of a chess position. Of course there are many different definitions of complexity
and many different reasons for wanting such a measure. For example, when
analyzing OTB games and looking at the blunders it would be interesting to see
if those happen in simple or complex positions and then adjust the training
I came to think about this topic while evaluating which of several moves to
choose when IDeA gives the same or very similar evaluations. Often against
an opponent that I think might be weaker than me (or has less time left or
plays many more games at the same time etc. ) I would like to choose the more
complex game in order to have better chances to actually WIN the game. Also
when allocating CPU time to different games, the more complex ones should
probably get more time.
Following are some ideas:
 Number of pieces on the board:
Obviously a very simplistic approach but often the move not leading to
many exchanges will be preferred when trying to avoid a dead position, so
maybe it is not so bad after all...
 Difference in evaluation of the best move and the second best move
Certainly useful in determining how much time to spend on analyzing
a certain position but how to account for the fact that there are a often
(semi-)forced moves that eventually lead to a really complex position.
 Time an engine needs to reach depth X
More pieces and more possibilities certainly slow engines down in terms
of reaching a certain depth. Again forced moves distort the picture of
 IDeA Branching
I think this is very useful but it is hard to standardize this measure
and to come to statements like "of all my current correspondence games,
this game is the one with the highest complexity and therefore I will
allocate more time to it".
I guess it must be a combination of some sort. Maybe executing the "easy"
moves (e.g. where the difference between the best and the second best is above
a certain threshold) and then looking how long an engine takes to reach a
The first challenge is to precisely define what you mean by complexity, as this could be looked at in different ways.
Example: think in EGTB terms for the whole game, assuming you had 32-piece EGTBs. In how many positions must a player play one and only one move in order to achieve a winning outcome or avoid a losing one? Clearly, the more legal moves there are available, the more likely the wrong move would be played. Secondly, how many moves would be required to achieve mate or compel a draw? The further away, the more complex. When a person starts thinking in such terms terms he rapidly enters the realm of theoretical mathematics.
>Clearly, the more legal moves there are available, the more likely the wrong move would be played.
Not necessarily, whether a position is sharp or complex does not need to be related to how difficult it's to play. Players actually perform excellently in sharp positions because even while the moves of the opponent are very few, they are hard to find, so the player blunders and the other wins. The losing side may think the position is very complex but I think those positions should be defined by the winning side.
>Secondly, how many moves would be required to achieve mate or compel a draw?
I don't think line line length should be necessary to measure this, number of non-transposing variations should be enough, regardless of how long they are.
>The further away, the more complex.
I disagree, of course the more they go on the chances the more lines are there, but if they are decided games that go to very deep, they aren't going to be very complex. I think those 7men tablebases endgames that are won and go on for ages playing moves go into this (they're the sharpest.)
> The first challenge is to precisely define what you mean by complexity, as this could be looked at in different ways.
I, too, have thought quite a lot about this topic and have settled on the idea that a complex position is one with latent possibilities. Games of advanced chess are often decided via incremental improvement over the course of many moves so I feel it's important to reach positions that are conducive to error. Before I lost my taste for computer chess, my primary goal in every correspondence game was to make it difficult for a lazy player to simplify the position.
Regarding #3 in the original post, one of the reasons I have little faith in engines like Houdini and Stockfish is that they tend to breathlessly reach extreme depths even when analyzing the most difficult positions. In such instances, I can't shake the feeling that engine is pruning its way into trouble. With an engine like Rybka, or the venerable Zappa Mexico, it's always quite obvious when the engine is chewing on a difficult position and I find that strangely comforting.
> Real men drink Scotch and don't use Facebook.
That is REALLY comforting as I do *not* have a Facebook account (because I think it is narcissistic and gay) and I *do* drink scotch (because it goes great with a cigar). In fact, I have a bottle of 18-year-old stuff in the pantry right now. I drive a Jaguar, too. Now, how have I managed to squander all this hyper-machismo? With computer chess, what else. That drives testosterone levels to rock-bottom and emasculates us all. Experts agree, as Alan might say.
Your thoughts on the original question are interesting, but I think in the second paragraph you betray a preference for breadth (or thoroughness) over depth. That's characteristic of a correspondence player. Engines are optimized for engine play because that's what moves merchandise. As computer power escalates I sense the advantages of greater depth will diminish (in non-endgames) relative to those that might accrue from less pruning. This isn't evident now because of the efficient testing methods Vas himself introduced to the hobby which everyone else has copied. (Good ideas tend to spread, bad ones tend to die out.) I suppose the alternative to testing hundreds of thousands of games at instantaneous time controls is finding a suite of super-hard positions and having the engine solve a higher percentage of them than anyone else. Then you have to figure out how to do it faster.
