|
|
Post by matejst on May 31, 2022 22:16:20 GMT
The more I play against it, the more I find it interesting. What encourages me is that I can easily predict the lines Maia will chose: Maia 1900 is really positionally sound, but tactically as bad as most humans -- working hard enough you can trick her. Engines have probably a similar NN -- I don't even think it is stronger -- but they calculate much deeper, and it can confuse you -- you have to thoroughly analyze a position to understand why some ideas don't work.
I will try to use it outside of LucasChess, or at least to use its weights with Lc0, to have a better insight. But so far, the impression is good.
|
|
|
|
Post by Ozymandias on Jun 3, 2022 20:16:49 GMT
What time control do you use? Does it lose on time equal and favorable positions against you?
|
|
|
|
Post by matejst on Jun 3, 2022 21:38:36 GMT
What time control do you use? Does it lose on time equal and favorable positions against you? I play at 50' + 30" against 30'+30"; lately I switched to 90'+30" for me. The engine has enough time not to lose on time, same for me. I avoid playing blitz -- it does not help you much improving. |
|
|
|
Post by matejst on Jun 5, 2022 9:58:22 GMT
The more I play against it, the more I find it interesting. What encourages me is that I can easily predict the lines Maia will chose: Maia 1900 is really positionally sound, but tactically as bad as most humans -- working hard enough you can trick her. Engines have probably a similar NN -- I don't even think it is stronger -- but they calculate much deeper, and it can confuse you -- you have to thoroughly analyze a position to understand why some ideas don't work.
I will try to use it outside of LucasChess, or at least to use its weights with Lc0, to have a better insight. But so far, the impression is good.
Update: After playing several games at LTC against Maia 1900... What did I write: "working really hard you can trick her"? Then, I don't work hard enough. Yes, one can trick her, but I don't seem to be this "one". When I analyze the positions I get from the opening, I reach usually very good positions, and different engines give me a clear advantage. But after that, converting this advantage is hard.
At the time controls I use Maia, she calculates 7-10 plies deep. The weights are different than the Lc0 weights, it should play weaker, but really... how much weaker? So, I started cheating and playing... Centaur chess. I check my ideas with Rebel (I analyze finished games with Eman) and see why they don't work. Anyway, it is a frustrating but useful learning experience. My openings are robust, the bishop pair is really an advantage, Rebel plays well positionally, and I slowly improve. The downside is that I started checking variations a bit too often with an engine. I have to restrict the number of times I cheat.
I think I wrote it already somewhere here, but Maia 1900 at 30' plays well over 1900. It is FM level at least. I faced FMs and IMs at LTC, it is somewhere there. There is probably a dip in endgames (where Maia does not calculate deep enough), but that's all.
To answer your question: I noticed that after a certain point, the time controls don't have any influence on Maia's use of time and game. 30'+30" seems to be too much for the engine, and I could shorten this TC.
|
|
|
|
Post by Ozymandias on Jun 5, 2022 10:28:02 GMT
Sounds to me like what you're finding out is that... chess is difficult. For my part, I keep focusing on tactics, it's too much infuriating for me to outplay my opponent strategically, to then drop a piece. On that front, I finally breached the 2000 barrier at ChessTempo, where I had 1700 just three years ago.
|
|
|
|
Post by matejst on Jun 5, 2022 11:46:20 GMT
Sounds to me like what you're finding out is that... chess is difficult. For my part, I keep focusing on tactics, it's too much infuriating for me to outplay my opponent strategically, to then drop a piece. On that front, I finally breached the 2000 barrier at ChessTempo, where I had 1700 just three years ago. When I am disciplined and do my work the whole game, playing against Maia is a great tactical training. The other thing I can advise is to do the tactical exercises in Lucas chess. They are not too difficult, and there are a lot of different type of exercises. I also work with the old Convekta software "Chess tactics for intermediate players", which is a whole other level of difficulty. CTIP gives you a rating too -- just to stimulate you. I did these exercises for one hour a day for months, and it helps tremendously both on the plans of awareness and ideas. Add to this playing LTC once a week and you'll gain 200 Elos in just a few months.
Yes, chess is difficult. Against Maia 1400-1700 I had only to wait for the tactical error. Now, I am forced to play complicated endgames, to gain experience in the middlegames in my new repertoire, against an opponent that, comparing it to the players I faced, gives you so little.
