Optimal SNR Depends on Data Conditioning

[jacobgolomb]

This is likely not issue per se but may be a point about notation/definition.

When computing the optimal SNR using the routine here bilby uses the PSD array, which will include the strain_data.window_factor if the data were read in from a frame file and FFT'ed. This makes sense as a way to correct for the power loss in the noise specifically when computing the log likelihood, which calls the optimal SNR and matched filter SNR functions.

However, when using bilby to report the optimal SNR of signals as an output of an analysis, does it make sense for this to depend on how the data are conditioned given the optimal SNR is (h|h)? This may make sense if we are scaling the PSD to reflect the fact that the noise level in the data gets reduced by the window, but given that the optimal SNR is independent of the data, would it make more sense as a convention to report the optimal SNR with respect to the user-supplied PSD without scaling?