> I do *not* have a Facebook account (because I think it is narcissistic and gay)
Huh? What does having a facebook account have to do with sexual preference? Or is gay the new lame?
I just had to create one to play some online games, but I never visit it nor have any personal data in there, I wonder if that makes me narcissistic.
In their day, men didn't type unless it was directly related to their job. They gave dictation to secretaries! Tell me that life is really better now.
> I guess that life is not better for the secretaries, either, since fewer of them have jobs.
Only the title has disappeared. They're now called administrative assistants or any of a dozen other vague titles that are designed to look good on resumes. At the end of the day, they're still secretaries. Sadly, in this day and age, it's not uncommon for a man to apply for the job. Admittedly, this saps all of the joy out of addressing ones secretary with lines such as, "Sweetheart, will you please pour us a drink?"
> The whole concept of "friending" someone was utterly alien to them.
Well, duh - they would have thought it ungrammatical!
When some stranger can be a friend by just a few clicks, the word has lost its meaning.
> Real men drink Scotch and don't use Facebook.
>That is REALLY comforting as I do *not* have a Facebook account (because I think it is narcissistic and gay) and I *do* drink scotch (because it goes great with a cigar). In fact, I have a bottle of 18-year-old stuff in the pantry right now. I drive a Jaguar, too.
well,in that case,im gay,and a poor gay...
but i think some of the gays here wont agree with your words,Nelson...at least the 'true' gays
I find IDEA's branching useless.
Other than that I go along the lines looking at average depth reached in my IA column.
Here are 5 more ideas for example:
5) Evaluation imbalance
It is the discrepancy between the amount of material and engine evaluation. If one player is a pawn down, but engine evaluates him better by 0.53, then the imbalance is 1 + 0.53 = 1.53. But if the eval were exactly -1.00, then the imbalance would be zero. In positions with even material the imbalance is always equal to evaluation.
The number of pieces along the x-axis. In the start position all pieces are symmetrically placed and the value is 16/32.
In this position only 3/6 pieces are symmetrically placed: Kh2/7, Pg2/7 and Pf3/6.
7) The number of more-or less equal moves
