29 Alpha-Beta's against Leela...

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Rebel, chrisw

User avatar
mclane
Posts: 18747
Joined: Thu Mar 09, 2006 6:40 pm
Location: US of Europe, germany
Full name: Thorsten Czub

Re: 29 Alpha-Beta's against Leela...

Post by mclane »

I am talking about the fact that these engines have no clue about chess but they are capable to search 40 plies.
As if this is the sense of chess.

I can buy a car that has 30 PS or a car that drives with 300 ps.

Yes the car with 300 ps can drive faster.


But is this the target ??

It’s only a waste of energy.

A dedicated chess computer plays chess with 5 MHz and runs on batteries. It gets 1950 ELO.
Compare this to the amount of energy your pc uses .
It makes no sense .
What seems like a fairy tale today may be reality tomorrow.
Here we have a fairy tale of the day after tomorrow....
jp
Posts: 1470
Joined: Mon Apr 23, 2018 7:54 am

Re: 29 Alpha-Beta's against Leela...

Post by jp »

I think people here want to know what you'd suggest doing instead. Ban all existing AB and NN chess engines?
Last edited by jp on Sun Aug 25, 2019 6:21 pm, edited 1 time in total.
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: 29 Alpha-Beta's against Leela...

Post by zullil »

mclane wrote: Sun Aug 25, 2019 5:58 pm I am talking about the fact that these engines have no clue about chess but they are capable to search 40 plies.
As if this is the sense of chess.

I can buy a car that has 30 PS or a car that drives with 300 ps.

Yes the car with 300 ps can drive faster.


But is this the target ??

It’s only a waste of energy.

A dedicated chess computer plays chess with 5 MHz and runs on batteries. It gets 1950 ELO.
Compare this to the amount of energy your pc uses .
It makes no sense .
But Lc0 running on my abacus loses a lot ... :roll:
Uri Blass
Posts: 10216
Joined: Thu Mar 09, 2006 12:37 am
Location: Tel-Aviv Israel

Re: 29 Alpha-Beta's against Leela...

Post by Uri Blass »

mclane wrote: Sun Aug 25, 2019 4:08 pm
Uri Blass wrote: Sun Aug 25, 2019 3:35 pm
mclane wrote: Sun Aug 25, 2019 10:53 am My brain is also doing massive amounts of calculations when I play chess, but my mind only looks onto the board without building huge search trees.
The gag is that humans do play chess with eyes seeing patterns and trajectories, while computers search into the 30-40 search today and do huge combinations out of moves to find out what humans see with their eyes. Also humans automatically see important things due to experience.

The process of learning is thinking with the backbones.
First you need to register everything with conscious, later the unconsciousness is doing the job,
Alpha beta chess engines do not search 30-40 plies in order to find out what humans see with their eyes but in order to find out more than what humans see with their eyes.
Yes but they run around without any plan,
They do enormous amounts of lines of play but have no clue what to do,
That’s the reason the draw rate is increasing and increasing. Also most programs are clones of other programs, they make so many draws.

Where is it good for ??

What is the sense of all this computation if the engines have no clue about chess ?

They do lines that maybe won’t lose material in 30 plies.
But it’s not the target of chess not to lose material.
The target of chess is to mate.

Or do you know any fide rule that says the player with more material wins the game ??
The player who mates wins the game,
Or if the opponent has fallen asleep or dead or time out or resigns or it is technical draw or whatever,
But there is no rule that says you win if your 30 plies search makes you win a pawn.

What the programs make is finding moves.
They are not finding intelligent plans.
They have none.
top humans also have many draws in their game.

I do not see engines make more draws than humans at similiar level.
I think that it is the opposite and ralatively weak engines at the level of 2800 make less draws relative to humans in engine-engine games.

chess engines have some weaknesses that humans do not have but their evaluation is clearly not only material.

When I look at one of the games pablo drew against stockfish I can see that stockfish evaluated obviously drawn position with equal material as almost 3 pawns advantage.

