BayeSN

bayesn.bayesn_model module

BayeSN SED Model. Defines a class which allows you to fit or simulate from the BayeSN Optical+NIR SED model.

class bayesn.bayesn_model.SEDmodel(num_devices=4, load_model='T21_model', filter_yaml=None, fiducial_cosmology={'H0': 73.24, 'Om0': 0.28})

Bases: object

BayeSN-SED Model

Class which imports a BayeSN model, and allows one to fit or simulate Type Ia supernovae based on this model.

Parameters:

num_devices: int, optional

If running on a CPU, numpyro will by default see it as a single device - this argument will set the number of available cores for numpyro to use e.g. set to 4, you can train 4 chains on 4 cores in parallel. Defaults to 4.

load_modelstr, optional

Can be either a pre-defined BayeSN model name (see table below), or a path to directory containing a set of .txt files from which a valid model can be constructed. Currently implemented default models are listed below - default is T21.

M20_model: Mandel+20 BayeSN model (arXiv:2008.07538).: Covers rest wavelength range of 3000-18500A (BVRIYJH). No treatment of host mass effects. Global RV assumed. Trained on low-z Avelino+19 (ApJ, 887, 106) compilation of CfA, CSP and others.
T21_model: Thorp+21 No-Split BayeSN model (arXiv:2102:05678).: Covers rest wavelength range of 3500-9500A (griz). No treatment of host mass effects. Global RV assumed. Trained on Foundation DR1 (Foley+18, Jones+19).
W22_model: Ward+22 No-Split BayeSN model (arXiv:2209.10558).: Covers rest wavelength range of 3000-18500A (BVRIYJH). No treatment of host mass effects. Global RV assumed. Trained on Foundation DR1 (Foley+18, Jones+19) and low-z Avelino+19 (ApJ, 887, 106) compilation of CfA, CSP and others.

fiducial_cosmologydict, optional

Dictionary containg kwargs {H0, Om0} for initialising an astropy.cosmology.FlatLambdaCDM instance. Defaults to Riess+16 (ApJ, 826, 56) cosmology: {H0:73.24, "Om0":0.28}.

filter_yaml: str, optional

Path to yaml file containing details on filters and standards to use. If not specified, will look for a file called filters.yaml in directory that BayeSN is called from.

Returns:

out: bayesn_model.SEDmodel instance

Attributes:

cosmo: `astropy.cosmology.FlatLambdaCDM`: Defines the fiducial cosmology assumed by the model when training
RV_MW: float: RV value for calculating Milky Way extinction
sigma_pec: float: Peculiar velocity to be used in calculating redshift uncertainties, set to 150 km/s
l_knots: array-like: Array of wavelength knots which the model is defined at
t_knots: array-like: Array of time knots which the model is defined at
W0: array-like: W0 matrix for loaded model
W1: array-like: W1 matrix for loaded model
L_Sigma: array-like: Covariance matrix describing epsilon distribution for loaded model
M0: float: Reference absolute magnitude for scaling Hsiao template
sigma0: float: Standard deviation of grey offset parameter for loaded model
RV: float: Global host extinction value for loaded model
tauA: float: Global tauA value for exponential AV prior for loaded model
spectrum_bins: int: Number of wavelength bins used for modelling spectra and calculating photometry. Based on ParSNiP as presented in Boone+21
hsiao_flux: array-like: Grid of flux value for Hsiao template
hsiao_t: array-like: Time values corresponding to Hsiao template grid
hsiao_l: array-like: Wavelength values corresponding to Hsiao template grid

