syzygy question (for Ronald)

[lucasart]

Is there a non trivial gain (elo-wise) in the root_probe() stuff ?

No one to my knowledge has ever shown a non-trivial gain from any tablebases with an engine with decent endgame evaluation. We need a good implementation of tablebases to do proper testing, but the multitude of different Stockfish versions is inhibiting this. It can be done, of course, but it takes some will power.

I used to think that too. But recently, with the implementation of syzygy tb in stockfish, it has been shown that there is a measurable gain. Of course, it all depends on what you consider to be a non trivial gain. Jean Francois Romang measured +8 elo in 60"+0.05" (single threaded).

Yet Stockfish has an enormous amount of endgame specific knowledge. But all the endgame knowledge of Stockfish is in the eval. It mostly attempts to recognize draws or drawish positions, and sometimes helps to win some tricky ones (eg. KBNK).

I think that most of the gain lies in TB probing, which allows a lot of 100% accurate pruning in the search, when there are few pieces left on the board. With 6-men TB, and assuming the cost of probing is negligible, it makes sense that there is a measurable ELO improvement.

If you look at Ronald's implementation in Stockfish, you'll see that the cost of probing is not big at all. At the root node, he sets a variable

Code: Select all

use_tb = nb of pieces if <= 6, 0 otherwise

and TB probing looks like

Code: Select all

if &#40;nb of pieces <= use_tb&#41; &#123;...&#125;

That way the cost in the general case is negligible enough for the tradeoff to be positive.

Also, using a TB makes late endgames look cleaner. Often it doesn't change the outcome of the game, but it improves the manner in which this outcome is achieved. That's why users like it, even though the elo gain is tiny.

[Michel]

I have added a BSD license and removed some clauses that I thought were too restrictive and updated the git repo.
I don't know how restrictive this license is but I would be glad to modify it as required because I don't want to drive anyone away.

Ok great. Hopefully they will now be reconsidered for Debian. Gaviota TB's
have recently been imported into Debian.

[jshriver]

What is the address for the scorpio githuh? Ive been hosting a set for years but people say its and older version. Would like to gen the 5 man set and replace my copy.

-Josh

[Daniel Shawul]

Sorry, I didn't realize that Scorpio's EGBB were open source. I thought the probing code came in binary form only, as a Windows DLL. 200 MB is very compact indeed: is that bitbase information comparable to a WDL table (result of the game, accounting for 50-move rule) ?

Yes it is very compact many have used it on mobile devices even on older ones. I am not all that optimistic about 6-men bitbases because it looses all the nice properties you mentioned. It becomes bulky and slow so you might as well have TBs there. I don't like TBs so it will be WDL for 6 men too even if it may need effort to win most games with WDL alone. Diep have used them before so they must be Ok to use alone. The source code for probing, generation, and compression is there if you need it but you don't need to see all that junk. It is very non-intrusive because all you need to do is add a 100 lines interface to egbbdlls. If it was up to me every engine should have had them just because of this neat approach, compared to bulky Nalimov TB probing code that is statically linked. I can understand why people would need convincing for those but 100 lines of code that is much faster than TBs, come on? So people who rejected Scorpio bitbases on the grounds of philosophy, better have some good reason for making a switch now, instead of repeating stuff that is not new as new like you just did. You seem to be brand new to this TB stuff from the way you present the probing when tb < 6 as a novel approach, but I can guarantee if there is any improvement over scorpio bitbases it will be very minimal, if at all.

The probing code has many other features as well. Using scorpio bitbases alone can can win many difficult endgames by modifying the egbb scores, which I think is the first to do so. There is caching if you need it besides the system cache. There is also multi-thread support if you need it. My bitbases will give same elo gain as any bitbases out there for 5-men ,or your money back? I am not interested in the buisness of TBs anymore but the uninformed stuff that keeps written here is unbelivooble

[Daniel Shawul]

The format have changed since I modified the code for 6 men and above, which I never completed, so please don't regenerate. But since you are hosting the code, could you please take the license file and add it to the package you host? I am happy both ways but it seems to have prevented some from using it as they wish in the absence a liscense.
Thanks for hosting
P.s: Address of github is in my signature