[d]4b1k1/rp4rp/p1p1p1pP/P1PpPpP1/1R3P2/4PB2/P4K2/7R b - - 0 85

Chess engines clearly have weaknesses and there are positions that they do not understand but I do not think that the problem of top engines is that they are materialistic.

There are trivial positions that they do not understand but there are also positions that they understand better than most humans.

Stockfish solve easily most of the nolot test suite when some of the positions are clearly positional problems and not tactics.

[d]r2qk2r/ppp1b1pp/2n1p3/3pP1n1/3P2b1/2PB1NN1/PP4PP/R1BQK2R w KQkq - 0 1

Stockfish solves Nolot number 3 clearly by positional evaluation.

other engines that do less pruning than stockfish cannot see white's advantage at depth 21 after the following line
1.Nxg5 Bxd1 2.Nxe6

https://www.chessprogramming.org/The_Nolot_Suite

Bruce moreland wrote in the past that he does not think that anyone has ever solved this position and I am sure that he gave Ferret a lot of hours.
Stockfish can get more than +2 easily and I do not believe it is because of seeing more tactics.

Next example for the great positional power of stockfish is the following position nolot number 5.

[d]r2qrb1k/1p1b2p1/p2ppn1p/8/3NP3/1BN5/PPP3QP/1K3RR1 w - - 0 1

Ferret needed 93 hours to get +5 score
Stockfish can get more than +7 in less than a minute.

How it is possible?
I do not believe stockfish could see in 1 minutes depths that Ferret could not see in 93 hours so again it is a great positional power of stockfish that does not only count material and the high score is also result of positional factors.

Nolot number 6 is also positional problem in the nolot suite and again stockfish can solve it easily(depth 17 on my hardware) when Bruce said
"I don't think that anyone has ever solved this."

[d]rnbqk2r/1p3ppp/p7/1NpPp3/QPP1P1n1/P4N2/4KbPP/R1B2B1R b kq - 0 1
supersharp77
Posts: 1242
Joined: Sat Jul 05, 2014 7:54 am
Location: Southwest USA

Re: 29 Alpha-Beta's against Leela...

Post by supersharp77 »

Glarean wrote: Sun Aug 25, 2019 2:00 am
supersharp77 wrote:Nice try Walter but unfortunately my friend these "results" are 100% meaningless..."Apples with Apples and Oranges with Oranges"
supersharp77 wrote: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)
the best teNope, I wanted the best test environment for the Alpha Beta's andst environment for the Leela's. And I didn't want to compare opening books, so just a book with 5 moves.
You Really wanted wanted the "Best Test Environment environment for the Leela's" "Your No Friend Of Stockfish (or any other non LC0 engine/NN" and your "Tests are 100% Biased" as It shows on your German Page... :lol: :lol: :wink:



"Apples with Apples..Oranges with Oranges"
jdart
Posts: 4363
Joined: Fri Mar 10, 2006 5:23 am
Location: http://www.arasanchess.org

Re: 29 Alpha-Beta's against Leela...

Post by jdart »

stockfish could see in 1 minutes depths that Ferret could not see in 93 hours
Today's engines are better searchers but also don't forget that Moore's law has vastly increased processing power from Ferret's time to now.

--Jon
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: 29 Alpha-Beta's against Leela...

Post by zullil »

Glarean wrote: Sun Aug 25, 2019 10:02 am
These are two clearly different MultiPV's.

You can try to reach overnight depth 53 (MultiPV) on your computer yourself, maybe you'll get something like this:
1. -+ (-4.00)
2. -+ (-3.00)
3. -+ (-2.00)

I'm not saying that this position is a perfect test position. I'm just claim that 24... c3 is the most efficient and interesting move in this position.
This has nothing to do with any criteria for "fast winning" in Engines.
And the most important thing is, it's a tournament move played only by AI engines, never by A/Bs.
That's my point.
It seems you are correct! A new 10-hour search with MultiPV = 2 shows two distinct lines, with the c3 line clearly ahead. No sign of two equivalent lines at any time during this new search. Very different from yesterday, which is strange since the only change I made was to update Stockfish to include the most recent commit, a patch tweaking null move pruning. Seems unlikely that this one patch would so strongly affect this position.

