For NARROW CMD cube around RC (go in very center of RC)--hope cmd avoids intense extinction
Use all 55 + 55 coefficients + G-J ---> error bar (mean flux over error)
Train on 8-fold testing thingy4.
MAKE validation figs* on APOGEE
Run on all XP spectra in neighborhood
Make figs* for XP spectra
Outline paper
Science case: make abundance maps better than Keith
Technical things APW needs to resolve:
Current likelihoods assume diagonal cov. matrices for labels and xp
switch to f params from APOGEE
don't believe f param uncertainties: assess noise params of f param (we dont need our noise model to be correct, we just need it to be responsible -Hogg)
FIGS:
make publication-ready plots of label vs. predicted
teff vs logg photos
mean spectra, individual stuff for each
Then select same box for all XP sources and then apply it.
LATER:
Neighborhoodize- each star is in 4 neighborhoods
Fix size or number in CMD (SNR will be mess)
Made decision about what we do about stars in not any neighborhood (nans, no label, etc) -- don't hit white dwarfs)
-Make decision: is extinction a label? apogee-wise extinction?
-Discuss: what other photometry could we use?
Hogg wants labels to be sines and cosines?
Hogg asks: why not use APOGEE spectra instead of APOGEE parameters
Catherine asks: can we try some GALAH labels of elements only found in the optical?
catherine-manea
commented
1 year ago
Train, test, and validation:
Technical things APW needs to resolve:
FIGS:
LATER:
Hogg wants labels to be sines and cosines? Hogg asks: why not use APOGEE spectra instead of APOGEE parameters Catherine asks: can we try some GALAH labels of elements only found in the optical?