Methods

dust_model:	Defines numpyro model for inferring dust parameters with population level SN parameters fixed based on previously trained model.
dust_redshift_model:	Defines numpyro model for inferring dust parameters with population level SN parameters fixed based on previously trained model, allowing the means of the RV and AV distribution to linearly evolve with redshift
dust_model_split_mag:	Defines numpyro model for inferring dust parameters, splitting the population in two based on host galaxy stellar mass as well as allowing an intrinsic magnitude offset between the bins. Population level SN parameters are fixed based on previously trained model.
dust_model_split_sed:	Defines numpyro model for inferring dust parameters, splitting the population in two based on host galaxy stellar mass as well as allowing an intrinsic SED difference between the bins. Population level SN parameters are fixed based on previously trained model.
fit_model_globalRV:	Defines numpyro model for fitting latent SN parameters including distance, conditioned on fixed population level parameters based on previously trained model. Assumes single global RV across population.
fit_model_popRV:	Defines numpyro model for fitting latent SN parameters including distance, conditioned on fixed population level parameters based on previously trained model. Assumes truncated Gaussian population RV distribution.
get_flux_batch:	Get integrated fluxes for BayeSN SED model across a large number of SNe, phases and bands.
def get_flux_from_chains:	Get model photometry from BayeSN SED model for posterior samples from model fitting chains.
get_mag_batch:	Get magnitudes for BayeSN SED model across a large number of SNe, phases and bands.
get_spectra:	Get spectra for BayeSN SED model across a large number of SNe and phases.
initial_guess:	Defined method used to initialise chains for model training.
parse_yaml_input:	Parse the input yaml file along with any command line arguments to define the job being run.
postprocess:	Postprocess the output of the MCMC run if required and save the chains and summaries.
process_dataset:	Process a set of data for use by the BayeSN model.
run:	Run an inference job using the BayeSN model.
sample_AV:	Sample AV from the population distribution based on a pre-trained model.
sample_del_M:	Sample delta_M from the population distribution based on a pre-trained model.
sample_epsilon:	Sample epsilon from the population distribution based on a pre-trained model.
sample_theta:	Sample theta from the population distribution based on a pre-trained model.
simulate_light_curve:	Simulate a light curve or set of light curves from the BayeSN SED model.
simulate_spectrum:	Simulate a specrum or set of spectra from the BayeSN SED model.
spline_coeffs_irr_step:	Vectorized version of spline coefficient calculations in spline_utils.
train_model_globalRV:	Defines numpyro model to train the BayeSN SED model assuming a truncated Gaussian RV population distribution.
train_model_popRV:	Defines numpyro model to train the BayeSN SED model assuming a single global fixed RV value across the population.

dust_model(obs, weights)

Numpryo model used to infer dust properties conditioned on fixed SN population parameters from a previously trained model.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band weights based on filter responses and MW extinction curves for numerical flux integrals

Returns:

dust_model_split_mag(obs, weights)

Numpryo model used to infer dust properties conditioned on fixed SN population parameters from a previously trained model, split into different mass bins above and below 10^10 solar masses. This model allows for a constant intrinsic magnitude offset between the two mass bins

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band weights based on filter responses and MW extinction curves for numerical flux integrals

dust_model_split_sed(obs, weights)

Numpryo model used to infer dust properties conditioned on fixed SN population parameters from a previously trained model, split into different mass bins above and below 10^10 solar masses. This model allows for a intrinsic difference in baseline SED (independent of light curve stretch) between the two mass bins

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band weights based on filter responses and MW extinction curves for numerical flux integrals

dust_redshift_model(obs, weights)

Numpryo model used to infer dust properties conditioned on fixed SN population parameters from a previously trained model, allowing the mean of the RV and AV distributions to linearly evolve with redshift.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band weights based on filter responses and MW extinction curves for numerical flux integrals

Returns:

fit(t, flux, flux_err, filters, z, ebv_mw=0, peak_mjd=None, filt_map={}, print_summary=True, file_prefix=None, drop_bands=[], fix_tmax=False, fix_theta=False, fix_AV=False, RV=False, mu_R=False, sigma_R=False, mag=False)

Method to fit light curve data loaded into memory with BayeSN model

Parameters:

