To post one more test from the positional Fischer suite:
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?
Amazon published my review of your book. I saw now several grammatical errors.
https://www.amazon.de/gp/product/B0768G8R2C/ref=oh_aui_d_detailpage_o01_?ie=UTF8&psc=1
Lyudomil, it is good that your ideas for an evaluation function have been enthusiastically received. However, you rashly claimed that your ideas are so good that if implemented they would result in a 4600 Elo engine, which is ridiculous.
Unfortunately, your claim to have been singlehandedly responsible for a 200 point improvement in Stockfish appear almost as implausible, given the graph of its Elo over the years:
Could you clarify in which release(s) your ideas were incorporated?
Starting from SF 4, but even before that, I published my Little Chess Evaluation Compendium in early 2010, and all programmers read it, of course, Houdini, for example. Marco and Joona for sure too.
In late 2012, before SF 3 and the Framework, SF already was implementing some of my ideas(I published a revised extended edition of the compendium in 2012), for example bonus for rooks on advanced ranks.
Then, a bunch of ideas were certainly incorporated in SF 3.
My real contribution however started with SF 4 and DD, when, based on my suggestion on the Talkchess forum, SF added bonus for connected pawns(defended and duos) in terms of rank. That was a major break-through, which thoroughly changed SF's playing style from attacking to positional attacking, due to the nice space bonus in terms of advanced pawns.
This, and some other 10 or so ideas I suggested were implemented in 2014(will not enumerate them all, you can find some of them in the introduction to 'The Secret of Chess'), which helped SF become the TCEC champion for the first time, the new number 1! I doubt this would have happened without my contributions, so I/we were the champions then.
New patches, based on my sugesstions, were implemented in SF 6 and 7.
I did not contribute directly to SF 8, but my ideas were still there to influence the SF developers, and indeed, some of the evaluation terms in SF 8 are based on my prior ideas.
I have gifted copies of 'The Secret of Chess' to some SF programmers, but they have made such a mess of the code in the latest years, with so many redundancies, that it is difficult to implement anything sensible at all.
Apart from that, you need examples to understand what is going on, and that is what I have been providing in earlier SF versions, but no one to do that now.
Please note, that you will find 30 or so threads on Talkchess and Fishcooking, where SF programmers thank me for a specific patch that has just succeeded, but my name is mentioned just couple of times officially on SF pages.
You really don't need to get that much official, do you? I guess we are all friends, in the SF Framework.
It's great that your ideas on quantifying positional factors have been useful to the Stockfish developers - I am convinced of this - but it would be very hubristic to claim that you are responsible for all of the increase in rating between Stockfish 4 and 8 (which amounts to 178 Elo points in total). Do you really think there were no other improvements and that no-one else deserves any credit, including all of the programmers?
The truth is that despite the success of Stockfish, the hand-crafted evaluation function approach is not the way forward! This is limited by human understanding and imagination. The route to future improvement is evaluation functions that are learnt, as in the case of AlphaZero.
It is surprised that more supervised learning has not been used before DeepMind took a leap beyond that. If the sort of features you describe are valuable (which they are) this can be learnt by a general purpose machine learning algorithm.
I also wrote a review in Chessable (again, a lot of grammatical mistakes, sorry about this); this is an updated version of it:
Tsvetkov, the guy who can defeat Stockfish
Torrubirubi PRO 3,010,940
I am curious to hear if you read one of Lyudmil Tsvetkov's books on chess. Although his books received some negative reviews in Amazon and chess.com-forum, I think it is worth to get them. I recommend the books "Human versus Machine". This is a collection of his winning games against Stockfish and Komodo. The games were played in the period 2013-2017 against different Stockfish versions and Komodo. He played over 50 thousand engine games. The time control for the game differs, but those are mostly blitz or rapid games, with the author having 2 times more on average on the clock. As he wrote, there are no takebacks, and time control has been strictly applied.
After going some of the games I see some common patterns. Tsvetkov usually play rather quite openings, tried to close the centre and attack the king with a pawn-storm, often after having castle on the kingside and attacking on the kingside. The games are highly interesting from a strategic and tactic point of view. The guy really understand the weakness of these engines and know how to explore small (sometimes not that small at all) inaccuracies and convert them to great positional advantages.
