T
Guest
This image was generated from around 20,000 16xx+ arabia RM 1v1 games. Basically you get the image by taking the frequencies that civilizations open either premill drush, postmill drush, scouts, maa, or straight archers and you map civs which have similar percentages across these openings close to each other using some kinda sophisticated tools. The colors of the civs are then determined by software which finds clusters in the original data (the opening play rates).
For instance for these five openings and from the selection of games (from this last month roughly) Bohemians open with percentages below
premill drush: ~0.109 of the time
post mill drush: ~0.063 of the time
men at arms: ~ 0.499 of the time
scouts: ~0.0417 of the time
straight arch: ~0.287 of the time
Vikings on the other hand open with the below frequencies:
premill drush: ~0.139 of the time
post mill drush: 0.0763 of the time
men at arms: ~0.549 of the time
scouts: ~0.054 of the time
straight arch: ~0.183 of the time
You can imagine these rates are much closer than if I compared Bohemians to say Franks (where scouts openings tend to dominate across elos), and this is why Bohemians and Vikings are close in the image while Franks are far away from both these civs. For a similar example and details on how the image is actually generated, here is an example from the web for basketball players clustering by position/playstyle (which I don't really know anything about). I used slightly different tools than the ones they use there, but the same ideas apply roughly.
I think the plot mostly agrees with my experience with these civilizations, and there are some interesting groupings (for instance Bulgarians, Japanese, and Burmese all grouped presumably because their maa opening rate is larger than the other civs).