Regression pipeline of IWF-KNOW (ranked 4th). The light curves on the left are two examples in |$\boldsymbol{\mathit{\widetilde{X}}}_{p,k}$|⁠. The minima of the light curves were estimated using the 1st, 5th, and 10th percentiles. Subsequently, the minima were used to calculate the dips of the light curves ΔF_{p, k, j, r}. The square root of all light curve dips ΔF_{p, k, j, r} belonging to the same planet p (i.e. including all wavelengths j and all stellar spot instances k), and additionally the stellar and planetary parameters z_{p, 1}, …, z_{p, 6}, were then gathered in the feature vector |$\boldsymbol{\mathit{f}}_p^{*}$|⁠. The feature vector was z-score normalized (not shown in the graphic). Eventually, linear regressions were used to calculate the relative planet radius for each wavelength j.