search

andycasey / AnniesLasso

The Cannon 2: Compressed sensing edition
MIT License
10 stars 8 forks source link

regularization at the test step? #12

Open andycasey opened 8 years ago

andycasey commented 8 years ago

consider using ridge regression for regularization at the test step:

\argmin[\chi(l)^2 + \lambda(v(l).w)^2]

andycasey commented 8 years ago

@mkness when you ran all APOGEE spectra through with the 17 label model, did you get any weird results (e.g., weird abundances) in the test phase? @davidwhogg thinks we may have to regularize at the test step as well to penalize large \theta deviations from zero, and we thought it would be useful to consider if you saw any issues

davidwhogg commented 8 years ago

not "from zero" but from fiducial values.