A guide's knots smoothed to their own noise: the cubic smoothing spline of
the knots' in-plane places over their stations (Reinsch 1967), so a guide
follows the reference medians where they agree and averages them where
they scatter.
Each region's knots are medians read one share of the stations at a time,
and the regions are read apart, so neighbouring knots carry scatter of
their own, crowd where two regions meet and fold where an extreme changes
branch. Interpolated exactly, all of it became the guide's shape. The
smoothing spline g minimises
sum_i |y_i - g(t_i)|^2 + alpha * integral |g''(t)|^2 dt
whose solution is a natural cubic spline with knots at the stations:
(R + alpha Q^T P Q) gamma = Q^T y, g = y - alpha P Q gamma, with Q
the second-difference matrix of the stations, R the tridiagonal of the
spline's continuity (Green and Silverman 1994, 2.3) and P the inverse
weights, zero at the two end knots: a guide's ends are its seams with the
web ring and the ankle ring, so they stay where they stand. The weight alpha is
set by the discrepancy principle: the residual's sum of squares equals the
knots' count times their noise variance. The noise is read from the knots
themselves, as the robust spread (1.4826 times the median absolute value)
of each knot's distance from the chord of its two neighbours, normalised to
unit variance (Gasser, Sroka and Jennen-Steinmetz 1986), so a guide's own
bends, which few knots carry, do not count as noise.
A guide's knots smoothed to their own noise: the cubic smoothing spline of the knots' in-plane places over their stations (Reinsch 1967), so a guide follows the reference medians where they agree and averages them where they scatter.
Each region's knots are medians read one share of the stations at a time, and the regions are read apart, so neighbouring knots carry scatter of their own, crowd where two regions meet and fold where an extreme changes branch. Interpolated exactly, all of it became the guide's shape. The smoothing spline
gminimiseswhose solution is a natural cubic spline with knots at the stations:
(R + alpha Q^T P Q) gamma = Q^T y,g = y - alpha P Q gamma, withQthe second-difference matrix of the stations,Rthe tridiagonal of the spline's continuity (Green and Silverman 1994, 2.3) andPthe inverse weights, zero at the two end knots: a guide's ends are its seams with the web ring and the ankle ring, so they stay where they stand. The weightalphais set by the discrepancy principle: the residual's sum of squares equals the knots' count times their noise variance. The noise is read from the knots themselves, as the robust spread (1.4826 times the median absolute value) of each knot's distance from the chord of its two neighbours, normalised to unit variance (Gasser, Sroka and Jennen-Steinmetz 1986), so a guide's own bends, which few knots carry, do not count as noise.