Feature Request: Rudimentary CLEAN for CubeLineMoment Line Extraction Efficiency

[jmangum]

CubeLineMoment requires that you provide it with what you think are the "known" lines in a cube. It then extracts the moments from this cube based just on your input. It would be nice to be able to assess how well one has extracted all of the spectral lines from a line cube. Perhaps something as simple as a "CLEAN-like" algorithm would be sufficient. Just calculate the total integrated intensity from all "signal" in a line cube and compare it to the total extracted integrated intensity from that line cube. Calculating the residual would then allow one to see what was missed (or if line overlap has resulted in some over-removal of signal).

[keflavich]

Interesting idea, but this should be done on a single-spectrum basis first. I'm skeptical whether a useful metric for "line ID completeness" can be developed; based on my own experience, the difference between a "real line" and noise is often unclear.

[jmangum]

I think that we are really looking for a metric which tells us how much of the "signal" has been extracted from a cube, rather than how well we have done in identifying spectral lines. What I can envision happening is iteratively giving CubeLineMoment a list of lines to extract, looking at the residuals, noting a low-level signal at a particular frequency, investigating whether this low-level signal corresponds to a known transition, the updating the lines to extract list accordingly.