snpgm

Originally a hack for the Computing the Universe 2015 workshop. Now a repository for testing out ideas for forward-model/hierarchical supernova cosmology inference.

About

We're performing inference on this Probabalistic Graphical Model (PGM):

It's specific to the SALT2 light curve model, in terms of the parameters that describe each light curve.

Dependencies:

• astropy
• sncosmo
• emcee
• triangle
• daft

... and the usual numpy/scipy/mpl business.

Scripts:

• gen_pgm.py: Script to draw the PGM. Generates snpgm.png.

• gen_dataset.py: Generate a test light curve data set, write each light curve to a file in the testdata directory. (In all scripts, directories are created if they don't already exist.)

• plot_testdata.py: Make a plot of the light curve data for each file in testdata, save to lcplots directory. Just to visualize the light curve data a bit.

• naive_sampling.py: Throw all the SN parameters and global parameters into a big MCMC and let it run. That's 4*N_SN + 4 parameters.

Importance sampling is two steps:

• sample_lcs.py: Run an MCMC on each light curve in testdata individually, save samples to samples directory as numpy binary files.

• importance_sampling.py: Run impotance sampling using the individual SN samples already created in previous step.

Importance sampling papers:

• Sonnenfeld et al "SL2S Paper 5" (strong lens ensemble)
• Schneider et al: Hierarchical WL