specsrbench.figures.fig4_mae_summary

Figure 4 – global reconstruction fidelity, three ways.

Three bar panels over the nine methods: raw MAE, MAE normalised by the reference’s own flux uncertainty, and per-spectrum RMSE. The table printed alongside is Table A2 of the paper, and tests/test_paper_consistency.py checks the manuscript against it row by row.

Read the MAE panel with the amplitude column beside it. SR2 leads it by 30%, and does so by producing a spectrum at 0.54 of the reference’s scale: absolute error against a noisy reference falls when you shrink toward zero, whatever the reconstruction quality. MAE_scalefree in summary_final.csv is the same comparison with that route closed, and there SR2 places eighth of nine.

Module Attributes

GAUSSIAN_ABS_FLOOR

E[|N(0, sigma)| / sigma] -- where the uncertainty-normalised panel would sit for a reconstruction that is perfect up to the reference's own noise.

Functions

build([cache, outdir])
compute(cache)	The rows of Table A2, plus the two reference lines the panels draw.

specsrbench.figures.fig4_mae_summary.GAUSSIAN_ABS_FLOOR = 0.7978845608028654

E[|N(0, sigma)| / sigma] – where the uncertainty-normalised panel would sit for a reconstruction that is perfect up to the reference’s own noise.

specsrbench.figures.fig4_mae_summary.compute(cache)

The rows of Table A2, plus the two reference lines the panels draw.

specsrbench.figures.fig4_mae_summary.build(cache=None, outdir=None)

Parameters:

outdir (Path | None)

Return type:

Path