Another book is "The Secret of Chess". It is rather unusual, you have to get used with his style and his allegedly poor English (but better than my English, so I don't care) and evaluations based on centypawns. IMO the book is really useful to learn strategy. He often gives a diagram, not telling which side is to play, and show his evaluation, often correcting engines' evaluations.
Although a lot of weak players are killing the author in chess.com-forum, there is a surprising positive review by the David Smerdon (hey, also a Scandi-specialist, but with the Portuguese Gambit and not ...Qd8). By the way, Smerdon is not exactly somebody you can call a patzer: he is a GM and the second highest ranked chess player of Australia.
In any case, if you are interested in insightful but unconventional chess books, get a copy of them (but be sure that you read first a sample in Kindle to see if you will accept his style - don't blame me afterwards if you think the book is different from all your other chess books - it is, but in a positive way, IMO).
Tsvetkov believes that close games are the future of chess: "they exhibit the deepest lines and most refined positional characteristics. Than is one of the reasons top engines still have considerable difficulties with them".
This guy is a kind of human AlphaZero, and I have a huge respect to his dedication to improve his game against the engines and his readiness to share is knowledge with us!
It is really sad to see how people criticise him (some rather even below 1000 in chess.com), often without even having a look on his books. But should be really surprised about this? Remember when the chess player Radi Danov was accused of cheating? IM Valeri Lilov did a video showing several evidences that Danov was actually using an engine to defeat much stronger players than him, on the one hand by the obvious coincidence between his lines and the lines given by an engine, on the other hand by the fact that even his draws were arranged by an engine. And you know what? A lot of people commented the YoutTube video defending Ivanov - they simply could not understand the arguments showing that Ivanov was cheating. For them was absolutely normal that in one day Ivanov was playing something like a GM, in the next day (without the help of an engine) he was playing something like a 2000-player or worst. The video had 191 likes, but 129 dislikes! By the way, Danov has been later banned by BCF after refusing a metal detector search - of course. Lilov's video is a classic, but also an impressive example of the huge gap concerning chess understanding between strong and weak players (for me, one of the most interesting arguments showing that Danov was using an engine is the way how Danov was playing for a draw - exchanging everything in a very systematic way, completly different from what you are used in the draws between human players.
What happen here is something similar, but now we have a guy able to play and to analyse his games in a high level is accused be a cheater, and this often by weak players who mostly did not read any of Tsvektov's books. This is a sad and tragic thing.
What percentage score would you estimate you would achieve against Stockfish in a straight match, Lyudmil? (You can pick the time control and take as much rest between games as you like).
@torrubirubi: Nice review, but in my verification, SF plays different (same version of course), so the claim LT can defeat it (and also AZ if you read his recent posts) is delusional.
What percentage score would you estimate you would achieve against Stockfish in a straight match, Lyudmil? (You can pick the time control and take as much rest between games as you like).
Such a match will never take place, because
1. objective, see my previous post
2. subjective he fears celebrity, like Fischer (!) just read his posts.
Amazon published my review of your book. I saw now several grammatical errors.
https://www.amazon.de/gp/product/B0768G8R2C/ref=oh_aui_d_detailpage_o01_?ie=UTF8&psc=1
Thanks Torrubirubi, excellent!
I will drop you a message later on.
Lyudomil, it is good that your ideas for an evaluation function have been enthusiastically received. However, you rashly claimed that your ideas are so good that if implemented they would result in a 4600 Elo engine, which is ridiculous.
Unfortunately, your claim to have been singlehandedly responsible for a 200 point improvement in Stockfish appear almost as implausible, given the graph of its Elo over the years:
Could you clarify in which release(s) your ideas were incorporated?
Starting from SF 4, but even before that, I published my Little Chess Evaluation Compendium in early 2010, and all programmers read it, of course, Houdini, for example. Marco and Joona for sure too.
In late 2012, before SF 3 and the Framework, SF already was implementing some of my ideas(I published a revised extended edition of the compendium in 2012), for example bonus for rooks on advanced ranks.
