We're Confused

• What metric will we use to assess 'working' for various inference tasks? Is this essentially saying 'get MCMC working' for the generative model we specify?
• How are we drawing a comparison between the VB and MCMC based inference methods?
• Are the generative models the for both MCMC and VB?
• What is Sky Flux?

Implementation Questions:

• What inference experiments have been conducted so far?
• How are chains mixing? Any simple ESS or R hat on the sampler?
• How successful are birth/death and merge/split moves? have been achieved yet?

Astro Details:

• How are pixel space coordinates converted to equatorial coordinates precisely.

  • The reference points are laid out in the FITS file headers (CRPIX*, CRVAL*, CD*_*), but which direction does that matrix, CD take you? According to this: http://stsdas.stsci.edu/documents/SUG/UG_21.html#6, it takes you from pixel coordinates to equatorial coordinates.

• What units are the equatorial coordinates in?

• What FITS params correspond to Kappa_n ("constant that scales nanomaggies"), and eta_n ("level of background noise"). I'm guessing eta_n is SKY and kappa_n is CALIB (or gain??)

• It looks like the point spread function is backwards in Fits_File_Details.ipynb (very consistent across multiple bands - will check across more sources). WHY!?