Which openings are related?

Caro-Kann and French are similar. I hate them both and play neither.

@Molurus said in #11:
> Caro-Kann and French are similar. I hate them both and play neither.

Then, if I understood, you would not have contributed to them being linked enough to make it to the figures.
I keep trying to explain things but it takes room.. I think the bi-partite graph visualization would be a good important step to show. it means using 3D perspective.. or invoking an ambient space where we could look at the opening id "plane" facing the player Id plane, and bi-partite graph being about links in that ambient space (3D) linking nodes of one plane to another.

There would be no ambiguity there with respect to either the paper or the blog text. but both decided not to show such figure. The context of the paper may have made that superfluous. But not here.

I am moving this to some place in my musing.. It is adressed to all interested in this topic here or skeptical. I think it is a good proposal. But it needs pedagogical development. to go beyond people who can talk shop, i.e. lichess chess aware audience but not necessarily people who could reproduce what the op or the paper did, but be able to follow and criticize the reasoning and accepot of not the suggestion/hypothesis/conclusion of paper and blog (which was careful in putting a header of "limitatoins").

I describe what I think is mssing for the chess audience interested in the chess questoin from this to be equal reasoning participants, and therefore able to accept or not this new way of thinking about chess. beyond the individual player, and the individual game.

This might help fill in the gap between the text describing the simple player ID to opening ID bi-partite graph and the displayed transformed graphs showing only opening ID nodes (and other visual transformation about their presentation). Think of X as players maybe and Y as opening ID.

the Y-projection being the opening-projection of players "range of opening manifested preference" (quotes might be répertoire, still trying to get some definition about a player repertoire (anyone can help there, toward the paper or the blog understainding). The particular contribution of the X node and their linkage would likely be different but some functions of counting them. I still have not looked in detail enough. Op, might chip in here, having computed such thing already. Or I might find some time to read the math of the paper, and translate back here.

en.wikipedia.org/wiki/Bipartite_network_projection
"Possible projections of a simple bipartite network"

This graph shows the possible one-mode projections of a simple bipartite network.
By Tao Zhou, Jie Ren, Matúš Medo, and Yi-Cheng Zhang - Physical Review E 76(4): 046115, 2007, Public Domain, commons.wikimedia.org/w/index.php?curid=19456549
upload.wikimedia.org/wikipedia/commons/f/f9/Bipartite_network_projection.png
commons.wikimedia.org/wiki/File:Bipartite_network_projection.png#/media/File:Bipartite_network_projection.png

Ok, so "relatedness" does not mean "tendency to transposition" but rather: 2 openings are "related", if a lot of players play both of them. So somebody who plays the Ruy Lopez often also plays the Scotch (which makes sense). It does not mean that a Ruy Lopez position often transposes to a Scotch position. Did I understand this correctly?

@Hurluberlu2 said in #14:
> Ok, so "relatedness" does not mean "tendency to transposition" but rather: 2 openings are "related", if a lot of players play both of them.

I think you're right. What I would love to see is more about transpositions and similarity of structures maybe. For example I play London in reverse colors against certain d4 schemes, but I don't play it as white. Or I don't play Caro-Kann, although I play a similarly-structured Slav defenceman. And maybe article like that could uncover more things like that in my repertoire with some ideas.

I'm quite sure, though, that in near future there will be simply AI that goes through all your games and points out exactly what your weaknesses are, suggests repertoire, and maybe even constructs excersices that are ideal for your training.

Interesting article. I suppose the real question is if a player plays the French (say) then which openings do they tend to play against d4/c4/Nf3 and what do they play as white?

Wouldn't exchange caro kann for example, be closer to d4 openings since the carlsbad pawn structure? and also pirc, philidor to KID?

Team french feels the caro as too slow, team Caro feels the french as a Caro minus one bishop.

the study was poorly made, they didnt choosed the openings well so there was false overlaps (e6 with dutch, nimzo and french for example) and they made the game pool very small as well.

I think there is a systematic fault in the study, in that it apparently assumes that two openings played by the same player are somehow "similar" by default. If picking a single defense each against 1.e4 and 1.d4, that's a reasonable assumption since it would in principle mean more familiarity with the structures and ideas. However, if you play two different white openings or have two (or more) defenses against 1.e4 it makes no sense for anyone to pick ones leading to similar play - that's literally just more study without much return.

What makes sense is instead to play two dissimilar openings, for example one sharp and tactical while the other is slow and positional. This would give the strong option to pick an opening depending on what is known or assumed about the opponent.