Then, a bunch of ideas were certainly incorporated in SF 3.
My real contribution however started with SF 4 and DD, when, based on my suggestion on the Talkchess forum, SF added bonus for connected pawns(defended and duos) in terms of rank. That was a major break-through, which thoroughly changed SF's playing style from attacking to positional attacking, due to the nice space bonus in terms of advanced pawns.
This, and some other 10 or so ideas I suggested were implemented in 2014(will not enumerate them all, you can find some of them in the introduction to 'The Secret of Chess'), which helped SF become the TCEC champion for the first time, the new number 1! I doubt this would have happened without my contributions, so I/we were the champions then.
New patches, based on my sugesstions, were implemented in SF 6 and 7.
I did not contribute directly to SF 8, but my ideas were still there to influence the SF developers, and indeed, some of the evaluation terms in SF 8 are based on my prior ideas.
I have gifted copies of 'The Secret of Chess' to some SF programmers, but they have made such a mess of the code in the latest years, with so many redundancies, that it is difficult to implement anything sensible at all.
Apart from that, you need examples to understand what is going on, and that is what I have been providing in earlier SF versions, but no one to do that now.
Please note, that you will find 30 or so threads on Talkchess and Fishcooking, where SF programmers thank me for a specific patch that has just succeeded, but my name is mentioned just couple of times officially on SF pages.
You really don't need to get that much official, do you? I guess we are all friends, in the SF Framework.
It's great that your ideas on quantifying positional factors have been useful to the Stockfish developers - I am convinced of this - but it would be very hubristic to claim that you are responsible for all of the increase in rating between Stockfish 4 and 8 (which amounts to 178 Elo points in total). Do you really think there were no other improvements and that no-one else deserves any credit, including all of the programmers?
The truth is that despite the success of Stockfish, the hand-crafted evaluation function approach is not the way forward! This is limited by human understanding and imagination. The route to future improvement is evaluation functions that are learnt, as in the case of AlphaZero.
It is surprised that more supervised learning has not been used before DeepMind took a leap beyond that. If the sort of features you describe are valuable (which they are) this can be learnt by a general purpose machine learning algorithm.
As said, my contributions already started with SF 2.62.
Some are documented, of others I simply know.
I guess the elo gain between SF 4 and 8 is bigger, at least 300 elos or so(SF DD was another intermediate version, so you have 5 instead of 4) at 1 min. + increment, the way they test on the Framework.
Is that really important if it is 200, 150 or somewhere close?
Sometimes, you can never progress without specific knowledge features being implemented, for example, before SF became really strong attacker, it applied the following 2 pawn storm patches, based on my ideas:
- differentiation of bonus for storming pawns in terms of whether the pawns are blocked or not
- penalty for a/h file edge storming pawn, blocked by opponent king, for example white pawn on h6, black king on h7
Without those 2 patches, reflecting real chess knowledge, SF would have extreme deficiencies in its attacking code, and you possibly can not lessen those with better search. So, you simply get stuck at some point, and if you don't have a knowledgeable person to guide your chess evaluation, you simply remain there.
So, mine is just a general assessment, and I know what I am talking about.
And what about other engines, who have implemented many of my ideas and became very strong because of that? Andscacs, Texel, Rodemt, to name just a few.
And what about Komodo and Houdini? They are following the progress on the SF Framework too.
Did not you see Mark Lefler posting a review on my book on Amazon? Why do you think he posted it?
So that, I know what I am talking about, when I say 200 elos.
Btw., Amazon slashed the paperback to just 11 bucks, maybe because of the holidays:
https://www.amazon.com/Secret-Chess-Lyudmil-Tsvetkov/dp/1522041400
What percentage score would you estimate you would achieve against Stockfish in a straight match, Lyudmil? (You can pick the time control and take as much rest between games as you like).
Well, I told you, I am very good when it is quiet, but apart from that:
- I am saving a lot of energy when playing in domestic circumstances, as I am relieved of the pressure to think how the game is going to end
- I save a lot of energy, as I don't care what the opponent thinks of me, etc.
So that, this saves a lot of energy and gives a nice psychological boost, enabling you to beat even the strongest.
