Column/metadata updates after `getps`

[astrofle]

getps does not update some of the metadata of the result after using timeaverage on the result.

Example:

from dysh.fits import GBTFITSLoad
sdf = GBTFITSLoad(file)
getps = sdf.getps(scan, plnum=0)
ps = getps.timeaverage()[0]
ps.meta["DURATION"] # Is that of a single integration.
ps.meta["TUNIT7"] # This one is still "Counts"

The following columns are not being updated:

• DURATION
• TUNIT7
• LASTON
• LASTOFF

The following columns are updated, but do not agree with GBTIDL:

• RVSYS # GBTIDL keeps this one equal to 0 -- This difference should be fine as long as dysh is self consistent and the gbtgridder understands what this means if needed.

The following columns were not found in dysh but are in GBTIDL:

• FOFFREF1
• ZEROCHAN
• TWARM
• TCOLD
• INT # Not sure what this one is.

The columns below are keywords in the header, not sure why GBTIDL makes them columns now

• BACKEND
• PROJID
• TELESCOP
• SITELONG
• SITELAT
• SITEELEV

This one I don't think we care about

• NSAVE

[mpound]

We should come to a general consensus about how metadata should be treated when averaging spectra.

[mpound]

General Principle

When averaging integrations or ScanMixin derivatives, the header values of the resultant Spectrum should contain averages of [most] numeric values in the individual integrations/Scans.

The exceptions are:

• Values that are enumerations or indices, like CRVAL4 (Stokes parameter), PLNUM, ROW, IFNUM, will be aggregated, possibly changing type, e.g., header[‘PLNUM’] = [0,1] for polarization averaging or ‘first found’ (see handle_meta argument of Spectrum1D.mean)
• Time-related variables that should be summed (EXPOSURE, DURATION)
• Values that are NaN in the FITS header (e.g., ZEROCHAN) will not be propagated to the spectrum header (and currently are not, this appears to be a default astropy behavior).

Primary Header Disposition

Primary header variables such as TELESCOP, SITELONG, etc could be made into columns in the SDFITS table, though since they are constant it is unlikely that anyone would want to use them for selection. It makes more sense to have an user-level accessor in SDFITSLoad for the primary header. Also the primary header values would be added to spectral metadata (averaged or not) in Spectrum.make_spectrum()

Miscellaneous

TUNIT7 can take on whatever the flux unit is of the spectrum, but perhaps is better to be replaced with the more conventional FITS keyword BUNIT.

It is unclear what the purpose of LASTON and LASTOFF are; the description simply says “LAST ON/OFF for position switching.’ We can update them I guess, but that info would also be in to-be-implemented command logging.

INT is integration number (intnum) which GBTIDL allows selection on. A series of rows in SDFITS that are all part of the same integration (e.g., cal on, cal on, pols, sig, ref) all share the same integration number. Peter and I discussed this yesterday in a different context and agreed intnum was important to keep as a concept. Likely implementation would be that it is created as an SDFITS column on GBTFITSLoad instantiation, e.g., everything with the same OBSDATE gets the same intnum. I consider this a separate issue.

[astrofle]

Split into EXPOSURE and DURATION as high priority, the rest medium priority.