CCRL Discussion Board

Anyone know if Mr. Nalimov is still working on egtb code or plans to start work on 7men egtb generation? If so any talks on sharing the cpu load to get some tb's out?

-Josh

bumping this up

What would this take? I would be willing to pay a coder to generate this code. assume a huge amount of memory would be needed, perhaps a sun system running unix could do this?

I cannot imagine the size of these databases, but would be a very interesting undertaking.

Any ideas?

Matt

kp1139 wrote: I cannot imagine the size of these databases, but would be a very interesting undertaking.

Well, from 4men to 5men and and from 5men to 6men, the increase of the size was roughly 200 fold. Extrapolating this to 7men will give something in the order of 200-250 TB.

Toni

What would be the size of the 7men bitbases?

The offer from 'kp' of funding for work on 7-man EGTs is most welcome. IMHO it would be best directed to the team of Yakov Konoval in Oreval and Marc B in the USA who are well into 7-man work, repeatedly breaking records, and contemplating generating 7-man P-ful EGTs.
Note that they are using the DTZ (== Depth to move-count Zeroing move) metric rather than thte DTM metric: it generates smaller EGTs and the payoff increases with number of men, number of Pawns and imbalance, e.g. 5-1p endgames.
Their code is very hot, presumably because it is written in Pentium Assembler and employs a bunch of ideas not used in the Nalimov approach.
Funding could also be pointed at Eiko Bleicher, whose FREEZER software builds on Nalimov EGTs to build EGTs for slices of endgames with one more man but with facing or blocked Pawns. Thus, some 7-man studies can be rigorously analysed with the use of FREEZER software.
Funding could also be pointed at the development of software for data-mining the available EGTs. FREEZER and WILHELM do a job but not the whole job.
g

guyhaw wrote:The offer from 'kp' of funding for work on 7-man EGTs is most welcome. IMHO it would be best directed to the team of Yakov Konoval in Oreval and Marc B in the USA who are well into 7-man work, repeatedly breaking records, and contemplating generating 7-man P-ful EGTs.
Note that they are using the DTZ (== Depth to move-count Zeroing move) metric rather than thte DTM metric: it generates smaller EGTs and the payoff increases with number of men, number of Pawns and imbalance, e.g. 5-1p endgames.
Their code is very hot, presumably because it is written in Pentium Assembler and employs a bunch of ideas not used in the Nalimov approach.
Funding could also be pointed at Eiko Bleicher, whose FREEZER software builds on Nalimov EGTs to build EGTs for slices of endgames with one more man but with facing or blocked Pawns. Thus, some 7-man studies can be rigorously analysed with the use of FREEZER software.
Funding could also be pointed at the development of software for data-mining the available EGTs. FREEZER and WILHELM do a job but not the whole job.
g

Hi Guy,

Listen since I am not up to speed with the vast ways of generating these egtb's I need to know if they use the DTZ metric method you speak about can we still be able to generate the TBS stat files the way Eugene does?? I mean would it be possible to get all the TBS stats based on the DTZ metric output?

Thanks,

Les

Marc B did a wonderful job generalising Eugene Nalimov's program TBGEN to produce GTBGEN - so that endgame EGTs could be produced
- for Western Chess on various sizes of board
- for various forms of Fairy Chess with 'old' and/or 'different' pieces
- for various metrics which had been in discussion

The program is not on 'open release' but was used by John Tamplin to generate all 3-to-5-man and 6-man P-less EGTs to the metrics:
DTC == Depth to Conversion, i.e. to change of force on the board and/or mate
DTZ == Depth to (move-count) Zeroing (move), i.e. to P-push and/or conversion
DTZ50 == DTZ but recognising draws assumed claimed under the 50-move rule
GTBGEN generated Nalimov-style stats which were summarised in papers, see http://www.tinyurl/law6k. http://www.icga.org - Western Chess - Endgames is the source for almost all the stats available now.
The impact of the 50-move rule in terms of denied and delayed wins has been analysed.
The 7-man stats coming out now are to the DTZ metric. As dtz =< dtm, we are in the interesting position of knowing that the old maxDTM record of 262 has been broken, and is now >= maxDTZ = 517 (in KRBNKQN). However, it is not known what maxDTM actually is.
g