Page 1 of 2

7 piece TB Stockfish kibitzer in TCEC

Posted: Tue Oct 23, 2018 1:21 pm
by Jouni
Why it have less TB access than playing 6 piece SF even with 88 cores?

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Tue Oct 23, 2018 2:31 pm
by jpqy
Maybe using wrong probe depth and it slow down.

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Tue Oct 23, 2018 2:44 pm
by Uri Blass
Not sure if it has to have more tb accesses in every position.

Maybe it is the opposite and if you see 1 tablebases hit with 7 pieces you do not search forward and get 10 different positions with 6 pieces so you do not have 10 tablebases hits.

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Tue Oct 23, 2018 4:16 pm
by Jouni
The fixed start-up speed in Komodo doesn't make it much stronger, my guess was 1 ELO maybe :) .

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Tue Oct 23, 2018 9:00 pm
by Dann Corbit
When there are 7 chessmen on the board, a single 7 man probe takes the place of a zillion 6 man probes, for engines which have access only to the 6 man files.
If you have 32 man probes, you would only need a single probe on the first move of the game.
OK, I'm being a little silly. But if we did have the nearly infinite storage to hold them, and someone had the nearly infinite computer power to build them, then the 32 man files would only need a single probe for any position.

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Sat Oct 27, 2018 6:43 am
by Jouni
No. 7 piece access nees all 6 piece tables and so on. So there must be more hits. Or do You think they are unnecessary?

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Sat Oct 27, 2018 6:55 am
by Dann Corbit
Jouni wrote:
Sat Oct 27, 2018 6:43 am
No. 7 piece access nees all 6 piece tables and so on. So there must be more hits. Or do You think they are unnecessary?
If you do not have 7 man files, you must do lots of searching through the six man files to find the same information. And you do not even know the good places to look, because those are encoded in the 7 man files.

It can also be the result of the configuration chosen for probing. But it does not surprise me that you see less probes with 7 man files and it does not mean that something is broken if you see that.

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Sat Oct 27, 2018 9:09 am
by Sven
Jouni wrote:
Sat Oct 27, 2018 6:43 am
No. 7 piece access nees all 6 piece tables and so on. So there must be more hits. Or do You think they are unnecessary?
1) Accessing a 7-men TB itself does not require presence of 6-men or lower TBs. It may be true for *generating* them, though.

2) When reaching a game position where the search tree hits a 7-men position (after some captures) then the search does not go any further beyond those nodes so even here no 6-men TBs are required.

3) When reaching a game position with less than 7 men, the engine would need the corresponding 6-men/5-men TBs etc. to play the optimal moves without search, unless these 6-men/5-men positions are contained in the 7-men TB (which they usually aren't). This will not happen with conditions that include TB adjudication, of course.
Dann Corbit wrote:
Sat Oct 27, 2018 6:55 am
If you do not have 7 man files, you must do lots of searching through the six man files to find the same information. And you do not even know the good places to look, because those are encoded in the 7 man files.

It can also be the result of the configuration chosen for probing. But it does not surprise me that you see less probes with 7 man files and it does not mean that something is broken if you see that.
The main difference between using 6-men or 7-men TBs that I would expect would be that roughly the same number of hits occurs a bit earlier in the game when using 7-men. For instance, searching in a 10-men position might have roughly the same number of TB hits with 7-men TBs as searching in a 9-men position when using 6-men TBs. Therefore, if you compare the total number of TB hits occurring during a whole game then I would *not* expect that number to be smaller when using 7-men. It is the part of the game where search will not yet benefit from TB hits that keeps getting smaller with increasing number of pieces in your TB.

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Sat Oct 27, 2018 12:46 pm
by Uri Blass
Sven wrote:
Sat Oct 27, 2018 9:09 am
Jouni wrote:
Sat Oct 27, 2018 6:43 am
No. 7 piece access nees all 6 piece tables and so on. So there must be more hits. Or do You think they are unnecessary?
1) Accessing a 7-men TB itself does not require presence of 6-men or lower TBs. It may be true for *generating* them, though.

2) When reaching a game position where the search tree hits a 7-men position (after some captures) then the search does not go any further beyond those nodes so even here no 6-men TBs are required.

3) When reaching a game position with less than 7 men, the engine would need the corresponding 6-men/5-men TBs etc. to play the optimal moves without search, unless these 6-men/5-men positions are contained in the 7-men TB (which they usually aren't). This will not happen with conditions that include TB adjudication, of course.
There can be a problem with positions with 7 pieces without the 6 piece tablebases because the engine may not know the correct capture that lead to win the game(there may be 2 captures when one lead to draw and one lead to win but without the 6 piece tablebases you may choose the wrong capture).

Note that the size of the 7 piece tablebases is bigger than the size of the 6 piece tablebases so I do not think that it is logical to have the 7 piece tablebases without the 6 piece tablebases.

Re: 7 piece TB Stockfish kibitzer in TCEC

Posted: Sat Oct 27, 2018 2:29 pm
by yurikvelo
Sven wrote:
Sat Oct 27, 2018 9:09 am

1) Accessing a 7-men TB itself does not require presence of 6-men or lower TBs. It may be true for *generating* them, though.
I was thinking the same for TB-6 and TB-5, but Ronald de Man explained on this forum that TB-5 are neccesary for Syzygy-6 to work properly in all cases.