Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Human versus Machine
1 2 Previous Next  
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-19 10:51
Again, why should people who have not read the book be allowed to post reviews?
Parent - By wem511 (**) [us] Date 2017-12-22 03:26
Because they are fake reviews
Parent - - By MachineLearning (*) [au] Date 2017-11-19 12:11
"and I bet it is Machine Learning here"


Sorry, It's not me. Do you want me to write one for you?
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-20 00:31

>Sorry, It's not me. Do you want me to write one for you?


You bet.

I guess I already know who you are on talkchess, I mean your real name.
Pity.

What do you care about my books.
You might not like them, but why would you want that they go down the drain, I simply
don't understand, and really can not find words to qualify that.

When you posted 3 1-star reviews under the same name, did not you ask yourself,
why would a person who has bought a book and finds it absolutely awful,
fully unreadable as riddled with grammatical mistakes and illegal chess positions,
buy another book from the same author, and then a third?

Does not make sense, does it?
But then, your purchase is not verified and you might simply go unnoticed...
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-24 05:58
Part III is out: https://www.amazon.com/dp/B077PN5QT8/ref=sr_1_8?s=books&ie=UTF8&qid=1511501584&sr=1-8

This will be the last part of the series, for the time being.

I would not like that this turns into a boring book,
the purpose has never been to publish as many games as possible,
but just to demonstrate the possibility to win against the top engines,
and cover the most common winning options.

I can play more games in the future, so there might be 4th part, but this will be only in a couple of years, when much stronger engines appear.
It does not make sense to repeat one and the same stuff. When much stronger engines appear, and people say again, well, it might have been
possible to beat the tops couple of years ago, that crazy guy Lyudmil used to do it, but not now anymore, then I will play some games to
renew my collection.

I would like to thank all those on this forum, who, by chance or willingly, have bought different of my books. Thanks a lot! You have helped me to at least keep part of my face and hope in what I am doing.

The third part features handicap wins. All the credit goes to Larry Kaufman(thanks, Larry, for all those Komodo handicap matches), as, whenever Komodo would play a match against some human, I would try my hand with
precisely the same imbalance against the tops.
It proved that handicap games are a great fun, one of the most interesting things in chess, and also are very helpful to your tactical training, for the reason that materially imbalanced positions increase the necessity for more calculations.

I would also like to thank the mods here for their patience and understanding.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-25 05:16
I forgot to say that, this time around, all diagrams are accompanied by fens, so it will be possible to check the lines with your engines.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-30 08:08
One thing I forgot to mention is games in all 3 parts have been played with ponder=on, so this somewhat decreases the time difference, but I guess, especially for closed positions, the effect would be rather small, as engines would be pondering mostly the wrong moves.

I have always wondered what the effect of ponder on game play is, I presume rather insignificant.
Parent - - By Banned for Life (Gold) Date 2017-11-30 15:16
I have always wondered what the effect of ponder on game play is, I presume rather insignificant.

In cases where the ponder hit rate is high, it can almost double the time spent on each move. Unfortunately, when the ponder move is wrong, unless there is a transposition the hash table may get wiped out, so the effect is actually substantially negative.

This leads to the age old discussion of whether it is best to work on the ponder move or to consider all opponent moves (in the same way the opponent is working)...
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-30 17:42

>In cases where the ponder hit rate is high, it can almost double the time spent on each move.


Right, I was wondering about the amount of elos gained from pondering.

Engine A draws a match against engine B, both don't ponder.
How many elos will engine A gain, if only it ponders?
Parent - - By Banned for Life (Gold) Date 2017-12-29 13:20
If the ponder hit rate was 100%, the effect would be equivalent to a doubling of time per move. This gain will be a lot larger at fast time controls, but would be in the range of 30 to 70 Elo. As the ponder hit rate falls, misses that don't transpose cause degradation in Elo. This would be hard to quantify without actual testing, but a very low ponder hit rate could actually result in lower Elo with pondering.
Parent - - By bob (Gold) [us] Date 2017-12-29 16:20
Note that this depends on how things start.  A plays B.  A goes out of book first and thinks.  And makes the move and starts to ponder.  This kicks B out of book.  100% prediction rate here will, as you said, double the time.

BUT...

A plays B.  B goes out of book first.  Same scenario except that now B gets the time bonus, not A.  And ONLY B gets the benefit here, not A.  Every time A makes a move, B replies instantly and A is right back on the clock.  The exact opposite of the first case.  100% prediction rate is not all the data you need.

