StockFish : best move not found due to MultiPV ?

[Sesse]

I let it run overnight on the b4 position, and it reported:

d61 (928 GN): -0.68 (Bxb4 Rg1 d5 Bb2 c5 Qg4 Nf5 e6)

Interestingly enough, in the process of getting to d62, it seems to have failed high three times already?

info depth 62 seldepth 90 multipv 1 score cp 60 upperbound nodes 1391836578573 nps 37745048 hashfull 990 tbhits 0 time 36874680 pv c5b4 e1g1
info depth 62 seldepth 90 multipv 1 score cp 53 upperbound nodes 1625089871301 nps 37636634 hashfull 991 tbhits 0 time 43178406 pv c5b4 e1g1
info depth 62 seldepth 90 multipv 1 score cp 42 upperbound nodes 1915882860831 nps 37434564 hashfull 994 tbhits 0 time 51179515 pv c5b4 e1g1

[Sesse]

info depth 62 seldepth 93 multipv 1 score cp 8 nodes 2666451877227 nps 37448430 hashfull 996 tbhits 0 time 71203302 pv c5b4 e1g1 d7d5 c1b2 c7c5 a2a3 b4a5 d1e2 b7c7 f4f5 e7f5 a1f1 f5g7 e5e6 d5d4 e6f7 c7f7 g5f7 f8f7 g1g6 d8e7 e2e7 f7e7 h2g2 b6b5 g6g5 a5b6 b2c1 c5c4 g5b5 e7d7 b5b4 c4c3 b4c4 d4d3 c4c3 d3d2 f1d1 d2c1q c3c1 d7f7 d1f1 g7f5 g2h2 g8g7 c1c6 b6d4 c6e6 d4c5 a3a4 c5d6 h2g1 d6c7 f1f3 c7b6 g1h2 f7f6 e6f6 g7f6 h2g2 f6e5 f3d3 e5f4 d3d7 b6d4 d7f7 a7a6 f7d7 f4e4 a4a5 f5e3 g2g3

So only -0.08 at d62. Maybe there's something in there.

[Eelco de Groot]

No I don't think White can achieve more than a draw here. Sometimes he has just three isolated pawns left and Black two plus pawns. But Black also does not seem to be able to use those extra pawns, maybe because the Bishop is a bit powerless on b4 or a5. In the 4. a3 Ba5 5. Qe2 line of your Stockfish I looked at another continuation and white seems to have a draw there too.

[pgn] [Event "?"] [Site "?"] [Date "2018.07.11"] [Round "?"] [White "?"] [Black "?"] [Result "*"] [SetUp "1"] [FEN "3q1rk1/prppnp2/1p4p1/2b1P1N1/5P2/7P/PP5K/R1BQR3 w - -"] 1. b4 Bxb4 2. Rg1 d5 3. Bb2 c5 4. Qg4 (4. a3 Ba5 5. Qe2 Rc7 6. f5 (6. Qg4 Nf5 7. e6 d4 8. exf7+ Rcxf7 9. Ne6 {$11 diagram})) 4... Nf5 5. e6 d4 6. exf7+ (6. Nxf7 Rbxf7 7. exf7+ Kxf7 8. Qxg6+ Ke7) 6... Rbxf7 7. a3 Bd2 8. Nxf7 Kxf7 9. Qxg6+ Ke7 10. Bc1 Bc3 11. Ra2 Qd7 * [/pgn]

[d]3q1rk1/p4r2/1p2N1p1/b1p2n2/3p1PQ1/P6P/1B5K/R5R1 b - -

62 817:33 -0.67 9...Qf6 10.Qxg6+ Qxg6 11.Rxg6+ Kh7
12.Nxf8+ Rxf8 13.Rg2 c4 14.Rd1 b5
15.Rg5 Bd8 16.Rxf5 Rxf5 17.Rxd4 Bf6
18.Rd7+ Kg6 19.Bc1 a5 20.Kg3 b4
21.axb4 axb4 22.Rc7 Rb5 (116.347.161.961) 2371

62 817:33 0.00 9...Qh4 10.Qxg6+ Kh8 11.Nxf8 Qxf4+
12.Kh1 Rxf8 13.Raf1 Qe4+ 14.Kh2 Qe5+
15.Kh1 (116.347.161.961) 2371

61 817:33 0.00 9...Qd6 10.Qxg6+ Rg7 11.Nxg7 Qxf4+
12.Kh1 Qf3+ 13.Kh2 (116.347.161.961) 2371

best move: Qd8-f6 time: 817:34.000 min n/s: 2.371.818 nodes: 116.347.161.961

[Sesse]

No I don't think White can achieve more than a draw here.

Yes, I agree the two likely outcomes should be draw or black win. Of course, this was a game with short time control, so all three results are obviously possible.

[tissatussa]

I just re-read all of the comments on this forum subject and I want to hint some other approach : could the evaluation be based on some other "pattern" of any position? Eg. How may pawns are 'involved', what percent of "movement space" has a piece on its square, ranges of "claimed squares" by pieces .. i don't know how MultiPV works out with depth .. and what its intention is .. also UCI options exist like "Slow Mover" and "Contempt" .. and maybe mimic Leela by alien patterns, found by deducing our game : chess.

Now this b4! position calme along and it feeds my mind - told by Leela.

Now i'm a bit into Python and chess - i made some nifty TKinter GUI which lists UCI output in a unique way. Experimenting when several UCI engines, i encountered asmFish from CounterPly, see https://github.com/Counterply/asmFish/t ... CounterPly , a very fast StockFish-like engine. It finds b4! rather soon, at depth 17 or so .. many other engines will find b4! only after depth 21 or higher, and ONLY with MultiPV 28 or more, so that takes a lot of time .. not for Blitz

some screenshots i made to state my idea : http://imoma.eu/pubgfx/UCI2CHE_MultiPV_Depth.zip

You see the output of several UCI chess engines in my Linux SCID, for that very FEN position, after some depth. I'm a long time chessplayer and for me the position holds some secret best move which is not encountered by most engines, even when using high depth search by the UCI command "go depth #" : the engines' "best move" output also depends on MultiPV. In the position b4! will only be encountered by *ALL* engines when depth is more then 20+.

For now, I tested about 30 Linux engines i found on the net.
Does any alternative non-self-learning engine exists with a different "best move" approach ?

Anyone ?

[Sesse]

I just re-read all of the comments on this forum subject and I want to hint some other approach : could the evaluation be based on some other "pattern" of any position? Eg. How may pawns are 'involved', what percent of "movement space" has a piece on its square, ranges of "claimed squares" by pieces ..

That's called an evaluation function. Every chess program has one.

i don't know how MultiPV works out with depth .. and what its intention is .. also UCI options exist like "Slow Mover" and "Contempt" ..

I don't understand what you are trying to say. Try again with fewer words and less rambling, maybe.

[tissatussa]

Caissa once gave me this video for you to see :
https://www.youtube.com/watch?v=7dkQH0GTmrQ
own edit -- read YT description.

[tissatussa]

--- That's called an evaluation function. Every chess program has one. ---

Clear, but HOW do all engines evaluate?
What's your comment on the screenshots in the linked ZIP?

[tissatussa]

Another one from my archive : https://www.youtube.com/watch?v=r-9696xm-04 and watch their second game.

[tissatussa]

A creative and meaningful chess video : https://www.youtube.com/watch?v=CbTs-QgJc4E -- own edit, see the YT description.