Apart from that, I told you, for me, by far the most important factor is deep concentration, if I have that, I am really very very strong; failing that, I play much weaker, sometimes like a patzer.
Most of my lost casual games are due to kibitzers, suggesting moves and commenting during the game, so I really hate kibitzers.
If I am in good form, I can do very well, drawing against SF is not a problem, winning is.
@torrubirubi: Nice review, but in my verification, SF plays different (same version of course), so the claim LT can defeat it (and also AZ if you read his recent posts) is delusional.
It is you who are delusional - give specific examples.
In any case, you don't have precisely the same hardware I have, so I am at a loss how you are doing your checks.
Just post one single position from my games in 'Human versus Machine', where the engine played a very weak/strange move or where you think it understands the closed position better than me.
On the contrary, you will see(and I can post them) 100 test positions of closed character, where SF invariably does not see my winning moves.
What percentage score would you estimate you would achieve against Stockfish in a straight match, Lyudmil? (You can pick the time control and take as much rest between games as you like).
Such a match will never take place, because
1. objective, see my previous post
2. subjective he fears celebrity, like Fischer (!) just read his posts.
That is what you think.
I am eager to crush the hell out of a 200-elo stronger SF/Komodo, but that will not be immediately.
Anyone who read your posts can see it is not "what I think" but exactly the situation. Only games you won are own set-ups against SF.
What percentage score would you estimate you would achieve against Stockfish in a straight match, Lyudmil? (You can pick the time control and take as much rest between games as you like).
Such a match will never take place, because
1. objective, see my previous post
2. subjective he fears celebrity, like Fischer (!) just read his posts.
That is what you think.
I am eager to crush the hell out of a 200-elo stronger SF/Komodo, but that will not be immediately.
I'm looking at the keyboard and trying to figure out how trying to type the word "ever" ends up as "immediately".
Hi, I wrote a review on the book in Amazon.com. It is an update from a review I wrote in Amazon.de.
I copy pasted the text here:
Most chess players noways avoid playing against strong engines like Komodo or Stockfish. With their fully strength, these engines are basically invincible. I am not talking about chess amateurs: it is well know that even strong GMs have problems to win any game against them, even with material odds. We are know the stories of cheaters using engines and winning tournaments, ahead of several GMs.
Stockfish has so a strong reputation among chess players that the fact that this engine was recently clearly defeated by another program (AlphaZero) was perceived as one of the most unforeseeable events in modern chess history. The shock was even stronger due to the fact that AlphaZero won after only four hours of self-play. AlphaZero didn't lose a single game!
There were probably only few people not surprised by the win of AlphaZero: people working in the DeepMind team (the developers of AlphaZero) and Lyudmil Tsvetkov, the author of "Human Versus Machine: How to Beat Stockfish". Tsvetkov shows why strong chess players still have a chance when playing such strong engines . What Tsvetkov did can perhaps be considered an even more amazing event: After playing and analysing games against Stockfish and Komodo (more than 50,000 games!) he is able to defeat regularly these engines in blitz games. In his book he commented 15 winning games played against the silicone monster Stockfish 4.
After going some of the games I see some common patterns. Tsvetkov usually play rather quite openings, tried to close the centre and attack the king with a pawn-storm. By the way, in the first games that I went through both "players" castled on the kingside, but Tsvetkov did not see as a problem to send his kingside pawns in the direction of the enemy king. The games are highly interesting from a strategic and tactic point of view. The guy really understand the weakness of these engines and know how to explore small (sometimes not that small at all) inaccuracies and convert them to great positional advantages.
The book can certainly be used to improve strategic game, but also show how to attack. Also his book "The Secrets of Chess" is highly recommended. I decided already that I will put all my chess books on strategy aside and use exclusively Tsvetkov's books in order to improve my game in 2018. As a weak player, I am basically learning new things in every page of his books. But I can imagine that even strong players will find several useful insights in his books, especially his "The Secrets of Chess" (if you are interested there is for example a favourable review by the GM David Smerdon - you can google "davidsmerdon.com The Secrets of Chess").
