making basic competition unit to be a pair of games with white player the choice of backrank, would by pool sampling be another way to randomize the experience.
the danger might then be that rabid winners would each on their turn also choose unbalanced, I guess. I did not think like that...
so yes.. then a middle ground would be to have pool based randomization among consensus balanced positions or with a range that is similar to the standard white initial turn imbalance of first turn per some moving target like engine scoring. Perhaps some safety margin there.
my main concern is a within life span ability to learn (but not depth first turn by turn knowledge as recipes to playable deep, one can learn other ways as someone else said, "immediately" from the very first position playing chess through either learned intuition or reasoning (or both, usually or dare I say, always both).
For me 960 is for the already experienced to find the bare commonalities from standard, as there are limits to the statistical brain sampling, if machine don't sweat, humans do, and they don't use systemic back-prop for their neural networks, timing of repeated (and diverse to a sufficient extent) needs to be within biological learning time frames. I would refer to basic animal statistical learning of the Bell ringing appearing association to food, the logic of the experimental design, being imposed on the learning brain there, as food you will get if hearing the bell (it might be a when but it acts like an if in this case, and its reproducibility increases the logical behavior consequence to the event of the bell ringing, in my interpretation).
a fluke, that repeats itself in the environment in association with another event, loses "fluke" status.. Correlation might not be causation, in general, but in the real world, it acquires associative importance, and if the real world is chess board logic at turn by turn clock, then it might converge to the board mechanical logic.. correlation would not be a fluke there, eventually. in the learning trajectory of my ideal conception right here.
So, now that I motivated why 960 full random is like an impossible learning program at any brain type level, deduction or induction (as in intuition under the hood).
Now, that the random obsession can be delegated to the pool sampling uncertainty of either player choice in pair of games as the unit.
we just need to adjust (while including standard, why exclude it, sure some of the pool might be shy and always choose it at first, but then there is the other game when they don't choose the set up.). This would seem like a smooth learning challenge transiton from standard. and would redfuce the spoiler lower ELO current limiting but winning learning strategy that might have annoyed high level players (that were past that kind of play) like Fisher.
maybge the pools were small then (the pools envisioned behind each game being some random generator driven).
trust in the pool....
so all the arguments in this balanced discussion might be to filter among the 960..
I would prefer also a representant of each mindboggling outliers (although some might think it would be about imbalance, I think it might also be about how distant from standard backrank consequence emerging board logic (i.e.. that which the human need to develop beyond the mere turn by turn logic, which I would say to be of both cosncious and subconscious learning processes).
"I" being the simplest voice choice here. I do not pretend to have it right.. but I think I hear little of this kind of consideration and revisiting of "expected" immutable. I got to go... but those arguments I wish to have made finally clearer that over the past 5 years, that I might have tried to express them...
PS: sorry for sentences left hanging... I don't even have time for another rambling temptation trying to fix that.
position 518 can be removed, but I don't find reason to remove position 80. Although delicate I think that position is not lost for black. I give you one line included in article
BBNNRKRQ
1.g4, Nd6, 2.b3,Ne6, 3.Nd3,f6 4.c4, b6!
I don't have access to a super computer anymore but I got some lower depth (20 on stockfish) analysis done on my personal machine. I am running some more right now but feel free to message me if you want python scripts or the txt files. I don't know if this is just a result of low time and low depth but some positions even have black as having a small advantage (around 7-17 centipawn).
Each centipawn value is part of a larger evaluation, that involves the entire position. The opening phase cp is not as relevant as an endgame cp value. As phase as I know we don't give a cp value for phases of the game. If engines incorporated phase-specific weights like: 0.25 cp for the opening, reflecting tactical positions and piece placement; 0.50 cp for the middlegame, due to the increased volatility; and 0.75 cp for the endgame, where the positional advantage has a more tangible impact. Maybe then, engines might take more risks, or even resign earlier.
A chess position normally needs to be better than a ±1.0 to indicate a winning advantage. It's more of a gambit than anything else. A 2.0 difference might not sound like much in an opening, but by the time you reach the endgame, it's often enough to tip the scales toward victory. Think of how 2 pawns can mutually protect each other against an overload king, that doesn't have the tempo to reach both in time. The speed of a pawn runs a bit faster than a king. The king is the slowest piece on the chessboard, and sometimes can travel more than a pawn.
