Win-Draw-Loss statistics for 6-men Nalimov

[Ola Mikael Hansson]

Hi,

I'm looking for the number of positions that are Won / Drawn / Lost for the different 6-men Nalimov endgames. Does anybody have these statistics, or know where to find them?

[Kirill Kryukov]

Some statistics can be found here. Also you can check the tbs files available from this page.

[Ola Mikael Hansson]

Also you can check the tbs files available from this page.

Thanks, I've been gradually extracting the statistics from the tbs files. However, upon reaching KBN-KNN, I encountered an oddity:

Code: Select all

White Wins           Draws                       Black Wins	
42018594   1215638   -1857279612   -1693721661   77936   514887

Negative number for draws, with both wtm and btm - I assume this is some kind of overflow issue. Does anybody know what the correct numbers would be, and whether other tbs files are afflicted with such errors as well, or are they trustworthy?

[Ola Mikael Hansson]

KBNKNN also has a negative number in the tbs statistics.

What I've been using these statistics for, is coming up with my own priority for which table-bases to download. In addition to frequency (taken from posts linked on the Endgame Tablebases Online site - I use a 2:1 weighted mix of human and computer frequency), I wanted to take into account file size (if two tablebases occur with the same frequency, but one takes up 11 GB and one takes up 2 GB, having the 2 GB is obviously more efficient, as it leaves 9 GB available for more tablebases), and how much impact the endgame has (e.g. KBNKBN is almost always drawn, and if it's not, it's likely to be due to tactics in the very first moves following conversion, which the search can easily find - contrast against KPPKPP which is usually decisive, so much more important to have knowledge of). Of course, the relevance of these can be disputed, and the method in which I used the available statistics is not necessarily the best one.

Still, whereas I did this for my own personal use, perhaps others can find the results interesting as well. For full details, see the attached spreadsheet - I have not done the statistics for all the tablebases (and don't know if I'll bother with the rest, as the remaining ones have very low occurrence frequencies, so I won't download them). Actually, turns out I couldn't attach the spreadsheet - "The extension ods is not allowed" - but I'd be happy to email it to anybody curious.

So, here is my personal order to download them in, along with the effiency rating my spreadsheet gives for each of them:

Efficiency Ending 100000 kppkpp 22657 kpppkp 18391 krpkrp 10591 kpppkr 8044 krppkr 7960 kbpkpp 6686 knpkpp 3799 kpppkb 2652 kpppkn 2649 krpkpp 2289 knpknp 2216 kbpknp 2105 kbppkb 2058 kbpkbp 1657 knppkb 1635 kqpkqp 1493 kbppkr 1450 krpkbp 1448 knppkp 1184 knppkr 1116 knppkn 1059 kbppkp 953 kqppkq 691 krpknp 662 krppkq 600 kpppkq 479 kbppkn 447 krrkrr 439 krbkrp 299 kqrkqr 296 krrkrb 284 kqqkqq 249 krnkrp 241 krrkrn 220 kbnkpp 219 krbkrn 208 kqpkrp 195 krnpkr 190 krbpkq 180 krbkrb 163 knnpkr 147 krppkp 137 kbbkpp 132 krbpkr 130 kqpkrr 127 krpkbn 126 krrkrp 116 krnkrn 115 kbbpkr 110 kbppkq 109 krrpkq 104 kbnpkr 101 krnpkq

[Kirill Kryukov]

Ola Mikael, very interesting. Sorry I don't have the correct statistics for Nalimov's tables, hopefully someone else can contribute since obviously the numbers should be known.

Actually, turns out I couldn't attach the spreadsheet - "The extension ods is not allowed" - but I'd be happy to email it to anybody curious.

I now added "odt" to the list of recognized extensions. Also, you can always attach a zip archive.

[Ola Mikael Hansson]

Also, you can always attach a zip archive.

Okay, here is the spreadsheet zipped up. Should be compatible with Open Office and Libre Office at least - not sure if whatever the current Microsoft spreadsheet program is can import it or not - compatibility was never high amongst MS's priorities.

The colours indicate bands - the first 5 bands are how much can be fitted into 30 / 60 / 90 / 120 / 150 GB, then the next 3 bands are 200 / 250 / 300 GB.

There are some parameters in the top two rows that can be changed for different results - e.g. if one only want frequencies from computer games to be used, simply change the weight for human games to 0. The scaling factors are purely cosmetical - I used them to get numbers that were easy to handle and understand as a human.