It is frustrating -- not because chess is suddenly difficult, it always was -- but because I don't feel that I calculate faster, deeper, and while my understanding of certain positions has clearly improved, my technique is still awful. In simple positions, I can't say that I am lost, but suddenly I am often completely without ideas. I don't know how to combine attacks on several weaknesses. I also lack the ability so far to see what I want to achieve, what should be the next maneuver, e.g.: the rook is best on this case, the place of the knight should be there, etc. I just keep on searching, calculating blindly trying the usual stuff -- a threat, or pushing a pawn. Then, I consult the lines given by Rebel, and it becomes clear to me what the plan is, where this rook should go, what is the best place for the knight.
I still am just another patzer.
|
|
|
|
Post by matejst on Jun 15, 2022 20:26:41 GMT
Very happy! I played a very good game against Maia 1600. Just one big mistake, but lots of good moves, ideas, and a creative attack! Of course, Maia could have played better, me too, but I worked hard, and the tactical exercises I did helped a lot. I felt I understood the position. I was not watching the board without ideas.
|
|
|
|
Post by Ozymandias on Jun 16, 2022 5:14:56 GMT
Define "mistake". Is it a blunder? (As in a tactical one leading to loss of material).
|
|
|
|
Post by matejst on Jun 16, 2022 10:09:08 GMT
Define "mistake". Is it a blunder? (As in a tactical one leading to loss of material). Here is the game with some comments.
[pgn]
[Event "Play against an engine"]
[Site "Lucas Chess R 1.30d"]
[Date "2022.06.15"]
[White "Stanojevic, Boban"]
[Black "Maia-1600"]
[Result "1-0"]
[ECO "B00"]
[TimeControl "1800+30"]
[TimeExtraWhite "3600"]
[TimeStart "2022.06.15 21:15:45"]
[Termination "Mate"]
[Opening "King's pawn opening"]
[PlyCount "75"]
[TimeEnd "2022.06.15 21:56:21"]
1.e4 b5 2.d4 Bb7 3.Bd3 e6 4.Nf3 a6 5.O-O c5 6.c3 cxd4 7.cxd4 Nf6 8.Re1 Be7 9.Nc3
O-O 10.a3 d6 11.Qc2 Nbd7 12.e5 dxe5 13.dxe5 Nd5 14.Bxh7+ Kh8 15.Be4 Rc8 16.Qd1
Nc5 17.Ng5 Bxg5 18.Qh5+ Bh6 19.Bxh6 Nxe4 20.Bg5+ Kg8 21.Bxd8 Nexc3 22.Bf6 gxf6
23.exf6 Nxf6 24.Qg5+ Kh7 25.Re3 Ne2+ 26.Kh1 Nf4 27.Qxf4 Nh5 28.Qg5 Rg8 29.Qxh5+
Kg7 30.Qe5+ Kf8 31.Qd6+ Ke8 32.Rxe6+ fxe6 33.Qxe6+ Kf8 34.Qf6+ Ke8 35.Re1+ Be4
36.Rxe4+ Kd7 37.Re7+ Kd8 38.Qd6# [/pgn]
16. Qd1 was a blunder. I intended to play B:d5 first, then Qd1 and keep my extra pawn. Then, after thinking several minutes, I was happy that I found some ideas with Qd1, Ng5, that I forgot to play with the bishop first.
16... Nc5 was a mistake -- N:c3 wins the pawn back with a better position for black. From there, I saw Ng5, Qh5+, and, I thought Maia would play 19...g6. The reason why I was so satisfied, I also saw 22.Bf6, although I could not calculate everything from there, then 30.Qe5 with the mating attack.
|
|
|
|
Post by Ozymandias on Jun 16, 2022 13:17:11 GMT
Looks more like an strategic mistake to me. A big one, but not a blunder. Blunders are for when you have the chance to obtain enough material to win (or a mate sequence). Here you were wasting a crushing positional advantage, by placing the Queen on a square which allowed black to exchange Knights with tempo and then the Bishops on e4. There's no attack for white after those exchanges, and black's pawn structure is better. No material loss tough, but Nd4 was overwhelmingly better, Nxc3 makes little sense with the d1-h5 diagonal free.
|
|
|
|
Post by matejst on Jun 16, 2022 13:44:08 GMT
Nonetheless, it really was a blunder. I started the variation with the second move. I did not even take Nd4 in account, btw, I saw it later in the analysis. My idea was 16.Bd5, Bd5 17.Qd1 and then Ng5.
