Mean flux evolution in the training set (with redshift bins of size Δz = 0.1; red triangles) is compared to FG08 measurements from high-resolution and high signal-to-noise spectra (black circles). The continuum of the 78 SDSS spectra in the training set were fitted by hand using spline interpolation. Error bars from our measurements were computed by bootstrapping pixels of our sample. Both measurements are consistent, which makes us confident in our continuum estimate.