Closed googlebaba closed 3 years ago
@googlebaba The true HSIC cannot be computable with finite samples. We use the unbiased finite sample estimator of HSIC, which can be negative with a very small value. Note that the biased finite sample estimator will give you non-negative value. We describe the details in the pydoc: https://github.com/clovaai/rebias/blob/master/criterions/hsic.py#L22-L30
Thanks a lot.
Dear authors, When I conducted my experiments using your HISC implementation, I found the HSIC value could be negative. Is it reasonable? Do you find the phenomenon in your experiments? Thanks in advance!