-3.69 1... c3 2. Na3 Rec8 3. Nc2 Nb2 4. a3 h5 5. f3 hxg4 6. hxg4 Nc4 7. Kh2 Ra6 8. Kg3 f6 9. e4 dxe4 10. fxe4 Be8 11. d5 Bg6 12. a4 Rxa4 13. Nd4 c2 14. Qxc2 Ra3+ 15. Kf2 Rca8 16. Ne2 Ra2 17. Qd1 Nb2 18. Qd4 R8a3 19. Bc1 Nd3+ 20. Kf1 Rxe2 21. Kxe2 Nxc1+ 22. Kf1 Rf3+ 23. Kg2 Rd3 24. Qb6 Bxe4+ 25. Kg1 Kh7 26. Qxb5 Bxd5 27. Kf2 Kh6 28. Qd7 g6 29. Ke1 Rd4 30. Kf2 Kg5 31. b5 Rd1 32. Qc7 Nd3+ 33. Ke2 Rc1 34. Qd6 Nf4+ 35. Ke3 Rc3+ 36. Kd4 Rc4+ 37. Ke3 Re4+ 38. Kd2 Rd4+ 39. Kc3 Rc4+ 40. Kb3 Rc6+ 41. Qxd5+ Nxd5 42. bxc6 Kf4 43. Kc4 Ke5 44. Kc5 Nc7 45. Kb6 Kd6 46. Kb7 Ne6 (depth 56, 10:00:19)

-2.92 1... Rec8 2. Nd2 c3 3. Nb3 c2 4. Nc5 Nxc5 5. dxc5 Rxa2 6. Be5 d4 7. exd4 Bc6 8. d5 Bxd5 9. Qg5 f6 10. Bxf6 c1=Q+ 11. Qxc1 gxf6 12. Qd1 Rd8 13. c6 Raa8 14. c7 Rd7 15. Qd4 Kg7 16. g5 Kg8 17. h4 fxg5 18. Kh2 Rc8 19. Qf6 Rcxc7 20. Qxg5+ Kf7 21. Qf5+ Ke7 22. Qe5+ Be6 23. Qg7+ Bf7 24. Qe5+ Kd8 25. Qxb5 Rc4 26. Qb8+ Ke7 27. Kg3 Rd3+ 28. f3 Rcc3 29. Qe5+ Be6 30. Qg7+ Kd6 31. Qf6 Rxf3+ 32. Qxf3 Rxf3+ 33. Kxf3 Kc6 34. b5+ Kxb5 35. Kg3 Kc4 36. Kf4 Bd5 37. Ke5 Kc5 38. Kf6 Kd4 (depth 56, 10:00:19)
Glarean
Posts: 262
Joined: Sun Oct 05, 2008 1:04 pm
Location: Switzerland
Full name: Walter Eigenmann

Re: 29 Alpha-Beta's against Leela...

Post by Glarean »

zullil wrote: Sun Aug 25, 2019 10:00 pm
Glarean wrote: Sun Aug 25, 2019 10:02 am
These are two clearly different MultiPV's.

You can try to reach overnight depth 53 (MultiPV) on your computer yourself, maybe you'll get something like this:
1. -+ (-4.00)
2. -+ (-3.00)
3. -+ (-2.00)

I'm not saying that this position is a perfect test position. I'm just claim that 24... c3 is the most efficient and interesting move in this position.
This has nothing to do with any criteria for "fast winning" in Engines.
And the most important thing is, it's a tournament move played only by AI engines, never by A/Bs.
That's my point.
It seems you are correct! A new 10-hour search with MultiPV = 2 shows two distinct lines, with the c3 line clearly ahead. No sign of two equivalent lines at any time during this new search. Very different from yesterday, which is strange since the only change I made was to update Stockfish to include the most recent commit, a patch tweaking null move pruning. Seems unlikely that this one patch would so strongly affect this position.