There is also something worth mentioning. There are some very negative reviews of the book, both in Amazon but also in chess.com, where Tsvetkov was so kind to present his books in the forum. Most of these negative reviews were probably written by people who did not really read the book, or in some cases these people did read it but were too biased to understand the value of the book. Some people for example criticised his English , or they are irritated by the fact that he is not a GM and at the moment not an active chess player.
Concerning the English I have to say that I do not see any problem to understand him - well, this has probably to do with my own rather poor English (as you can see here). But if I buy a chess book I am more interested in the chess content; for high standard English literature we have books written by guys like Alexander Dumas or James Joice.
Concerning the second criticism: several very strong coaches or experts in different aspects of the game are not GMs, some even haven't any title in chess. The Russian Mark Dvoretsky (1947-2016), probably the most famous chess coach in modern chess, was "only" an International Master (although a strong one). There was a time where most all players in top 100 trained at least for a while with him, and I am sure that they hardly noticed that he was not a GM, they were too occupied learning from him.
Although of course the vast majority of IMs and GMs earned their titles with hard work and at least some talent, we should also not forget that it seems that at least some chess players gained IMs or GMs titles by cheating, as for example by faking whole tournaments!. And according to some sources it seems also possible to pay a GM to lose intentionally a game and allow his opponent to get a title norm. We should keep this in mind when discussing the sense or nonsense of titles in chess, and the qualifications an author need to write a chess book. We probably should appreciate Tsvetkov's efforts to understand the engine's weakness and explore them in a time where he could play tournaments and make GM norms. It is probably impossible for most of us to dedicate fully to two different challenging projects.
There is another kind of criticism. Some people argue that what he is doing is just not especial, as strong engines are patzers concerning close, positional games: you have only to close the centre (probably weak players even don't know exactly what this mean), put arbitrarily your pieces somehow more or less in the direction of the enemy king and a win is unavoidable. Yes, sure. Let's say what really strong chess players has to say about playing against engines, as for the example John Nunn:
"The computer dominance was only emphasised when, in 2005, Michael Adams lost 5 1/2 -1/2 in a match against Hydra. This event demonstrated that computers can play extremely well not only in wild, tactical struggles, but also in quite positions." [...] "Computers are very useful for chess preparation and analysis and, if used properly, for training. What you should not do is play games against them, since the result is bound to be demoralising."
Nunn gave up playing against engines in 1995. This is 23 years ago, in a time when engines were much "weaker" than they are now! For people who don't know John Nunn, he cannot be exactly considered as a patzer, as he is a strong GM and had in this time an ELO of 2630.
Let's summarise: Tsvetkov's skills to defeat Stockfish and Komodo are not banal, and the way how he does it and his explanations of the relevant strategy and tactics is very insightful. If you want to improve your game I recommend you to purchase all his books and learn them carefully.
Anyone who read your posts can see it is not "what I think" but exactly the situation. Only games you won are own set-ups against SF.
At least I have my own setups.
Other people don't even have that.
I mean, I have my WINNING setups against the top engines.
How many GMs have those?
Kasparov gave up long time ago.
What percentage score would you estimate you would achieve against Stockfish in a straight match, Lyudmil? (You can pick the time control and take as much rest between games as you like).
Such a match will never take place, because
1. objective, see my previous post
2. subjective he fears celebrity, like Fischer (!) just read his posts.
That is what you think.
I am eager to crush the hell out of a 200-elo stronger SF/Komodo, but that will not be immediately.
I'm looking at the keyboard and trying to figure out how trying to type the word "ever" ends up as "immediately".
Those are 2 very different words: one consists of 4, and the other of 11 letters.
What concerns my 5000 elos claim, of course, such an engine could be built, using the concepts in 'The Secret of Chess', no doubt about that.
SF has 3 to 4 times few chess knowledge parameters currently, and the more good evaluation parameters precisely tuned, the better the engine, so my estimate should be more or less correct.
The difficulty is how to implement those and avoid redundancies.
Improving engines at the top, like Houdini, Komodo and SF are trying to do, is not an easier task than, say, a human beating those very same engines. You have to know the code by heart, be a very good chess player(or take the knowledge from somewhere) and conduct endless tests, really very very difficult.