It's defined as the number of moves that are not evaluated worse than a specified amount of centipawns from the best move, for example 0.25.
Here are 10 best moves at d21 by Houdini 3:
21/51 01:49 822.174.902 7.481.000 +0.41 Kh2h1 Rd5d2 Qf2g3 Qd8b8 Qg3xb8 Ra8xb8 Rc4a4 Rd2c2 Bc5d4 Bg6d3 Re1e7 Kh7g8 Bd4a7 Rb8d8 Ra4g4 g7g5 h4xg5 h6xg5 Rg4d4 Rd8xd4 Ba7xd4 Bd3f1 g2g4 Rc2xa2 Bd4xf6 Bf1g2+ Kh1h2 Bg2xf3+ Kh2g3 Bf3d5 Bf6xg5 Bd5xb3 Bg5e3 a6a5 Be3d4
21/51 01:49 822.174.902 7.481.000 +0.45 Re1e7 Ra8c8 b3b4 Qd8xe7 Bc5xe7 Rc8xc4 Qf2a7 Rc4xh4+ Kh2g3 Rh4d4 Be7xf6 Rd5d7 Qa7xd4 Rd7xd4 Bf6xd4 Bg6c2 Kg3f4 g7g5+ Kf4e5 Kh7g6 f3f4 g5xf4 Ke5xf4 h6h5 Bd4e3 Bc2d3 g2g3 Kg6f6 a2a4 Bd3c4 Be3d4+ Kf6e6 Kf4g5 Ke6d5
21/51 01:49 822.174.902 7.481.000 +0.49 Rc4c3 Rd5d2 Re1e2 Rd2xe2 Qf2xe2 Qd8d5 Rc3c1 Ra8e8 Qe2f2 Qd5d3 Rc1e1 Re8xe1 Qf2xe1 Qd3c2 Qe1a5 Bg6h5 Kh2h1 Qc2f5 b3b4 Qf5b1+ Kh1h2 Qb1c2 Qa5xa6 Bh5xf3 Qa6f1 Bf3c6 Qf1f2
21/51 01:49 822.174.902 7.481.000 +0.51 Rc4c1 Rd5d2 Re1e2 Qd8c7+ Kh2g1 Rd2xe2 Qf2xe2 Ra8d8 Qe2xa6 Qc7f4 Rc1e1 Qf4xh4 Bc5f2 Qh4b4 Qa6a7 Rd8d2 a2a4 Qb4xb3 Re1e7 Qb3g8 a4a5 Rd2a2 Bf2d4 Qg8f8 Re7d7 Bg6f5 Bd4c5
21/51 01:49 822.174.902 7.481.000 +0.54 Re1e6 Ra8c8 Bc5b6 Qd8d7 Re6e2 Rc8c6 Rc4xc6 Qd7xc6 Bb6e3 Qc6d6+ Qf2g3 Qd6e6 Qg3g4 Qe6xg4 f3xg4 Bg6d3 Re2e1 Rd5e5 Kh2g3 Re5a5 Re1a1 Ra5e5 Be3f2 Re5e2 Ra1d1
21/51 01:49 822.174.902 7.481.000 +0.66 Bc5b4 Rd5d7 Qf2g3 Ra8c8 Bb4c5 Qd8g8 Qg3f4 Rc8e8 Re1xe8 Qg8xe8 Bc5d4 Qe8e6 Rc4c1 Bg6d3 Rc1c7 Qe6f5 Qf4g3 Bd3b1 a2a3 Bb1a2 b3b4
21/51 01:49 822.174.902 7.481.000 +0.67 Re1c1 Ra8c8 Bc5e3 Rc8xc4 Rc1xc4 Rd5d7 Be3f4 Qd8a5 Rc4c8 Rd7e7 Rc8c5 Qa5b6 Bf4g3 Qb6e6 h4h5 Bg6d3 Bg3f4 Re7d7 Qf2d2 Bd3b5 Qd2e3 Qe6xe3 Bf4xe3 Rd7e7 Be3c1 Re7e2 a2a4 Bb5e8 Kh2g3 Be8f7
21/51 01:49 822.174.902 7.481.000 +0.67 Bc5e3 Ra8c8 Re1c1 Rc8xc4 Rc1xc4 Rd5d7 Be3f4 Qd8a5 Rc4c8 Rd7e7 Rc8c5 Qa5b6 Bf4g3 Qb6e6 h4h5 Bg6d3 Bg3f4 Re7d7 Qf2d2 Bd3b5 Qd2e3 Qe6xe3 Bf4xe3 Rd7e7 Be3c1 Re7e2 a2a4 Bb5e8 Kh2g3 Be8f7
21/51 01:49 822.174.902 7.481.000 +0.80 Bc5e7 Qd8g8 Re1c1 Rd5d7 Rc4c7 Qg8e6 Be7b4 Rd7xc7 Rc1xc7 Ra8c8 Rc7e7 Qe6f5 Bb4d6 Qf5d5 Bd6c7 Rc8e8 Re7xe8 Bg6xe8 Bc7g3 Qd5d3 Bg3f4 g7g5 Qf2c5
21/51 01:49 822.174.902 7.481.000 +0.83 Qf2g3 Rd5d7 Qg3f4 Ra8b8 Rc4c1 Rb8b7 Re1e2 Rd7d5 b3b4 a6a5 a2a3 Bg6f7 Rc1e1 a5xb4 a3xb4 Kh7h8 Kh2g1 Rb7d7 Re2e7 Rd7xe7 Re1xe7 Rd5d7 Re7xd7 Qd8xd7 Qf4e4 Qd7e8 Qe4g4
As can be seen, there are only 5 moves within 0.25 interval.
8) Evaluation stability
The difference between the highest and the worst evaluation of best move across all plies.
