Closed EiffL closed 4 years ago
No, this is correct, because we would compare to a noise-subtracted estimate of the observed signal in any parameter estimation. It doesn't actually matter either way here, though, because for Fisher we don't use the mean. And for the SNR metric it would add a constant offset to all the results I think.
Great thanks for the clarification :-) I hadn't thought the data vector would come in noise subtracted. that answers it.
So, I'm puzzled by this line:
https://github.com/LSSTDESC/tomo_challenge/blob/c89d57eb195c5dcdd1e54294678549590e4af86a/tomo_challenge/metrics.py#L212
Shouldn't it be
C_obs
instead i.e. including the noise power spectrum in auto-spectra? or am I missing something?For the initial metric of S/N I wasn't too sure, but for the FoM, I would have definitely included the noise.