“…For functional data, natural measures of proximity involve notions of shape as those given by the similarities of derivatives or FPC scores (Shang, 2016). This suggests the use of semi-metrics as distance measures for many kernel-based functional regression estimators, including functional local-constant/linear estimators (Ferraty and Vieu, 2002;Baíllo and Grané, 2009;Barrientos-Marin et al, 2010;Berlinet et al, 2011), functional k-nearest neighbours (Burba et al, 2009;Kudraszow and Vieu, 2013;Kara et al, 2017), functional recursive kernel estimator (Amiri et al, 2014), and Reproducing Kernel Hilbert Space (RKHS) methods (Preda, 2007;Avery et al, 2014). To avoid having to choose a semi-metric if several are available, Ferraty and Vieu (2009) and Febrero-Bande and González-Manteiga (2013) suggested to combine kernel estimators constructed with different semi-metrics.…”