Again, Houdini 3's evaluation in the same position:
4/14 00:00 5.092 2.546.000 +0.92 Bc5e3 Bg6d3 Rc4c6 Ra8c8
5/14 00:00 6.282 2.094.000 +0.76 Bc5e3 Bg6d3 Rc4c6 Qd8d7 Re1c1 Rd5e5
6/14 00:00 9.539 1.059.000 +0.74 Bc5e3 Bg6d3 Rc4c1 Ra8c8 Qf2g3 Rd5e5
7/14 00:00 12.069 1.097.000 +0.67 Bc5e3 Bg6d3 Rc4c1 Ra8c8 Qf2g3 Rd5e5 Rc1xc8 Qd8xc8
8/16+ 00:00 17.020 1.215.000 +0.77 Bc5e3
8/16 00:00 19.203 1.200.000 +0.81 Bc5e3 Bg6d3 Rc4c1 Ra8c8 Qf2g3 Rd5e5 Rc1xc8 Qd8xc8 Re1c1
8/18+ 00:00 31.840 1.447.000 +0.97 Qf2g3
8/18 00:00 53.766 1.991.000 +0.74 Qf2g3 Qd8a5 Rc4g4 Rd5xc5 Rg4xg6 Rc5c7 Re1e6 Ra8d8 b3b4
8/18 00:00 53.767 1.991.000 +0.81 Bc5e3 Bg6d3 Rc4c1 Ra8c8 Qf2g3 Rd5e5 Rc1xc8 Qd8xc8 Re1c1
9/18 00:00 64.038 2.208.000 +0.73 Bc5e3 Bg6d3 Rc4c1 Ra8c8 Qf2g3 Rd5e5 Rc1xc8 Qd8xc8 Re1c1 Qc8d7
9/18+ 00:00 80.074 2.502.000 +0.91 Qf2g3
9/18 00:00 92.125 708.000 +0.89 Qf2g3 Rd5d7 Bc5e3 Ra8c8 Rc4xc8 Qd8xc8 Qg3g4 Qc8c7+ Be3f4 h6h5 Qg4g3
10/19 00:00 118.970 862.000 +0.85 Qf2g3 Rd5d7 Bc5e3 Ra8c8 Rc4xc8 Qd8xc8 Re1c1 Qc8e8 Qg3f4 Bg6d3 Rc1c6 Kh7g8
11/22 00:00 207.634 1.322.000 +0.79 Qf2g3 Rd5d7 Bc5e3 Ra8c8 Rc4xc8 Qd8xc8 Re1c1 Qc8e8 Qg3f4 Bg6d3 Rc1d1 a6a5 a2a3 g7g5
12/22 00:00 427.708 2.204.000 +0.81 Qf2g3 Rd5d7 Rc4c1 Ra8c8 Qg3f4 Qd8g8 Bc5f2 Rc8xc1 Re1xc1 Qg8e6 Bf2e3 Rd7e7 Be3d2 Kh7g8 Qf4d4
13/25 00:00 517.031 2.462.000 +0.81 Qf2g3 Rd5d7 Rc4c1 Ra8c8 Qg3f4 Qd8g8 Bc5f2 Rc8xc1 Re1xc1 Qg8e6 Bf2e3 Rd7e7 Be3d2 Kh7g8 Qf4d4
14/26 00:00 1.248.713 4.028.000 +0.79 Qf2g3 Rd5d7 Rc4c1 Ra8c8 Qg3f4 Qd8g8 Bc5g1 Rc8xc1 Re1xc1 Qg8e8 Bg1e3 Bg6d3 Rc1d1 a6a5 a2a3 g7g5
15/30 00:00 2.830.518 5.539.000 +0.79 Qf2g3 Rd5d7 Rc4c1 Qd8g8 Re1e7 Rd7xe7 Bc5xe7 Ra8c8 Rc1c7 Rc8xc7 Qg3xc7 Qg8e6 Be7d6 Qe6e2 a2a4 Qe2c2 Qc7b7 a6a5
16/30 00:01 6.646.255 7.047.000 +0.77 Qf2g3 Rd5d7 Rc4c1 Qd8g8 Qg3g4 Qg8d5 Rc1c4 h6h5 Qg4d4 Qd5xd4 Rc4xd4 Rd7xd4 Bc5xd4 Ra8d8 Bd4e3 Rd8e8 Kh2g3 Bg6d3 Kg3f2
17/32 00:01 9.688.440 7.464.000 +0.77 Qf2g3 Rd5d7 Rc4c1 Qd8g8 Qg3g4 Qg8d5 Rc1c4 h6h5 Qg4d4 Qd5xd4 Rc4xd4 Rd7xd4 Bc5xd4 Ra8d8 Bd4e3 Rd8e8 Kh2g3 Bg6d3 Kg3f2
18/35+ 00:02 21.539.066 8.258.000 +0.87 Qf2g3
18/35 00:02 22.324.621 8.271.000 +0.86 Qf2g3 Rd5d7 Qg3f4 Ra8c8 Rc4c3 Rc8c6 b3b4 Qd8c8 a2a3 a6a5 Kh2g1 a5xb4 a3xb4 Rc6d6 b4b5 Rd6d5 Re1c1 Qc8b7 b5b6 Kh7g8 Bc5f2
19/37 00:03 30.426.943 8.468.000 +0.84 Qf2g3 Rd5d7 Qg3f4 Ra8c8 Rc4c3 Rc8c6 b3b4 Qd8c8 a2a3 a6a5 Kh2g1 a5xb4 a3xb4 Rc6e6 Re1xe6 Rd7d1+ Kg1h2 Qc8xe6 b4b5 Qe6e1 Rc3a3 Qe1h1+ Kh2g3 Rd1d5 Qf4e3
20/42- 00:06 58.815.260 8.819.000 +0.74 Qf2g3 Rd5d7
20/51 00:17 152.368.790 8.995.000 +0.76 Qf2g3 Rd5d7 Qg3f4 Ra8c8 Rc4c3 Qd8g8 h4h5 Bg6d3 Rc3xd3 Rd7xd3 Qf4e4+ Kh7h8 Qe4xd3 Rc8xc5 Qd3g6 Qg8b8+ Kh2h3 Rc5e5 Re1xe5 Qb8xe5 Qg6g4 Qe5d5 Kh3h2 Qd5c5 f3f4 a6a5 Qg4g6 Qc5d5 Kh2g3 Kh8g8 Qg6e8+ Kg8h7
21/51 00:19 175.337.333 8.989.000 +0.79 Qf2g3 Rd5d7 Qg3f4 Qd8a5 Re1e7 Ra8a7 Re7e2 Ra7b7 Bc5f2 Bg6d3 Rc4c5 Rb7b5 Rc5xb5 a6xb5 Re2d2 Qa5d8 Rd2d1 Bd3g6 Rd1xd7 Qd8xd7 Bf2g1 Qd7d5 Qf4b4 Qd5e5+ f3f4
21/51 00:28 252.546.048 8.813.000 +0.79 Qf2g3 Rd5d7 Qg3f4 Qd8a5 Re1e7 Ra8a7 Re7e2 Ra7b7 Rc4c1 Rd7d5 Bc5f2 Qa5d8 Rc1c5 Rd5xc5 Bf2xc5 Rb7d7 b3b4 Bg6d3 Re2d2 a6a5 a2a3 a5xb4 a3xb4 Qd8e8 Qf4e3
