Follow-up thought, I wonder if this approach could be applied to multiple games from the same player in a way that draws out certain habits. Like, if a player is very fond of a certain opening that would be clear by looking at a combined version of this visualization across all their games for the first 6-10 moves, and hiding anything that didn't have high correlation across games.
To address the scannability, maybe the outlines could be different based on which piece was there in the final position? Circles or small circles for locations that have a lot of traffic? So much information about a game is indicated by the pieces' final position so I feel like this could reduce the opacity some (but again, I'm not an expert chess player so maybe I'm just inadequate at inferring much information from these).
You may be interested in the structural relationship explorer by JKU Labs [0], which condenses hundreds of chess matches in a lower-dimensional space, revealing multiple win/loss states and trajectories and the (dis)similar strategies of expert matches. It's one of my favourite visualisations.
Comments
Follow-up thought, I wonder if this approach could be applied to multiple games from the same player in a way that draws out certain habits. Like, if a player is very fond of a certain opening that would be clear by looking at a combined version of this visualization across all their games for the first 6-10 moves, and hiding anything that didn't have high correlation across games.
To address the scannability, maybe the outlines could be different based on which piece was there in the final position? Circles or small circles for locations that have a lot of traffic? So much information about a game is indicated by the pieces' final position so I feel like this could reduce the opacity some (but again, I'm not an expert chess player so maybe I'm just inadequate at inferring much information from these).
You may be interested in the structural relationship explorer by JKU Labs [0], which condenses hundreds of chess matches in a lower-dimensional space, revealing multiple win/loss states and trajectories and the (dis)similar strategies of expert matches. It's one of my favourite visualisations.
[0]: https://jku-vds-lab.at/publications/2022_embedding_structure...
Wow, that's super interesting! Thank you I will look into it!