29 Alpha-Beta's against Leela...

[Glarean]

I can't help but mention one issue I have with your article. You compare depth and nodes of Leela with other engines, which mean completely different things in MCTS and AB engines.

Nop, things are much easier: Alpha Beta's take 2 minutes, AI's takes 2 seconds. Calculate against knowledge...
And of course you're right: Depths and nodes are not so meaningful in Leela.

[jp]

Nop, things are much easier: Alpha Beta's take 2 minutes, AI's takes 2 seconds.

That's a hardware-dependent statement.

[mclane]

But why do we need a 30 or 40 depth search to play a game of chess. It sounds to me like using 30 tons of tnt to open a door we could easily open by putting a key into the lock that fits.

Yes we can use 64 CPU cores and compute up to depth 45 . But why ??

What do you want to demonstrate with this ??

That you can waste more energy then others ?

[Vinvin]

...
One of the most interesting positions from these tournament games is this one here:

[d]r3r1k1/3b1ppp/8/1p1p4/nPpP1BP1/4P2P/P4P2/1NQ3K1 b – – 0 24
Here Leela played 24... c3!
...

One more good and hard test position !

[Glarean]

Nop, things are much easier: Alpha Beta's take 2 minutes, AI's takes 2 seconds.

That's a hardware-dependent statement.

Of course: On very slow hardware Alpha Beta's need 15 minutes and Leela 15 seconds...

[mwyoung]

But why do we need a 30 or 40 depth search to play a game of chess. It sounds to me like using 30 tons of tnt to open a door we could easily open by putting a key into the lock that fits.

Yes we can use 64 CPU cores and compute up to depth 45 . But why ??

What do you want to demonstrate with this ??

That you can waste more energy then others ?

Obviously said by someone who knows nothing about the Game Tree Complexity of chess. .

[Milos]

But why do we need a 30 or 40 depth search to play a game of chess. It sounds to me like using 30 tons of tnt to open a door we could easily open by putting a key into the lock that fits.

Yes we can use 64 CPU cores and compute up to depth 45 . But why ??

What do you want to demonstrate with this ??

That you can waste more energy then others ?

In your classlessness you are probably totally unaware of the fact the MCTS-based engines (so Lc0 with all of it NN clones included) search much deeper than A/B engines. So if A/B engines in your analogy use 30 tons of tnt to open a door, Lc0 (and its NN clones) are using a 100 kt nuke for the same job.

[zullil]

im completely unsure though wether 1...Ra6/Rec8 (maybe 1...Re6 as well and even 1...h5 possibly) are really worse here...at least for Ra6/Rec8 the h5 idea (in mwyoung's SF-line) seems to work as well after a really quick overview from my side with bad hardware in use...maybe a SF-multiPV=5 search on strong hardware with decent amount of time could bring clearence...from my pov 1...c3 seems clearest as well though at first sight !?

According to Stockfish-dev, c3 and Rec8 are equivalent by transposition:

-2.81 1... c3 2. Na3 Rec8 3. Nc2 Nb2 4. a3 h5 5. f3 Ra6 6. Kg2 hxg4 7. hxg4 Nc4 8. Kg3 f6 9. e4 dxe4 10. fxe4 Be8 11. d5 g5 12. Be3 Bg6 13. Bc5 Re8 14. Ne3 Nxe3 15. Bxe3 c2 16. Qxc2 Rxa3 17. Qc1 Rd3 18. d6 Rxd6 19. Bxg5 fxg5 20. Qxg5 Rxe4 21. Qxb5 Rde6 22. Qb7 Re3+ 23. Kf2 R3e5 24. Qc8+ Be8 25. Kf3 Kf8 26. Qc7 Rb5 27. Qc4 Bc6+ 28. Kg3 Rbe5 29. Qd4 Kf7 30. Qc4 Re4 31. Qc5 Kg8 32. Kh4 Be8 33. Qg5+ Kh8 34. b5 R4e5 35. Qf4 Rxb5 36. Kg3 Kg8 (depth 45, 0:57:02)

