The code uses 2/k but I think this should be 1/k, in line with the correction of (m-1)/k for m continuous variables referred to in the MI estimator in this commit (which has not been applied here yet). Early testing with 2/k seems off by 1/k, which appears to confirm this theory is correct.
The code uses 2/k but I think this should be 1/k, in line with the correction of (m-1)/k for m continuous variables referred to in the MI estimator in this commit (which has not been applied here yet). Early testing with 2/k seems off by 1/k, which appears to confirm this theory is correct.