There are lots of tricks to try to claim this time bonus for yourself.  For example, in the above the worst case is when you go out of book first because your opponent played an unexpected move.  One can apply a simple idea here.  You are in book, and your opponent is not, so he is already thinking.  He thinks for 3 minutes and makes a move, you play an instant book move, UNTIL he makes an unexpected reply.  Now he is already thinking (pondering) and you have to start a search.  He wins the time war at 100% prediction rates.  In Crafty, for example, I make a list of all legal opponent moves.  I look up each one to see if I have a book move.  I set those moves aside.  I now start pondering on the remaining set of moves to pick the best, then re-start the ponder search on that move.  If the opponent makes a book move, I reply instantly.  If not, he has a high probability of playing the move I found best and I still win the time battle.

As I said, ponder hit rate doesn't tell much about time, just about how well A predicted B or vice versa.  BOTH could have 100% hit rate and one gets 6 mins per move, the other gets only 3.
Parent - By Banned for Life (Gold) Date 2017-12-29 19:02
I'll stipulate to everything you've said. Throwing the book in at the beginning of the game certainly complicates matters, especially when popular GUIs handle the book, and the engine just stops rather than using the time to keep the hash up to date. There is also the matter that you brought up regarding the algorithm employed when a ponder hit occurs, where if the opponent spent enough time on the move, a ponder hit might result in an instant reply. As you mention, this can result in one engine's clock running continuously while the other engine plays in a much shorter time (maybe instantly) because it was always pondering the correct move. In some of these cases with correct ponder hits, both engines are spending the same amount of time working on the position, but only engine's clock is running. Of course this kind of sequence frequently determines the outcome of a game by giving one side a large time advantage.

But these factors only exacerbate the importance of a good book and a high ponder hit rate. If the ponder hit rate is low, pondering can actually weaken engine performance by trashing the hash after each move. You might be aware of statistics that relate Elo gain to ponder hit rate, possibly normalized to a ponder hit rate of 100%. I'm not aware of this type of data, although I'd be surprised if it was never collected.

Your discussion of Crafty's pondering algorithm is interesting. The last time I looked at the Stockfish code, it used a rather simplistic algorithm for time control. I always believed that a few Elo could be obtained with improvements there.

This all reminds me of a pet peeve from the past, when an engine could spend a lot of time when there was only one legal move. The optimal strategy (absent book considerations) would certainly be to make the move instantly and then consider all possible responses, but many engines would spend time on the move (sometimes a lot of time) for the sole purpose of coming up with a good ponder move.

This strategy could be implemented in a reasonable manner if engines implemented the algorithm you discussed where instead of being a consisting of a single move, the ponder move was actually a ponder move list which could contain anywhere from one move to all legal moves...
- - By rocket (***) [se] Date 2017-11-30 15:40
Ponder is always on during man-machine matches.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-11-30 17:44

>Ponder is always on during man-machine matches.


It could be switched off, of course, for the engine.
I could also promise not to ponder during the engine thinking time. :)
Parent - - By rocket (***) [se] Date 2017-11-30 20:06
btw, sometimes a program will pick an inferior move with greater depth, simply because the search rotates around  lines, and it may arbitrarly be attracted to continuation alternative "49", but only because that line was not investigated enough in the alloted time, and only looked better on the surface.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-29 08:37
I will post this here, as it has some relevance to computer chess and human-engine competition.

After doing an extensive study of the Fischer game collection, with the help of Stockfish, I just published a book on the theme:
https://www.amazon.com/gp/product/B078NSTR3K/ref=s9_acsd_newrz_hd_bw_b194_c_x_6_w?pf_rd_m=ATVPDKIKX0DER&pf_rd_s=merchandised-search-5&pf_rd_r=JNC827XRAWWDGW4HYCNC&pf_rd_t=101&pf_rd_p=17355592-3237-51ea-a6a2-f95fdeea28ea&pf_rd_i=4406

While going through the positional test suite, including 112 test positions, I had to ascertain that Stockfish still fails to solve around one third, but maybe even close to half of the puzzles. With the tactical set, Stockfish has no problems at all, all solved.

Make the conclusions yourselves how weak actually Stockfish is and how strong Fischer.

So, you might just want to forget anything about alleged engine superiority in chess.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-30 06:31
Thanks everybody for your contribution to the ponder puzzle.