-2.81 1... Rec8 2. Na3 c3 3. Nc2 Nb2 4. a3 h5 5. f3 Ra6 6. Kg2 hxg4 7. hxg4 Nc4 8. Kg3 f6 9. e4 dxe4 10. fxe4 Be8 11. d5 g5 12. Be3 Bg6 13. Bc5 Re8 14. Ne3 Nxe3 15. Bxe3 c2 16. Qxc2 Rxa3 17. Qc1 Rd3 18. d6 Rxd6 19. Bxg5 fxg5 20. Qxg5 Rxe4 21. Qxb5 Rde6 22. Qb7 Re3+ 23. Kf2 R3e5 24. Qc8+ Be8 25. Kf3 Kf8 26. Qc7 Rb5 27. Qc4 Bc6+ 28. Kg3 Rbe5 29. Qd4 Kf7 30. Qc4 Re4 31. Qc5 Kg8 32. Kh4 Be8 33. Qg5+ Kh8 34. b5 R4e5 35. Qf4 Rxb5 36. Kg3 Kg8 (depth 45, 0:57:02)

Took a while for Stockfish to come to this conclusion, with c3 considered "better" until this iteration.

[EDIT] And now c3 retakes the lead!

-3.02 1... c3 2. Na3 Rec8 3. Nc2 Nb2 4. a3 h5 5. f3 Ra6 6. Kg2 hxg4 7. hxg4 Nc4 8. Kg3 f6 9. e4 dxe4 10. fxe4 Be8 11. d5 g5 12. Be3 Bg6 13. Bc5 Re8 14. Ne3 Nxe3 15. Bxe3 c2 16. Qxc2 Rxa3 17. Qc1 Rd3 18. d6 Rxd6 19. Bxg5 fxg5 20. Qxg5 Rxe4 21. Qxb5 Rde6 22. Qd5 Re5 23. Qc4 Be8 24. Qc8 Re3+ 25. Kf2 R6e5 26. Qc4+ Kg7 27. Qc7+ Re7 28. Qc5 R3e5 29. Qd4 Kh7 30. Qd3+ Kg8 31. Qc4+ R7e6 32. Qd4 Re4 33. Qd5 Kg7 34. Kg3 R6e5 35. Qd6 Kg8 36. Qb8 Kh7 37. Qc7+ Re7 38. Qc5 Rb7 39. Qf5+ Bg6 40. Qc8 Rexb4 41. Qd8 Rb3+ 42. Kh4 (depth 46, 1:08:11)

-2.86 1... Rec8 2. Nd2 h5 3. gxh5 c3 4. Nb3 c2 5. Kh2 Nc3 6. Be5 Rc6 7. Qg1 f6 8. Bf4 Bf5 9. h6 g6 10. e4 Nxe4 11. Qf1 Rc4 12. Qe2 Rxa2 13. Nc1 Ra7 14. h4 Kh7 15. f3 Nc3 16. Qe3 Rc8 17. Kg3 Rb7 18. Kh2 Na4 19. Kg2 Nb6 20. Qe2 Nc4 21. Kf2 Nb2 22. Kg3 Ra8 23. Kg2 Rg8 24. Kf2 Nd1+ 25. Kf1 Rc8 26. Kg2 Nc3 27. Qe3 Na4 28. Kf2 Rf7 29. Kg3 Nb2 30. Qa3 Nc4 31. Qa6 g5 32. hxg5 fxg5 33. Bxg5 Rg8 34. Kh4 (depth 46, 1:08:11)

[Spliffjiffer]

yes, as i thought....1...ra6 seems to be not really worse as well according a quick insight (if worse at all, which i cant see atm)...and if 1...rec8 is a tranposition then how long does it take SF to find it, lol...if it chooses 1...Ra6 it doesnt matter imo...its always the same (more or less) imo...bad testpos honestly as far as i dont miss something serious...thx Louis for reading my post and doing the stuff !!

