#7 men chess endgame tablebase download#
#7 men chess endgame tablebase free#

#7 men chess endgame tablebase free#

I have no idea why the slowdown (compared to 6 men usage), maybe SSD disc operation requires some more free space then I left, and maybe it is because of 7 men files are really large and that for some reason causes the slowdown (It reminds very much of the situation in the past I had when using 5 and 6 men Nallimov that were placed on my slow HDD). It appears that positions that are more likely to contain (somewhere down the Search) the 7-men endgame I have on my disc experience bigger slowdowns. I checked the usage of couple of late midgame positions so far, and while usage itself has no problems, I see a slowdown in nps figure, anywhere between 10-40% (figure measured after a couple of hours on infinite). EGTB are splitted in this way in (only *.rtbw files for your WDL purposes). Other thing, do pawnful EGTB (like KPvK) need pawnless EGTB (like KQvK) in order to convert promotions or are they completely independent? If that, joint analysis pawnful + pawnless may be needed. I also guess that the relative frequency of 6v1 will be very low, in spite of the smallest *vK.rtbw, which I have found KPPPPPvK (3.5 MB), that help to raise ratio frequency/GB. I would reject 6v1 EGTB from the previous analysis because I consider SF clever enough to not stalemate or drop enough pieces to fail to win the strong side (I am not considering the special case of all bishops on the same colour) if SF is playing the losing side, it would be the opponent's fault to not win, other than getting access to EGTB might save some time to SF.

Of course, there will be many games that finish earlier than a 7-man endgame is present OTB. I do not know if such a tool exist or how difficult to program is. Other method for obtaining relative frequencies can be the next one: parse each game of a large database from the start opening and reach the move when the first 7-men endgame is available, then count of what type it is (KRPPvKRP, whatever white or black is a pawn up or other of the 1,000 remaining endgames). I found a web with the 33 most frequent 7-man endgames obtained from 7M+ games according with their claims:Дisgracefully, the results are only shown in a graph and not in tables, so exact figures can not be computed, only approximations. So we need two items: frequency of the endgames (hard to obtain) and sizes of EGTB, that are easy to obtain, for example at Once said that, you ask some educated guesses of Elo/GB and I would go on frequency/GB. I have no idea of which endgames are more difficult for engines, that is, the ones that can gain more Elo from no EGTB to EGTB enabled, so I think in the frequentist approach. I know that there are some endgames that are more difficult for engines than others, so Elo could be not linear to Elo gain (in my example above, 37%/0.01% = 3,700 but for example you may win 5 Elo with the first EGTB and 0.2 Elo with the second one, so the ratio would be 5/0.2 = 25 < 3,700, because the second endgame is much more difficult than the first one although is far less frequent).

My reasoning is the following: more 7-man endgames of one type (let's say 37%) would have more impact on Elo that less 7-man endgames of other type (let's say 0.01%). I would go for a frequentist approach although it might not be an educated guess. Does anyone have any educated guesses as starting suggestions?

#7 men chess endgame tablebase download#

Note that some of the most useful and common endgames also produce very large files so it's not clear if they are the best to download first. The metric for most helpful I'm interested in is not just Elo but Elo/GB.

> Now that SF supports 7-men syzygy I would like to download somewhere between 200GB to 300GB of the most 'helpful' WDL files.