You could also try kirill's website.

I don't know if your interest is purely academic, but the practical use of such statistics is limited. E.g. end-games that are dead draws typically have win percentages of 40-60%...

[Dirt]

I think you can get what you need from the tbs files available on the Crafty site. For instance, the kqbk.tbs file is:

Code: Select all

wtm: Mate in   8:             8
wtm: Mate in   7:         10809
wtm: Mate in   6:         96483
wtm: Mate in   5:        205985
wtm: Mate in   4:        270415
wtm: Mate in   3:        224186
wtm: Mate in   2:        125781
wtm: Mate in   1:         52105
wtm: Broken positions:   526735
btm: Lost in   0:          8392
btm: Lost in   1:         27513
btm: Lost in   2:         77689
btm: Lost in   3:        184527
btm: Lost in   4:        348063
btm: Lost in   5:        488290
btm: Lost in   6:        351360
btm: Lost in   7:         79776
btm: Lost in   8:         14633
btm: Lost in   9:          2873
btm: Lost in  10:            10
btm: Draws:              164158

It would be some work to add up the numbers to get what you want.

[Brunetti]

the practical use of such statistics is limited. E.g. end-games that are dead draws typically have win percentages of 40-60%...

Right. In a first time I had the idea to use such statistics for my material table, but that's nonsense, since most positions of a TB have little or no chances to ever appear in a game. Then I considered to create a statistics for every possible practical configuration of the material, processing a big database of real games, but such a process would have required some months... Eventually, I wrote my values by hand

Alex

[Jouni]

Yes tbs file has just the info I wanted. But have I understand right: broken position is illegal position? Why they are counted at all? And does "wtm" in 3 vs 2 pieces contain both material balances for white?

Jouni

[hgm]

By convention, wtm in KBBKN (say) is the side with KBB.

Broken positions are not so much illegal, as impossible. (I.e. two pieces on the same square.) I don't think positions where both sides are in check are counted in there. Positions where both Kings attack each other are not counted at all, as with Nalimov indexing such positions are not part of the tablebase. Positions where one side is in check are illegal if the other side is to move, but can be perfectly legal if the in-check side has to move.

Note that the Nalimov tbs statistics is "distorted", because of the diagonal symmetry reduction. Positions with both Kings on the diagonal correspond to 4 real positions, while the others count for 8, (in Pawnless and-games), and you don't know how many of each contribute to the reported count. If it would be of interest to have the true numbers, you could use the EGTB generator on my website to print those. Or I could post the file somewhere.

[Edmund]

the practical use of such statistics is limited. E.g. end-games that are dead draws typically have win percentages of 40-60%...

Right. In a first time I had the idea to use such statistics for my material table, but that's nonsense, since most positions of a TB have little or no chances to ever appear in a game. Then I considered to create a statistics for every possible practical configuration of the material, processing a big database of real games, but such a process would have required some months... Eventually, I wrote my values by hand

Alex

This has already been done before and made public
by Alessandro Scotti:
http://www.talkchess.com/forum/viewtopic.php?t=18240
http://www.ascotti.org/programming/chess/mat_stats.html

If you are interested I could also give you my statistics, where I used Alessandro's parser to process a larger sample.

regards,
Edmund

[Brunetti]

This has already been done before and made public
by Alessandro Scott

Very interesting, thank you!

If you are interested I could also give you my statistics, where I used Alessandro's parser to process a larger sample.

Yes I'm absolutely interested, it can help me to concretize my original idea. Thank you very much!

Alex

[Edmund]

If you are interested I could also give you my statistics, where I used Alessandro's parser to process a larger sample.

Yes I'm absolutely interested, it can help me to concretize my original idea. Thank you very much!

Alex

If you pm me your email, I can send it to you (the raw data is about 1.2mb)

[bob]

By convention, wtm in KBBKN (say) is the side with KBB.

Broken positions are not so much illegal, as impossible. (I.e. two pieces on the same square.) I don't think positions where both sides are in check are counted in there. Positions where both Kings attack each other are not counted at all, as with Nalimov indexing such positions are not part of the tablebase. Positions where one side is in check are illegal if the other side is to move, but can be perfectly legal if the in-check side has to move.

Note that the Nalimov tbs statistics is "distorted", because of the diagonal symmetry reduction. Positions with both Kings on the diagonal correspond to 4 real positions, while the others count for 8, (in Pawnless and-games), and you don't know how many of each contribute to the reported count. If it would be of interest to have the true numbers, you could use the EGTB generator on my website to print those. Or I could post the file somewhere.

With Eugene's code, I believe "broken" includes _any_ position that can never be reached legally. Both sides in check included. And positions where there is no move for the side not on move that can be unmade to produce a still-legal position. I am not sure about things like pawn on the first rank and such, but I've never found a use for that information anyway and it wouldn't really bother me to find out that it is a random number with no meaning at all.