[supersharp77]

I organized a big tournament with 33 engines, including some derivatives of Stockfish and some networks for Leela.
(2m+2s permanent-brain on - AMD-Ryzen7-2700X & RTX2080 - 5-moves book)

The Ranking



I evaluated the whole tournament chess-wise and summarized it in a report (with download possibility of all PGN-files):
https://glarean-magazin.ch/2019/08/24/s ... ogramming/

(You can use the "Translate" button at the bottom left for an automatic english translation of the report).

One of the most interesting positions from these tournament games is this one here:

[d]r3r1k1/3b1ppp/8/1p1p4/nPpP1BP1/4P2P/P4P2/1NQ3K1 b – – 0 24
Here Leela played 24... c3!

An amazing move that the Alpha Beta engines don't see...

Greetings: Walter

>Stockfish can find the move!
>Analysis by Stockfish 230819 64 POPCNT:
>1...c3 2.Na3 Rec8 3.Nc2 Nb2 4.a3 h5 5.f3 hxg4 6.hxg4 Nc4 7.Kf2 f6 8.Qd1 g5 9.Bg3 Be8 10.f4 Nd2 11.Kg2 Ne4 12.Qb1 Kg7 13.Qa1 Bd7 14.f5 Rh8 15.Ne1 Ra7 16.Nc2 Rha8 17.Qc1 Rc8 18.Qf1 Rg8 19.Qa1 Rh8 20.Ne1 Kf7 21.Nc2 Bc6 22.Ne1 Kg8 23.Nd3 Rah7 24.Nf2 c2 25.Qc1
>Black is clearly better: -+ (-2.07) Depth: 36/82 00:02:18 4308MN, tb=18119

Yeah, of course - but only after 2 minutes. Too late!

See Leela (0 sec!):

Analysis by Lc0 0.22 (NW42595):

24...c3 25.Dc2 Ta6 26.Kg2 h6 27.Kg3 Tc8 28.f3 Tc4 29.Sa3 Txb4 30.Dd3 Ta8 31.Sc2 Tb1 32.e4 Tc8 33.exd5 Sb6 34.d6 Sd5 35.De4 Sxf4 36.Kxf4 Tb2 37.a3 Te8 38.Dd3 Tc8
Schwarz steht deutlich besser: -/+ (-1.42) Tiefe: 6/19 00:00:00
Schwarz steht deutlich besser: -+ (-1.65) Tiefe: 14/40 00:00:55 1929kN

Leela doesn't calculate, Leela knows...

Greetings: Walter

.

It was faster then 2 min. That is when I clip the analysis.
But of course I thought you tested a bunch of A/B engines and said.
"An amazing move that the Alpha Beta engines don't see..."

Yes, Lc0 sees the move instantly!!!

Nice try Walter but unfortunately my friend these "results" are 100% meaningless..."Apples with Apples and Oranges with Oranges"
1.Why not use LC0 (CPU) when they post these "Results"? Its always LC0 with a superfast GPU setup vs AB searchers with a limited or no opening book...(at least 8-9 moves book)
2. 2 minutes plus 2 seconds with ponder on? How come? Whats the hurry? Of course GPU has a huge speed advantage with your "setup" How about 5 minute plus 10 seconds or 5 minutes plus 15 seconds? We want to find the correct continuations..right?
3. I see one Brainfish and Stockfish 10 and a Cfish and 1 CorChess... wheres ASM Fish, wheres SugaR?
Wheres Thothsfish? Wheres Learning Stockfish? Wheres Sugar NN? Wheres Sugar MCTS? Wheres Stockfish Development (2019 versions)?...Scorpio 3.0 (or 2.9).etc etc.... Nice Try my friend...Back to The DRAWING BOARD!