Your network blocks the Lichess assets!

lichess.org
Donate

Identifying the Style of Different Players

nice post! interesting idea! i love it! thanks for the good blog :D

nice post! interesting idea! i love it! thanks for the good blog :D

Can one do the same for individual games? Like something I could add to LiChess Tools to show the playing style of one side on this chart?

Can one do the same for individual games? Like something I could add to LiChess Tools to show the playing style of one side on this chart?

I put this on your Substack, but I'll add it here in case any Lichess folks have thoughts: Very cool! I'd love to know what the full set of variables were and what the PCA scree plot looked like. Looking at this with a large sample of amateur players would also be neat, especially if you could do some clustering.

I put this on your Substack, but I'll add it here in case any Lichess folks have thoughts: Very cool! I'd love to know what the full set of variables were and what the PCA scree plot looked like. Looking at this with a large sample of amateur players would also be neat, especially if you could do some clustering.

@TotalNoob69 said in #3:

Can one do the same for individual games? Like something I could add to LiChess Tools to show the playing style of one side on this chart?

In theory, this could work for individual games, but some variables may be complete outliers. For example, I used the average game length, which is around 30-40 moves for all players I looked at, so I think that a 20 or 80 move game may skew the data quite a bit

@TotalNoob69 said in #3: > Can one do the same for individual games? Like something I could add to LiChess Tools to show the playing style of one side on this chart? In theory, this could work for individual games, but some variables may be complete outliers. For example, I used the average game length, which is around 30-40 moves for all players I looked at, so I think that a 20 or 80 move game may skew the data quite a bit

@NDpatzer said in #4:

I put this on your Substack, but I'll add it here in case any Lichess folks have thoughts: Very cool! I'd love to know what the full set of variables were and what the PCA scree plot looked like. Looking at this with a large sample of amateur players would also be neat, especially if you could do some clustering.

Here are all the variables:
'movesPerGame', 'kingsideCastling', 'queensideCastling', 'noCastling', 'oppositeSidesCastling', 'promotions'
'captures', 'recaptures', 'trades', 'centralMoves', 'flankMoves', 'movesInOppHalf', 'checks', 'inCheck',
'pawnMoves', 'aPawnMoves', 'bPawnMoves', 'cPawnMoves', 'dPawnMoves', 'ePawnMoves', 'fPawnMoves', 'gPawnMoves', 'hPawnMoves',
'pawnToRank5', 'pawnToRank6', 'pawnToRank7', 'pawnCaptures',
'pieceMoves', 'knightMoves', 'bishopMoves', 'rookMoves', 'queenMoves', 'kingMoves', 'pieceCaptures', 'forwardPieceMoves',
'movesToEnemyKing', 'movesToEnemyKingMan', 'opponentKingMoves', 'materialUp', 'materialDown', 'avgMaterial', 'materialImbalance'

And here is the explained variance ratio for the first 5 components: [0.33357048 0.25080529 0.10255517 0.07911614 0.07765889]

@NDpatzer said in #4: > I put this on your Substack, but I'll add it here in case any Lichess folks have thoughts: Very cool! I'd love to know what the full set of variables were and what the PCA scree plot looked like. Looking at this with a large sample of amateur players would also be neat, especially if you could do some clustering. Here are all the variables: 'movesPerGame', 'kingsideCastling', 'queensideCastling', 'noCastling', 'oppositeSidesCastling', 'promotions' 'captures', 'recaptures', 'trades', 'centralMoves', 'flankMoves', 'movesInOppHalf', 'checks', 'inCheck', 'pawnMoves', 'aPawnMoves', 'bPawnMoves', 'cPawnMoves', 'dPawnMoves', 'ePawnMoves', 'fPawnMoves', 'gPawnMoves', 'hPawnMoves', 'pawnToRank5', 'pawnToRank6', 'pawnToRank7', 'pawnCaptures', 'pieceMoves', 'knightMoves', 'bishopMoves', 'rookMoves', 'queenMoves', 'kingMoves', 'pieceCaptures', 'forwardPieceMoves', 'movesToEnemyKing', 'movesToEnemyKingMan', 'opponentKingMoves', 'materialUp', 'materialDown', 'avgMaterial', 'materialImbalance' And here is the explained variance ratio for the first 5 components: [0.33357048 0.25080529 0.10255517 0.07911614 0.07765889]

Interesting stuff although I'm a little confused. If players like Petrosian and Smyslov are higher on the Y axis and more left on the X asis than Topalov and Polgar.. Why are the latter labeled as attacking while the former are labeled as positional? Is this suggesting that Petrosian was able to blow people off the board but in a positional way? Or am I misunderstanding?

