Flop clustering​

​

In Texas Hold'em Poker, there are 1755 different possible flops. Grouping or "clustering" these flops allows us to simplify our gameplan by treating different flops with similar characteristics the same way.

​

But which method of clustering represents all possible flops the best?

We performed different tests where different parts of ranges played against each other like "25% top range vs. 25% top range" or "33% bottom range vs. 66% bottom range" to figure that out. Measuring the average error per hand to the size of the subset gives a function with a 1/x behavior: 

​

As you can see, GTO+ and PioSolver do a great job of finding clusters where the average error converges to 0 quite fast. The gap per subset-size is related to different subsets with the same size. 

​

​

​