SDSS V Milky Way Mapper: Science Requirements
What signal-to-noise ratios are required to deliver the requisite precision in stellar effective temperature, surface gravity, and chemical abundances from SDSS-V Milky Way Mapper?
Environment
conda create -n sdss python=3.6 anaconda
source activate sdss
conda install -n sdss -y numpy scipy matplotlib astropy ipython
git submodule init
git submodule update
cd AnniesLasso
python setup.py install
To-do
- Download Holtz training set
- Train model with existing Holtz training set
- Train model with strict(er) optimization requirements and compare to existing trained model
- Easily identify and extract individual visits
- Map out original label recovery as a function of S/N -- save results
Experiments
- Is DR14 optimized to the correct solution? Yes
- Is there weirdness going on because labels are so similar? Yes: REMOVE THEM&
- Map performance on individual visits with a trained model where we are confident that there is no weirdness going on
- Limit correlated information by prohibiting negative
:math:\theta
coefficients for absorption lines (which would add emission to the spectrum) - Test with and without windows
- Test with windows and RestrictedCannon
-
Test with windows and regularization - Test with RestrictedCannon and regularization
- Train using the ASPCAP best-fitting spectra for each star instead of data
- Script to make all comparison plots
Code
- tc.plot.theta to take label names
- tc.plot.theta to show censored regions (if they exist)
- tc.plot.theta to show bounded regions (if they exist)
- tc.plot.one_to_one to show in square format, if requested
- new progressbar
- Move SDSS-V MWM sci-req to SDSS github repository.