MultiLightning#

The main interface to multilightning is the MultiLightning class.

class multi_lightning.MultiLightning#

A class interface to fit “multi-region” spectral energy distributions: SEDs where multiple regions are resolved in some, but not all of the available bandpasses.

Parameters:

lghLightning object or list of Nregions Lightning objects

If a single Lightning object, the same model will be applied to each region (albeit with different parameters) and the Nregions keyword must be set. More generally, this can by a list of Lightning objects, one per region. Things like the redshift, luminosity distance, and filter set are expected to agree between different objects.

flux_obsnp.ndarray, (Nregions+1, Nfilters), float

An array giving the observed fluxes in mJy: the first row should contain the unresolved “global” fluxes, and subsequent rows should contain the resolved fluxes for each region. Missing or NA entries should be set to NaN. As an example, suppose we have 2 regions with 4 resolved optical measurements and 2 unresolved IR measurements. Our flux array would be structured like:

                  Opt1   Opt2   Opt3   Opt4  IR1  IR2
unresolved     [[  NaN,   NaN,   NaN,   NaN, 0.2,   1],
region1         [0.003, 0.002, 0.004, 0.002, NaN, NaN],
region2         [  NaN, 0.001, 0.002, 0.001, NaN, NaN]]

where we’ve supposed that region2 is undetected in band Opt1.

flux_uncnp.ndarray, (Nregions+1, Nfilters), float

An array giving the uncertainties on the fluxes in mJy. Here, missing and NA entries should be set to 0.0.

model_uncfloat

Fractional model uncertainty to apply to all bands.

Nregionsint

Number of resolved regions. Only necessary if lgh is a single Lightning object.

reg_nameslist of str

Names for individual regions, e.g. ‘core’, ‘disk’, ‘clumpA’, ‘clumpB’, etc. Defaults to ‘reg1’, ‘reg2’, etc. if not set.

Methods

`fit`(p0, priors[, Nwalkers, Nsteps, ...])	Fit the multi-region model with emcee.
`get_model_log_like`(params)	Calculate model log-likelihood.
`get_model_log_prior`(params, priors)	Calculate prior probability of parameters.
`get_model_log_prob`(params, priors[, p_bound])	Calculate model log probability.
`init_from_priors`(priors, Nwalkers)	Sample an initial state vector for emcee from the priors.
`print_params`([verbose])	List parameters (or pretty-print, when verbose=True)

__init__(lgh, flux_obs, flux_unc, model_unc=None, Nregions=None, reg_names=None)#

fit(p0, priors, Nwalkers=64, Nsteps=30000, init_sigma=0.001, progress=True, savefile=None)#

Fit the multi-region model with emcee.

Parameters:

priorsdict: A dictionary keyed on reg_names, where each entry is a list of the Nparamsx-many prior functions for each region. Prior functions should be specified by lightning.priors objects. Any prior == None is assumed to indicate a constant parameter.
Nwalkersint: Number of MCMC samplers for the emcee affine-invariant algorithm (Default: 64)
Nstepsint: Number of steps to run the MCMC (default: 30000)
init_sigmafloat: Sigma for gaussian ball initialization (default: 1e-3) UNUSED
progressbool: If True, print a tqdm progress bar. (Default: True)
savefilestr: Filename to use with the emcee HDF5 backend (if any). (Default: None)

Returns:

sampleremcee.EnsembleSampler

get_model_log_like(params)#

Calculate model log-likelihood.

Parameters:

paramsdict: A dictionary keyed on reg_names, where each entry is a numpy array with dimensions (Nmodels, Nparamsx), giving the parameters for each region. Note that in the most general case Nparamsx can be different for each region, while the first Nmodels axis is the vectorization axis, and should be the same for each region.

Returns:

lnlikenp.ndarray, (Nmodels,): -0.5 * chi2

get_model_log_prior(params, priors)#

Calculate prior probability of parameters.

Parameters:

paramsdict: A dictionary keyed on reg_names, where each entry is a numpy array with dimensions (Nmodels, Nparamsx), giving the parameters for each region. Note that in the most general case Nparamsx can be different for each region, while the first Nmodels axis is the vectorization axis, and should be the same for each region.
priorsdict: A dictionary keyed on reg_names, where each entry is a list of the Nparamsx-many prior functions for each region. Prior functions should be specified by lightning.priors objects.

Returns:

lnprior_probnp.ndarray, (Nmodels,): Prior log-probabilities

get_model_log_prob(params, priors, p_bound=inf)#

Calculate model log probability.

Parameters:

paramsdict: A dictionary keyed on reg_names, where each entry is a numpy array with dimensions (Nmodels, Nparamsx), giving the parameters for each region. Note that in the most general case Nparamsx can be different for each region, while the first Nmodels axis is the vectorization axis, and should be the same for each region.
priorsdict: A dictionary keyed on reg_names, where each entry is a list of the Nparamsx-many prior functions for each region. Prior functions should be specified by lightning.priors objects.
p_boundfloat: The magnitude of the log-probability of zero (default: np.inf)

Returns:

log_probnp.ndarray, (Nmodels,): Log probabilities

init_from_priors(priors, Nwalkers)#

Sample an initial state vector for emcee from the priors.

Parameters:

priorsdict: A dictionary keyed on reg_names, where each entry is a list of the Nparamsx-many prior functions for each region. Prior functions should be specified by lightning.priors objects. Any prior == None is assumed to indicate a constant parameter.
Nwalkersint: Number of MCMC samplers for the emcee affine-invariant algorithm

Returns:

x0np.ndarray, (Nwalkers, Ndim): Sampled initial state for emcee

print_params(verbose=False)#: List parameters (or pretty-print, when verbose=True)

MultiLightning

Contents

MultiLightning#