I think, cp values is an old system and now we have score cp which is not presented the same way. I think evaluations need to be orientated more towards mobility (tempo) and activity (threats) than anything else. Being proactive should have a value higher than the pieces that are forced to move. In the opening I prioritise safe piece activation. Rush a gambit and the risk of losing cp increases. Hanging pieces is neglecting the rules of thumb. There is wisdom in chess and it's hard to give it a score.
When it comes to chess960 there are positions that take longer to activate the pieces. The chess960 could be sorted by activation time. In standard chess we only needs two pawns moved to be able to legally mobilise all the pieces.
If we focus on castling, it can take from 1, 2 or 3 moves to castle. Each static position needs a situation score.
Some Chess960 starting positions inherently offer faster or slower development. Quantifying this "activation time" or "situation score" could be a valuable metric for analyzing Chess960 openings and understanding the strategic nuances of different setups.
@Toscani I'm only a club level player so perhaps I don't understand the nuances of the starting position, nor do I have experience in developing any engines or game theory. Your activity and tempo evaluation is really what I value as well and that mentality is why I keep playing the king's gambit.
I value counterattacking chances as black and always create asymmetry when I can, but I am genuinely confused how black can have a perceived advantage at the starting position. It's a turn based game where white starts first. Doesn't that give you tempo to generally be the first to develop and activate your pieces?
Well, evaluations are not the golden bullet, but those 3500+ rated engines surely show some authority...
At the end if the day, only mate scores are reliable. But an engine giving +2 is usually decisive, although the scale seem to be shifting, and the variations on that the eval is based get more and more obscure the stronger the engines get.
Given that there are positions that the engines hate but people get good results with (like KID, or maybe some lines of the Pirc), just being 0.8 worse is a rather random observation, that might also change significantly between engines and search depth.
@collective_enjoyer said in #25:
> It's a turn based game where white starts first. Doesn't that give you tempo to generally be the first to develop and activate your pieces?
Usually we consider White to have an advantage because of that. That is, unless it is zugzwang. It is easy to imagine that among the 960 positions, there might be some where Black always has a good (better) response to whatever White tries.
When we gather a couple millions games played, we might get to see some trends. But at the moment, databases for 960 are rather hard to get hold of.
@nadjarostowa Thanks for the perspective! The analysis at higher depth that I ran over night gave far fewer black positions with a small centipawn advantage for black (*Positions 30, 110, 433, 774 if we use the SP counting from 0-959*). I'm going to take a closer look at them if we can find a zugzwang or if most opening moves can have a meaningful counterattack.
The position with the highest evaluation according to the Lichess cloud evaluation is position 935 with an eval of +0.75. The position with the lowest eval is position 404 with an eval of -0.04. However, the position in which White wins the most games in the Lichess database is position 608 with a ratio of 12162/10793 white wins to black (~1.1268). The position in which Black wins the most games is 762 with a ratio of 11462/11421 white wins to black (~1.004). The most drawish position is 131 with 1133/23877 games being a draw (~4.75%) and the least drawish position is 863 with 5/159 games being a draw (~3.14%).
Finally, the ranking of position 518 for each of these properties (with 1 being the highest rank) is
White to Black win ratio: 160/960
Drawishness: 344/960
Evaluation: 705/960
So it seems that 518 is one of the least equal positions according to the database, but one of the most equal according to the evaluation (although this could also be due to the fact that position 518 has been searched to depth 65, with the average depth of the other positions being 33.5875).
Did anyone publich distributions of lichess games across the 960 setups. A priori it ought to be random (by the grace of chess gods willing it) uniform, but could abort put dents on that? also, non-rated games or from positions games where one is sure that it was all back ranks material occupancy, so likely a 960 as well, even if not using the tunnel of proper certified pure random setup across 960 etc... stuff of legends.. yipeeyeah! its random!).
Minor point, Why do I even care... sorry... I liked the last 2 posts. I wish we could have more rational discussions with more information about what is in the setups and how that affects the thinking or learning problem.. Not just envelope statistics.
@dboing said in #29:
> Did anyone publich distributions of lichess games across the 960 setups.
Unsurprisingly, position 518 has more than a quarter million times more games than the next most popular position, which is 343 with 24,440 games. The least popular position is 186 with 23,045 games, which isn't a huge difference (23045/24440 is about 94%), so it seems more or less an even spread.
This topic has been archived and can no longer be replied to.