Move 26b discussion: Everyone vs stockfish

[MooMoo2]

The poll for move 25 just closed, and the most popular move was Qh6+. Stockfish responded with 25...Kg8, and the FEN is:
4r1k1/pp3p1p/7Q/3p4/8/1N2nPK1/PPP1r1PP/8 w - - 7 26

Game so far: 1.e4 e6 2. d4 d5 3. Nd2 c5 4. Ngf3 Nf6 5. exd5 exd5 6. Bb5+ Bd7 7. Bxd7+ Nbxd7 8. O-O Be7 9. dxc5 Nxc5 10. Nb3 Nce4 11. Nfd4 O-O 12. Nf5 Re8 13. Nxe7+ Rxe7 14. f3 Nd6 15. Bg5 Nc4 16. Bxf6 gxf6 17. Qd4 Qb6 18. Rab1 Qe6 19. Rfe1 Qxe1 20. Rxe1 Rxe1+ 21. Kf2 Rae8 22. Qxf6 R1e2+ 23. Kg3 Ne3 24. Qg5+ Kf8 25. Qh6+ Kg8

Stockfish doesn't offer draws, but I'll offer a draw on behalf of Stockfish since it will accept a draw at this point. Of course, a draw has to be agreed on by both sides, so I'll let everyone discuss whether to accept the draw or play on. If the consensus is to accept the draw, congratulations on a game that was very well played! I'm surprised that everyone was able to force a draw this quickly

I'd also like to know the level of interest of an engine-assisted game against Stockfish. Proposed rules:

- Stockfish gets white, the forum gets black
- 3 days discussion, 3 days voting
- Engine use is allowed, as long as the engine used isn't Stockfish*. There are no limitations on the hardware used or on the time the engine spends on each move.
- Using multiple engines on multiple machines is allowed
- Getting assistance from people outside the forum (chess clubs, real life masters or grandmasters, etc.) is allowed
- Opening books and endgame tablebases are allowed
- Stockfish will not be using an opening book
* You may use an older version of Stockfish. Versions 1, 2, and 3 are allowed and may be downloaded at: https://www.dropbox.com/sh/75gzfgu7qo94pvh/NlXX-QLGu6

I don't think Stockfish can lose a game when it's given a lot of time and plays as white. As the time controls get longer and/or the players get stronger, the draw rate increases. White still can win a few games, but the number of games won by black drops to zero:
http://serverchess.blogspot.com/2015...m-is-long.html
https://www.chess.com/blog/ThomasJEv...--time-matters
http://www.randalolson.com/2014/05/2...d-elo-ratings/
http://chess-db.com/public/research/draw_rate.html

I'll also put my money where my mouth is. If the forum can somehow beat Stockfish, I'll gladly donate $10 to the forum, which can help cover server costs.

[LaurV]

Agree with the draw.

About beating SF, if you think it is not beatable, you may lose your money. It is beatable, and I have beat it 3-4 times, both with black and white (in over a hundred games or so) without using any engine (but starting from a preset opening, usually sicilian, which I can now play very well as black). I don't think SF has any chances of win, even against a weaker engine, but which uses a much stronger computer (or network). At the end, machine chess is a lot of computing and generating min-max trees, and all machines are the same here. But because adjudicating the full-size trees take centuries, some branches of the trees have to be cut and forgot. Where the machines differ, is at cutting branches. Some good engines (like SF) have a much better heuristic to cut out branches, being able to reach a better solution faster. "Bad" engines (like Win7's built-in Chess Titans, for example, which I beat regularly, at its max level) have a "not so good" strategy of cutting out those branches, and they are time-limited, if you increase the level they only will "think longer" in time (seconds to minutes) to still keep a reasonable playing time per move. Given enough time and enough computing power, "bad" engines will succeed in adjudicating their trees, which, at least in theory, are wider than the SF's trees, because less branches are eliminated. Therefore, at least in theory, they have better chances than SF, assuming enough power and time are given to them. That is because their chance to "miss" good moves by "tree cutting" is lower. Now, on the other hand, given the same amount of time to SF, he will think much deeper, and chess games are won not thinking wider, but thinking deeper. Like, you don't need to consider all possible moves, in some position only few moves are good, so you (I mean the engine) have to spot those moves as fast as possible, and reduce the tree weight a lot, being able to compute a higher number of semi-moves in advance that your adversary. You don't need a tree like Hitachi tree, you need one like a poplar or so... As narrow as possible and as tall (deep) as possible, therefore able to visit all branches and still reach its top fast. In fact, what chess engines do is more like a (link to come) tree, cutting branches at every level.