Concerning SF and Fischer, here is one position from the positional suite SF fails to see:
[Site "New York ch-US"]
[Date "1965.??.??"]
[Round "4"]
[White "Fischer, Robert J"]
[Black "Benko, Pal"]
[Result "1-0"]
[Opening "Ruy Lopez, Closed, Breyer, Simagin Variation"]
[ECO "C95"]
[MLNrOfMoves "36"]
[MLFlags "010000"]

1. e4 e5 2. Nf3 Nc6 3. Bb5 a6 4. Ba4 Nf6 5. O-O Be7 6. Re1 b5 7. Bb3 d6 8. c3
O-O 9. h3 Nb8 10. d4 Nbd7 11. Nh4 Nb6 12. Nd2 c5 13. dxc5 dxc5 14. Nf5 Bxf5 15.
exf5 Qc7 16. g4 h6 17. h4 c4 18. Bc2 Nh7 19. Nf3 f6 20. Nd2 Rad8 21. Qf3 h5 22.
gxh5 Nd5 23. Ne4 Nf4 24. Bxf4 exf4 25. Kh1 Kh8 26. Rg1 Rf7 27. Rg6 Bd6 28. Rag1
Bf8 29. h6 Qe5 30. Qg4 Rdd7 31. f3 Bc5 32. Nxc5 Qxc5 33. Rxg7 Rxg7 34. hxg7+ Kg8
35. Qg6 Rd8 36. Be4 Qc8 37. Qe8+ 1-0


SF does not see 17. h4 wins here.

Why would SF miss that?
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-30 06:34
Is it possible to post fens here, not whole pgns?
Parent - By Labyrinth (*****) [us] Date 2017-12-31 00:41
Yes. Just paste the fen into the post window.

>This is the fen rnbqkbnr/pp1ppppp/8/2p5/4P3/5N2/PPPP1PPP/RNBQKB1R b KQkq, with '- 1 2' cut off the end, otherwise it will parse and generate a board..


rnbqkbnr/pp1ppppp/8/2p5/4P3/5N2/PPPP1PPP/RNBQKB1R b KQkq - 1 2
Parent - - By rocket (***) [se] Date 2017-12-30 22:02
I don't know why. Stockfish has advanced pawn storm algorithm nowdays and pushes kingside pawns quite aggressively, just as humans would.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-31 07:14

>I don't know why. Stockfish has advanced pawn storm algorithm nowdays and pushes kingside pawns quite aggressively, just as humans would.


Probably because it also has king shelter bonus for keeping its pawns close by the king, it is sufficient that the h4 square gets bigger bonus and some stupid search algorithm prunes the rest.
If you know SF code by heart, you can fix that in under a minute.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2017-12-31 07:16
Thanks for the tip how to post fens.

Here one more, already has been posted on Talkchess forum:
r3r1k1/pp1q1p2/2p2npb/PPPp1bnp/3PpN2/2N1P1PP/1R1B1PBK/3Q1R2 b - - 0 19


How many engines would see 19...Bg4?

This is from Nikolic-Fischer, Vinkovci 1968
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-01 07:43
Here another test from the positional suite:
rnbqk2r/1p2bpp1/p2ppn1p/8/3NPP1B/2N2Q2/PPP3PP/R3KB1R b KQkq - 0 9

  
Any self-respecting engine should find here 9...g5! for full equality. 10. fg5 hg5!(Nfd7 Qh5! is weaker, as black can not recapture with the h pawn towards the center) 11. Bg5 Nfd7, with nice outpost square for the knight on e5, pawn less, but central phalanx.
9...e5 instead might be losing after 10. Nf5, but the lines are long.
9... Nc6 is simply worse than 9...g5, as no full equality here.

Any self-respecting engines on this forum?
Parent - - By Carl Bicknell (*****) [gb] Date 2018-01-03 20:49

> How many engines would see 19...Bg4?


19...Bg4! 20.hg just loses for white. But 20.Qb1 is a tough nut to crack.
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-06 20:17

>19...Bg4! 20.hg just loses for white. But 20.Qb1 is a tough nut to crack.


Tough nut, but still losing.
Parent - - By Boombaard (**) [nl] Date 2018-01-01 16:26
Why would 17.h4 win? After 17...c4 18.Bc2 Nfd5! my Stockfish doesn't see a win.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-02 07:41

>Why would 17.h4 win? After 17...c4 18.Bc2 Nfd5! my Stockfish doesn't see a win.


Mine too.
But rest assured, it wins.
That is what analysis is for, and I have done my job.
I also have my very good intuition to help me.

Just give me your main line, and I will tell you how white wins/where to deviate.
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-02 07:43
Here one more puzzle for you(in the hope someone will look at my book):

r4rk1/4bppp/pqbp1n2/1p2pPB1/4P3/1BNQ4/PPP3PP/R4R1K w - - 0 15

  
The point is to find 15. Bf6!(which SF finds, or maybe first a4 b4, which does not change the overall character of the position and assessment) Bf6, and then 16. Bd5!, which is the real key move, and which SF fails to see. It prefers 16. Nd5? instead, which after 16...Bd5 17. Bd5 should only lead to a draw in an ending with opposite colour bishops.

