Kernel CCA where the two input kernels can be specified.
The Randomized Dependence Coefficient (RDC). This is simply CCA on random Fourier features.
The linear-time independence test in this paper. Python code is available from the authors.
Block-estimator version of HSIC in this paper. This has O(n) runtime complexity (asymptotically).
Related to the previous point, in fact, all HSIC variants can be supplied with options of incomplete U-statistic estimators that the users can choose from, ranging from O(n) to O(n**2), depending on the computational budget (but trading off the variance). This will be very useful. More details on incomplete U-statistic estimators can be found here.
Is your feature request related to a problem? Please describe.
Describe the solution you'd like
Describe alternatives you've considered
Additional context (e.g. screenshots)