And, another thing. I analyzed the game, later, with Hiarcs 8 Bareev. I feel much less stupid when I use these old engines instead of new ones. SF immediately sees 40 plies deep, has a +3,54 or something alike when I have the impression of facing a hard task of making a draw. H8 sees 10-12 plies deep, a bit more than I do, and I can check for obvious blunders. I feel less stupid.
|
|
|
|
Post by Ozymandias on Jun 16, 2022 14:29:53 GMT
I don't let them intimidate me with their evals, I always follow the line to see what kind of advantage we're talking about. For example, I played a tournament this weekend and my best game (against the child who ended 2nd) was a draw because I missed a winning continuation SF immediately evaluates as +6, but it's NNPPvQ ending I would never go for. In fact, the game was agreed a draw with a QNPPvQBN, and with less of a disadvantage, that position was indeed 0.00.
|
|
|
|
Post by matejst on Jun 16, 2022 15:30:52 GMT
I don't let them intimidate me with their evals, I always follow the line to see what kind of advantage we're talking about. For example, I played a tournament this weekend and my best game (against the child who ended 2nd) was a draw because I missed a winning continuation SF immediately evaluates as +6, but it's NNPPvQ ending I would never go for. In fact, the game was agreed a draw with a QNPPvQBN, and with less of a disadvantage, that position was indeed 0.00. I wrote already about the misleading effect of hyper-fast engines' evaluation. In retrospect, one has to be very careful about using engines for his own improvement. And there is the question: which engines? With time, I tend to think that the slower they are, the better -- with the conditio sine qua non of a good positional play. Anyway, for me personally, the main problem is the "lack of effort" problem -- I tend to rely on engines too much when I should analyze for myself. For months, it has been a fight not to use them, to rely sparingly on them. They introduce doubts in positions I haven't played: "Oh, what's the best plan now? How should I play this kind of positions? Let's ask SF, or Rebel, or..." Or in positions I think are worse, but they show are better or equal -- indeed, they helped me understand how to use the bishop par, how activity often is more valuable than a good pawn structure (or even a pawn), but I feel like my niece, who avoided to solve maths problems whenever I was present, because it was easier just to ask for the solution. A better way should have been to play these positions against them with both colors, on both side, but... lack of effort. It is easier to write something or read instead.
When we are at it... There were these unending discussions about clones on CCC, but when I compare a bit the evaluation of Hiarcs 8 with some of much newer HCE engines, Mark Uniacke was obviously years ahead in this area. When we know all the tricks professional authors used to gain Elos or to win tournaments... I would not be surprised if some of his ideas -- even code -- was not present today in other, even open source engines.
|
|
|
|
Post by Ozymandias on Jun 25, 2022 7:14:22 GMT
Speaking of tactic problems, here's the last one I failed. Rated 2170 at Chess Tempo. I saw all the moves and themes, except for a mate in 4 which allows quite a nice fork. Try it before hitting "next", white to move:
[ChessTag]{5bk1/6pp/3qP3/p2Pp3/2Q4P/P4RP1/5PK1/3r4}{5bk1/4P1pp/3q4/p2Pp3/2Q4P/P4RP1/5PK1/3r4}{6k1/4b1pp/3q4/p2Pp3/2Q4P/P4RP1/5PK1/3r4}{6k1/4b1pp/3q4/p2Pp3/Q6P/P4RP1/5PK1/3r4}[/ChessTag]
SPOILER: saving the rook on d1 leads to mate.
|
|
|
|
Post by matejst on Jul 4, 2022 19:20:05 GMT
Last time I did not see the diagram. Now it is here. Of course, with your instructions, the tactical motif was easy to find. The problem in this kind of puzzle is that you always assume that there is a forced mate. When I solve puzzles, I always found it much, much easier when I could read the kind of problem it was.
Meanwhile, I slowly find the joy of playing. I still blunder, and very often (yesterday was an example, when I missed a potential queen sac) I make a move, and then see new possibilities for my opponent. But I see more, I am more aware of tactics. I take it slower, restarting with Maia 1500, and the fact that these weaker Maias blunder a lot helps me stay focused trying to find a mistake from their side and searching more moves.
I toyed with the settings a bit to find how much my opponent should play from the book, found some adequate settings. I am not where I want to be yet, but I am improving.
|
|