Of course, when more people play, the moves are "averaged" by voting, but I would happily pick up your glove here. I have a Rybka engine somewhere (about 4-5 years old) and I will install it when the game starts.

[gd_barnes]

I also agree with the draw. It was the outstanding move by LaurV on move 23 which resulted in a very quick draw!

I am looking extremely forward to playing a game with engines as black against SF. I too would like to try for a win. Since LaurV is so good at the sicilian, perhaps we can choose that opening if SF plays 1. e4. I think that many computers and people working together will have a decent chance (maybe 10-20%) of beating SF as black.

[axn]

Originally Posted by LaurV

I have a Rybka engine somewhere (about 4-5 years old) and I will install it when the game starts.

Critter is another strong engine that is freely available.

Komodo 8 is also available for free and is an _extremely strong_ engine (https://komodochess.com/)

[WMHalsdorf]

I agree with all the previous posts.

[LaurV]

Originally Posted by gd_barnes

It was the outstanding move by LaurV on move 23 which resulted in a very quick draw!

Actually the merit is yours. In that particular position they were only 3 moves, from which one was clearly bad and the second one was analyzed by you. Therefore I looked to the third...

[richs]

Congrats, everyone, especially LaurV and Gary!

[The Carnivore]

Originally Posted by axn

Critter is another strong engine that is freely available.

Komodo 8 is also available for free and is an _extremely strong_ engine (https://komodochess.com/)

Chess engine rating list:
http://www.computerchess.org.uk/ccrl..._list_all.html

SF is at the top, but Komodo and a few others are not far behind.

We'll need lots of fast machines in order to pull this one off. Every doubling of positions searched results in a gain of 40-50 ELO. The comps themselves can be used as nodes of min-max trees to extend thinking time. For example, let's say we had no opening book, SF played 1. e4, and we're thinking of playing 1. c5. While voting is going on, we could have the following setup:

Comp 1 works on our response if SF plays 2. Nf3
Comp 2 works on our response if SF plays 2. Nc3
Comp 3 works on our response if SF plays 2. c3

Comp 4 works on our response if SF plays 2. Nf3, we respond with 2. d6, and SF plays 3. d4
Comp 5 works on our response if SF plays 2. Nf3, we respond with 2. d6, and SF plays 3. Bb5+

Comp 6 works on our response if SF plays 2. Nf3, we respond with 2. Nc6, and SF plays 3. d4
Comp 7 works on our response if SF plays 2. Nf3, we respond with 2. Nc6, and SF plays 3. Bb5

Comp 8 works on our response if SF plays 2. Nc3, we respond with 2. Nc6, and SF plays 3. g3
...

Avoiding transpositions may be tricky, but the concept scales really well and is very suited for distributed computing. Ideally, by the time SF moves, there is at least one machine that has been analyzing that position non-stop for the past 2 weeks or more.

[MooMoo2]

Seems like the forum accepts the draw Great job, everyone!

Here's Stockfish's game analysis. I'll look through the older threads later this week and post anything interesting that I find.

Move No.	Depth (ply)	Nodes Searched	Eval	Next Move Expected
1		34		 2,754,754,703	-0.17	d4
2		35		 2,775,255,936	-0.13	Nd2
3		37		 7,823,890,345	-0.10	e4xd5
4		36		 4,748,231,994	-0.14	e4xd5
5		34		 3,046,969,375	-0.15	Bb5+
6		35		 1,707,757,893	-0.15	Bxd7+
7		36		 1,800,523,731	-0.12	0-0
8		37		 4,257,172,692	-0.21	dxc5
9		37		 3,632,048,539	-0.17	Nb3
10		36		 5,122,109,377	-0.21	Nfd4
11		35		 2,971,667,745	-0.23	Nf5
12		36		 8,759,943,111	-0.17	Nxe7+
13		35		 3,408,909,487	-0.21	f3
14		36		10,759,685,954	-0.27	Bg5
15		35		 2,182,846,568	-0.22	Qd4
16		38		 5,794,415,093	-0.27	Qd4
17		35		 3,870,663,402	-0.22	Rab1
18		37		 3,690,735,878	-0.25	Qf2
19		43		 6,001,644,327	 0.00	Rxe1
20		38		   887,005,521	 0.00	Kf2
21		41		 2,385,354,858	 0.00	Qxf6
22		42		 2,244,662,029	 0.00	Kg3
23		43		 1,696,109,092	 0.00	Ne3
24		44		 1,646,529,536	 0.00	Qh6+
25		38		   106,012,536	 0.00	Qg5+