Posterior predictive distributions obtained by fitting a "spiky" light curve using the cosine kernel (red solid line and shaded region) and periodic kernel (blue solid line and shaded region) defined in Equations 7 and 8, respectively. The light curve is sampled 4 times per day. Similarly to the sawtooth case in Figure 1, the cosine kernel is unable to properly describe the shape of the periodic signal, while the periodic kernel manages to capture it. The time t and signal f(t) are reported in arbitrary units.