Interesting stuff although I'm a little confused. If players like Petrosian and Smyslov are higher on the Y axis and more left on the X asis than Topalov and Polgar.. Why are the latter labeled as attacking while the former are labeled as positional? Is this suggesting that Petrosian was able to blow people off the board but in a positional way? Or am I misunderstanding?

Can you takl about the meaning of universal. Is it about a mixture of the other 2. or a different class.

it might go with how one defines style is it a whole game thing, a phase thiing, a bias in which type of tactical patterns or feature sets preferences.

are players necessarily using one style through a whole game. or could have adaptible behavior.

I am not aksing you all those questions. Just the first one. or anything you would like to share your thoughts about this.

Is there a way to find building blocks of styles that might have board patterns. wait. is that what you are exploring this way.. doh!

In that case it would be nice to have the list or link to the list of features you are considereing, and perhaps their algebraic defitions as functions of the positions or games. Since the ID is player, you might be having all sorts of types of variables, some integrating over whole games and many games.. I would learn a lot of chess I think seeing those features. Do you think you could have such work put on github?

Can you takl about the meaning of universal. Is it about a mixture of the other 2. or a different class. it might go with how one defines style is it a whole game thing, a phase thiing, a bias in which type of tactical patterns or feature sets preferences. are players necessarily using one style through a whole game. or could have adaptible behavior. I am not aksing you all those questions. Just the first one. or anything you would like to share your thoughts about this. Is there a way to find building blocks of styles that might have board patterns. wait. is that what you are exploring this way.. doh! In that case it would be nice to have the list or link to the list of features you are considereing, and perhaps their algebraic defitions as functions of the positions or games. Since the ID is player, you might be having all sorts of types of variables, some integrating over whole games and many games.. I would learn a lot of chess I think seeing those features. Do you think you could have such work put on github?

I would love to see this analysis applied to modern popular chess players: Magnus, Hikaru, Hans. Or even the YouTuber crowd: Eric Rosen, Gotham, Will Graif, that kind of stuff.

I would love to see this analysis applied to modern popular chess players: Magnus, Hikaru, Hans. Or even the YouTuber crowd: Eric Rosen, Gotham, Will Graif, that kind of stuff.

@TotalNoob69 said in #9:

I would love to see this analysis applied to modern popular chess players: Magnus, Hikaru, Hans. Or even the YouTuber crowd: Eric Rosen, Gotham, Will Graif, that kind of stuff.

That could be a project that would force one to figure out what was done for the plot we saw and get our hand on the full data set.

It might also make that a test set for a machine leanring prediction model. Although would there be arguments about their styles not being captured for chess culture evolution reasons ( I would not know, from my ignorant point of view that universe is a priori infinite or equivalently of uncertain scope).

I find this might be a nice machine learning exploration project. As you (op) are not really seeking any closed form deterministic equation but more exploration of which of your many raw metrics would contribute to the 3 values styles classification (definition of style or hypothesis).

@TotalNoob69 said in #9: > I would love to see this analysis applied to modern popular chess players: Magnus, Hikaru, Hans. Or even the YouTuber crowd: Eric Rosen, Gotham, Will Graif, that kind of stuff. That could be a project that would force one to figure out what was done for the plot we saw and get our hand on the full data set. It might also make that a test set for a machine leanring prediction model. Although would there be arguments about their styles not being captured for chess culture evolution reasons ( I would not know, from my ignorant point of view that universe is a priori infinite or equivalently of uncertain scope). I find this might be a nice machine learning exploration project. As you (op) are not really seeking any closed form deterministic equation but more exploration of which of your many raw metrics would contribute to the 3 values styles classification (definition of style or hypothesis).