Write out more metadata about fit

snfactory / cubefit

Fit supernova + galaxy model on a SNFactory spectral data cube

MIT License

5 stars 5 forks source link

Write out more metadata about fit #23

Open kbarbary opened 8 years ago

kbarbary commented 8 years ago

Most metadata of interest is already written to the log file, but it would be good to have it in a structured form. The header and/or table extension of the result file would be an obvious place.

[x] final chisq per epoch
[x] initial positions and position bounds
[ ] total chisq, regularization penalty
[ ] time and number of iterations (total or for each step?)
[ ] Other measures of residuals that don't depend on pixel weights? (RMS?)

kbarbary commented 8 years ago

Actually, until there is some plan to ingest such values into the database, it's unclear to me how this will be useful or what the best output format is.

ycopin commented 8 years ago

There are plenty of technical values which are not stored in the DB, e.g. extract_star effective parameters. They can still be very useful for a posteriori analysis. You can store them in ad hoc files (e.g. json or simple txt files) that would be stored in the DB, or directly in output FITS file headers. Note that while it's pratically impossible to add new keys to the DB, it's relatively easy to store new files (figures or data dump).

Anyway, I suggest you start storing all quantities that you think could be useful for posterior debug or analyses.

kbarbary commented 8 years ago

OK, I'll just use the result FITS file.

kbarbary commented 8 years ago

I added position bounds and chisq per epoch to the result file. I don't plan to add any more diagnostics until I feel like they're needed, but I'll leave this issue open.