The highest eval is 0.97 at d8 and the lowest one 0.67 at d7, the difference is 0.30 centipawns.
9) The difference in evaluations between the most logical or natural move and the best move
This one is a bit tricky to determine, as one would need to find a criterion for describing naturality or logicality of moves. Moving knight to c3 from b1 or exchanging pawns are very natural moves. Sacrificing exchange for positional advantages or making somewhat thematic bishop sacs on h2/h7 represent moves that are less obvious. And moves such as Fischer's 11.Na4!! or Keres' 72.Qe5!! against Fischer in Curacao 1962 are quite illogical and extremely to spot.
Also one must not forget that since engines and humans have very unlike move-choosing processes, certain positions may be quite easy for computers while being diffcult for humans, and vice versa.
>And moves such as Fischer's 11.Na4!! or Keres' 72.Qe5!! against Fischer in Curacao 1962 are quite illogical and extremely to spot.
deka can you post the whole game,please?
> can you post the whole game,please?
I find that view a bit problematic in that it assumes that the engine is evaluating the position correctly, which isn't always the case. An engine may evaluate its first and second choices as being near or far in value, but if it is misevaluating the position in the first place - underestimating the attack which is coming, for example - the degree of difference in eval is probably not all that important in deciding whether your opponent will be able to find a good response. Similarly if the engine is calculating quickly, it may just be that it is blind to the dangers in its position, pruning off key lines without examining them at all.
Re. Uly's definition of sharpness - He seems to be using the word "sharp" to describe Robin Smith's "box canyons" i.e. forced variations which can sometimes change the evaluation of a position in a drastic way. The definition of "sharp" found in chess dictionaries has to do with first of all, risk, and often tactical complications, which can be forced or not.
One thing that correspondence players are looking for is positions that are complex in the sense that engines will not find the best moves even if they analyze for days on end. The moves in the winning line benefit from being surprising in some way, perhaps getting pruned by the engines because it offends against their evaluation too much at the root. Sacrificing material is one possibility, or perhaps sacrificing space in exchange for a later attack.
The GUI Fritz 13 has "hotness" and "mate-o-meter" measures which may help pick out positions that are complex from the point of view of an over-the-board player analyzing their games. I wonder if there are other programs that have a feature similar to this. I believe Crafty gave direct access to its evaluation broken down by king safety and such using the 'score' command.
Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill