Postprocessing

Postprocessing#

lightning.postprocessing.postprocess_catalog(res_filenames, model_filenames, solver_mode='mcmc', model_mode='json', names=None, catalog_name='postprocessed_catalog.hdf5')#

Given lists of result files and model files, merge the results into a postprocessed catalog.

This script uses h5py to produce an output file in HDF5 format. We’ve made this choice to easily allow for non-homogeneous model setups, e.g. different numbers of bandpasses and parameters per source. The structure and content of the HDF5 file is as follows (for each source):

└──sourcename
    ├── mcmc
    │   ├── logprob_samples (Nsamples)
    │   └── samples (Nsamples, Nparams)
    ├── parameters
    │   ├── modelname
    │   │   └── parametername
    │   │       ├── best ()
    │   │       ├── hi ()
    │   │       ├── lo ()
    │   │       └── med ()
    └── properties
        ├── filter_labels (Nfilters)
        ├── lnu (Nfilters)
        ├── lnu_unc (Nfilters)
        ├── lumdist ()
        ├── mstar
        │   ├── best ()
        │   ├── hi ()
        │   ├── lo ()
        │   └── med ()
        ├── redshift ()
        └── pvalue ()

The “modelname” and “parametername” groups under the “parameters” group repeat for every model and parameter. For piecewise-constant SFHs the “properties” group also contains the age bin edges for the SFH. Quantiles (”/lo” and “/hi”) are computed at the 16 and 84th percentile.

For solver_mode=’mcmc’, the chains are also expected to be in HDF5 format, with the following structure:

mcmc
├── logprob_samples (Nsamples)
├── samples (Nsamples, Nparams)
└── autocorr (Nparams)

Whereas for solver_mode=’mle’, the results are expected to be formated as:

res
├── bestfit (Nparams)
└── chi2_best ()

I’ll provide a function to make such HDF5 result files soon if I haven’t already.

Parameters:
res_filenamesarray-like, str

A list of filenames pointing to the result files.

model_filenamesarray-like, str

A list of filenames poitning to the model files (either json or pickles).

model_modestr

Method for model serialization, either “json” or “pickle”. (Default: “json”)

namesarray-like, str

Names for each of the galaxies. If None (default), we’ll just guess based on the filenames of the chains assuming that they’re named something like [NAME]_chain.h5.

catalog_namestr

(Default: “postprocessed_catalog.hdf5”)

Returns:
None

Notes

TODO:

  • Allow user-supplied quantiles.

  • Add additional properties; allow user specification of properties? That might be a whole chore.

Contents