centering in autocorr #554
Description
The mean of x is subtracted before calculating the fft. However, the sample mean is ofcourse a biased estimate - especially if we have short chains. But we have a less biased estimate of the mean of a parameter if we use the mean from all chains+walkers. It should be slightly better for too short chains.
(An alternative idea is to calculate the ACF and the tau from the first-differences - Then there would be no need to center).