zpace / pcay Goto Github PK

KDEUnivariate in find_pcs.py tries to import factorial from scipy.misc. This fails, so working around by hard-coding KDE plotting as false in find_pcs.PCA_Result.qty_hist. This is a hot-fix for mostly-unused functionality, not a long-term solution.

allow results files to include most any attribute of the PCA_Results object

Veldisp space isn't well-populated enough, doesn't make sense to realize only once per CSP

• initial CSP spectra get no blur at all
• each CSP spectrum gets blurred according to a random sample of (10 or so) veldisps
• can also add in custom (spectrograph-specific) LSF

using the biweight midcov from astropy 2.0 may help with spectral covariance computation

generate all models with appropriate velocity dispersions

do we adjust bounds of random sampling, or take all models and sample them uniformly?

plot fitted veldisp against DAP veldisp results

find the average signal to noise of reobserved spectra, and use that to scale the signal to noise in the spectral covariance calculations.

replace main diagonal with ivar

steps:

1. initialize K_inst with eigenspectra, pre-compute all projections onto PC space with diag(K_inst) = 0
2. compute PC covar cube resulting from ivar separately
3. call instrumental K_PC based on starting index

These cause a problem with mass-weighted age calculations, and also are just generally messy. Could get around them with wrapping directly around fsps fortran code and specifying a piecewise SFH

Check in Conroy+ '09 and with Bob/Jay

Replace StellarMass_PCA.reconstruct_full() and similar methods with a more flexible reconstruction object. Put it in reconstruct.py or something. Transparently implement spaxel-wise and cube-wise reconstruction.

new ivar should have units compatible with 1 / f**2.

add property to dered object

Natively, fsps uses basel or miles spectral library, which is much lower-res than MaNGA data is. Want Castelli+Kurucz+Conroy library, which is unpublished and unavailable (in fact, throws errors when run)

M/L profile of typical Sc (or whatever else)

Doesn't account for bursts in FSPS pre-computation, just uses integrate.quad on SFR (doesn't actually use masses at each age through FSPS).