density_asymmetry(): the kernel density estimate is now wrapped around the circle (wrap = TRUE), so mass near +/- circ_space/2 is no longer truncated. This is a no-op to ~1e-13 for orientation or colour errors, but it matters for data with a 180-degrees-off reversal mode, such as motion direction.
density_asymmetry(): the +/- circ_space/2 point is now excluded from the positive and negative sums (exclude_antipode = TRUE), like zero. It is the same angle reached from two sides, and with an odd n it was counted twice.
density_asymmetry(): a bandwidth narrower than the grid spacing now rescales yvar and the bandwidth together (rescale_narrow = TRUE) so the kernel spans at least one cell. Without it, the rectangle sum aliases: on a narrow error distribution with bw.SJ()-selected bandwidth the returned asymmetry was off by 7.5% of its value at the default n = 181, now 0.4%.
density_asymmetry(): the two signs are now paired on a rounded abs(x), since seq() is not bitwise symmetric for every n (on n = 1801, 604 of 900 pairs failed to match and dropped out of the sums entirely).
density_asymmetry_discrete(): the same four fixes, with the same argument names and defaults. Wrapping there is done by evaluating stats::density() over the neighbouring periods and folding the images back, since augmenting the sample with shifted copies would run into density()’s binning range.
The fixes above default to the corrected behaviour; pass wrap = FALSE, exclude_antipode = FALSE, rescale_narrow = FALSE for pre-1.4.0 results.
circhelp 1.3.1
Added circ_mean(), circ_sd(), and angle_diff(), generalized versions of the _rad/_180/_360 family that accept an arbitrary period argument (e.g., 180, 360, or anything else).
Added a period argument to weighted_circ_mean() to compute the weighted circular mean directly on data in an arbitrary circular space, without manually converting to radians first.
circhelp 1.3
remove_cardinal_biases(): replaced gamlss dependency with a native P-spline + REML/EM implementation. Results are numerically equivalent; runtime is approximately 2× faster.