-3.69 1... c3 2. Na3 Rec8 3. Nc2 Nb2 4. a3 h5 5. f3 hxg4 6. hxg4 Nc4 7. Kh2 Ra6 [...] (depth 56, 10:00:19)
-2.92 1... Rec8 2. Nd2 c3 3. Nb3 c2 4. Nc5 Nxc5 5. dxc5 Rxa2 6. Be5 d4 7. exd4 Bc6 [...] (depth 56, 10:00:19)
Thx for your efforts.

Greetings: Walter
zullil
Posts: 6442
Joined: Tue Jan 09, 2007 12:31 am
Location: PA USA
Full name: Louis Zulli

Re: 29 Alpha-Beta's against Leela...

Post by zullil »

Vinvin wrote: Sat Aug 24, 2019 9:42 pm
Glarean wrote: Sat Aug 24, 2019 6:17 pm...
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 !
A deep new search with Stockfish-dev suggests that c3 might indeed be the "best" move here:

-3.69 1... c3 2. Na3 Rec8 3. Nc2 Nb2 4. a3 h5 5. f3 hxg4 6. hxg4 Nc4 7. Kh2 Ra6 8. Kg3 f6 9. e4 dxe4 10. fxe4 Be8 11. d5 Bg6 12. a4 Rxa4 13. Nd4 c2 14. Qxc2 Ra3+ 15. Kf2 Rca8 16. Ne2 Ra2 17. Qd1 Nb2 18. Qd4 R8a3 19. Bc1 Nd3+ 20. Kf1 Rxe2 21. Kxe2 Nxc1+ 22. Kf1 Rf3+ 23. Kg2 Rd3 24. Qb6 Bxe4+ 25. Kg1 Kh7 26. Qxb5 Bxd5 27. Kf2 Kh6 28. Qd7 g6 29. Ke1 Rd4 30. Kf2 Kg5 31. b5 Rd1 32. Qc7 Nd3+ 33. Ke2 Rc1 34. Qd6 Nf4+ 35. Ke3 Rc3+ 36. Kd4 Rc4+ 37. Ke3 Re4+ 38. Kd2 Rd4+ 39. Kc3 Rc4+ 40. Kb3 Rc6+ 41. Qxd5+ Nxd5 42. bxc6 Kf4 43. Kc4 Ke5 44. Kc5 Nc7 45. Kb6 Kd6 46. Kb7 Ne6 (depth 56, 11:13:21)

-2.92 1... Rec8 2. Nd2 c3 3. Nb3 c2 4. Nc5 Nxc5 5. dxc5 Rxa2 6. Be5 d4 7. exd4 Bc6 8. d5 Bxd5 9. Qg5 f6 10. Bxf6 c1=Q+ 11. Qxc1 gxf6 12. Qd1 Rd8 13. c6 Raa8 14. c7 Rd7 15. Qd4 Kg7 16. g5 Kg8 17. h4 fxg5 18. Kh2 Rc8 19. Qf6 Rcxc7 20. Qxg5+ Kf7 21. Qf5+ Ke7 22. Qe5+ Be6 23. Qg7+ Bf7 24. Qe5+ Kd8 25. Qxb5 Rc4 26. Qb8+ Ke7 27. Kg3 Rd3+ 28. f3 Rcc3 29. Qe5+ Be6 30. Qg7+ Kd6 31. Qf6 Rxf3+ 32. Qxf3 Rxf3+ 33. Kxf3 Kc6 34. b5+ Kxb5 35. Kg3 Kc4 36. Kf4 Bd5 37. Ke5 Kc5 38. Kf6 Kd4 (depth 56, 11:13:21)
Glarean
Posts: 262
Joined: Sun Oct 05, 2008 1:04 pm
Location: Switzerland
Full name: Walter Eigenmann

Re: 29 Alpha-Beta's against Leela...

Post by Glarean »

supersharp77 wrote: Sun Aug 25, 2019 8:40 pm You Really wanted wanted the "Best Test Environment environment for the Leela's" "Your No Friend Of Stockfish (or any other non LC0 engine/NN" and your "Tests are 100% Biased" as It shows on your German Page... :lol: :lol: :wink:
Please, don't keep wasting our time here!