Any engine seeing that sequence?
Parent - - By Boombaard (**) [nl] Date 2018-01-02 22:04
17.h4 Ne8 18.Ne4 Bxh4 19.Qf3 Nf6 20.Qh3 Nxe4 21.Qxh4 c4 22.Rxe4 cxb3 0,15
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-03 07:00

>17.h4 Ne8 18.Ne4 Bxh4 19.Qf3 Nf6 20.Qh3 Nxe4 21.Qxh4 c4 22.Rxe4 cxb3 0,15


Bram, is it you?

At first, you were suggesting 17...c4 and then Nfd5 as best defence, and now you have switched to 17...Ne8. :)
Obviously, your computer changes its score with larger depth.

Well, in any case, Bram Mourik posted on Talkchess forum, 4 out of 5 realistic defences after h4 should be losing,
and he thinks(his SF/Houdini actually, on a powerful hardware) 17...Ne8 might draw.

Well, I am not certain about that, I guess white should still be winning, but the position is very complicated and the lines
abundant and long, so that requires a lot of time to check.
Currently, I don't have much time for a deeper analysis, but, as soon as possible, I will post some feedback.

In the meantime, someone might find something promising.
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-03 07:01
In any case, win or not(I am still certain it is), 17.h4 is still the best move, and engines mostly fail to easily see that, so my puzzle is correct.
Parent - - By Boombaard (**) [nl] Date 2018-01-03 15:44
I'll await your refutation. Till then I consider the postion after 17.h4 as a draw.
Parent - - By Carl Bicknell (*****) [gb] Date 2018-01-03 20:15
I can't see a win here after 17.h4 either. In fact 17.Bc2, 17.Qf3 and 17.h4 all seem to be about as good as each other.
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-06 20:22

>I can't see a win here after 17.h4 either. In fact 17.Bc2, 17.Qf3 and 17.h4 all seem to be about as good as each other.


Not at all, h4 is by far the best, the problem is if it wins in all lines.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-06 20:20

>I'll await your refutation. Till then I consider the postion after 17.h4 as a draw.


Nice statement. :grin::grin:

I will be veryfiying it again, as soon as I get some time for it.
Does not 90% of lines winning qualify for a win? :smile:
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-06 20:24
Here one more:

3r1rk1/1p1n1ppp/pNbpp3/8/2P1P3/5P2/PPB3PP/2RR2K1 w - - 0 19


Single best winning move.
Can you find it?

This is from Fischer-Rossetto, Mar del Plata 1959.
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-06 20:25 Upvotes 1
Btw., if someone is interested, here one review for 'Human versus Machine':
https://www.expert-chess-strategies.com/human-versus-machine.html
Parent - - By Akbarfan (***) Date 2018-01-07 20:55
Nd5 :lol::lol:
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-01-08 09:27

>Nd5


Indeed.
I like the idea and the execution(a very surprising move), though Bram and Andreas, who are probably also present on this forum,
would claim this only leads to a draw with perfect play.
Well, maybe, but it is still the best possible move, by far, very elegant, so I guess it deserves to be included in a collection of Fischer's best moves,
don't you think so?
In any case, white should be winning at least 70% of all lines, most lines after cd5 ed5, most lines after Ne5 Ne3, for sure Ne5 Ne7 Kh8 Nc6 Nc6, etc.
White wins too many lines in this simple ending, although objectively, with perfect play, that might be very close to a draw.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-03-16 11:56
Currently, the 1st part is just 1 dollar: https://www.amazon.com/Human-Versus-Machine-Stockfish-Komodo-ebook/dp/B0768G8R2C/ref=tmm_kin_swatch_0?_encoding=UTF8&qid=&sr=
You can have the 3 for 7!
This will be just a limited offer, so don't miss it.
'The Secret of Chess' is also much cheaper.
Btw., anyone who has purchased the paperback of 'The Secret of Chess', can now download the ebook for free.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-04-17 07:12
Here is an interesting article on the book and anti-computer chess in general with 5 replayable games: https://glarean-magazin.ch/2018/04/14/anti-computerschach-human-vs-machine-report-carstens-tsvetkov/
In the lower left corner, there is a 'Translate' button, you can certainly read the English translation from German,
what you will understand is another matter.
In any case, the multitude of replayable games are interesting.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-08-10 10:14
Just to inform people that anyone with Kindle Unlimited can read for free
"Human versus Machine": https://www.amazon.com/Human-Versus-Machine-Stockfish-Komodo-ebook/dp/B0768G8R2C/ref=tmm_kin_swatch_0?_encoding=UTF8&qid=&sr=
Maybe someone will be interested.
Parent - - By Lyudmil Tsvetkov (**) [bg] Date 2018-08-13 09:42
Parent - By Lyudmil Tsvetkov (**) [bg] Date 2018-08-16 07:28
Up Topic The Rybka Lounge / Computer Chess / Human versus Machine
1 2 Previous Next  

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill