Benchmark#
Method#
The page lists the computer and the package versions with the results. The sample size goes from 100 to 100 000 (HSIC and KCI: to 10 000), and |Z| goes from 0 to 8. Each point runs 20 queries (HSIC and KCI: 10) one time, and the first call is only a warm-up call. The result is the mean time of the other calls, with its standard error. When a point takes more than 60 s, the package skips the larger points. The benchmark README gives the full method.