t: array-like: Set of MJDs/rest-frame phases for light curve data to be fit. If you pass MJD and also a peak_mjd, values will automatically be converted to rest-frame phases
flux: array-like: Set of fluxes/mags for light curve data to be fit. Despite the name, you can use mags and if mag=True data will be automatically converted into flux for fitting.
flux_err: array-like: Set of flux/mag errors for light curve data to be fit. Despite the name, you can use mags and if mag=True data will be automatically converted into flux for fitting.
filters: array-like: Set of filters that flux/flux_err are measurements for, telling BayeSN which filters to use when fitting data. Must be of same length as flux/flux_err i.e. specify the filter for each data point individually
z: float: Heliocentric redshift of SN to be used when fitting
ebv_mw: float, optional: Milky Way E(B-V) value of SN. Defaults to 0.
peak_mjd: float or Boolean, optional: Fiducial value for maximum MJD of SN, used to convert phases to rest-frame. Note that this value only needs to be rough as BayeSN will fit for the time of maximum. However, if you set fix_tmax=True then this will be fixed as the time of maximum. Defaults to False, meaning that the code will assume phases are already rest-frame rather than MJD and will not do any conversion
filt_map: dict, optional: Dictionary providing mapping between filter names in file and BayeSN filters. Defaults to empty dictionary
print_summary: Boolean, optional: Specifies whether to print fit summary
file_prefix: str, optional: Prefix of name for output files containing summary table and MCMC samples. Default to None, in which case output files will not be saved and only returned for use in script.
drop_bands: array-like, optional: List of bands to be ignored during fitting. Defaults to empty list
fix_tmax: Boolean, optional: If True, tmax will not be inferred and fiducial value in file meta will be fixed. Defaults to False.
fix_theta: float, optional: Value to fix theta at during fitting. Defaults to False, meaning that theta will be inferred during fitting rather than fixed.
fix_AV: float, optional: Value to fix AV at during fitting. Defaults to False, meaning that AV will be inferred during fitting rather than fixed.
RV: float, optional: Value to fix RV at during fitting. Defaults to False, meaning that default model RV treatment will be used.
mu_R: float, optional: Value of mean of RV distribution to be used during fitting. Defaults to False, meaning that default model RV treatment will be used. If specified, sigma_R must also be specified.
sigma_R: float, optional: Value of standard deviation of RV distribution. Defaults to False, meaning that default model RV treatment will be used.
mag: Boolean, optional: Specifies whether data is mag or flux. If True, data is assumed to be mag and is automatically converted to flux before fitting.

Returns:

samples: dict: Dictionary containing parameter names as keys and MCMC samples as values
sn_props: tuple: Tuple containing SN redshift and MW E(B-V), which can be useful to have in memory when making plots

fit_from_file(path, filt_map={}, peak_mjd_key='SEARCH_PEAKMJD', print_summary=True, file_prefix=None, drop_bands=[], fix_tmax=False, fix_theta=False, fix_AV=False, RV=False, mu_R=False, sigma_R=False, mag=False)

Method to fit light curve contained in SNANA-format text file using BayeSN model

Parameters:

path: str: Path to SNANA-format text file containing data to be fit
filt_map: dict, optional: Dictionary providing mapping between filter names in file and BayeSN filters. Defaults to empty dictionary
peak_mjd_key: str, optional: Key to be used for peak MJD in SNANA text file meta. Defaults to ‘SEARCH_PEAKMJD’
print_summary: Boolean, optional: Specifies whether to print fit summary
file_prefix: str, optional: Prefix of name for output files containing summary table and MCMC samples. Default to None, in which case output files will not be saved and only returned for use in script.
drop_bands: array-like, optional: List of bands to be ignored during fitting. Defaults to empty list
fix_tmax: Boolean, optional: If True, tmax will not be inferred and fiducial value in file meta will be fixed. Defaults to False.
fix_theta: float, optional: Value to fix theta at during fitting. Defaults to False, meaning that theta will be inferred during fitting rather than fixed.
fix_AV: float, optional: Value to fix AV at during fitting. Defaults to False, meaning that AV will be inferred during fitting rather than fixed.
RV: float, optional: Value to fix RV at during fitting. Defaults to False, meaning that default model RV treatment will be used.
mu_R: float, optional: Value of mean of RV distribution to be used during fitting. Defaults to False, meaning that default model RV treatment will be used. If specified, sigma_R must also be specified.
sigma_R: float, optional: Value of standard deviation of RV distribution. Defaults to False, meaning that default model RV treatment will be used.
mag: Boolean, optional: Specifies whether data is mag or flux. If True, data is assumed to be mag and is automatically converted to flux before fitting.

Returns:

samples: dict: Dictionary containing parameter names as keys and MCMC samples as values
sn_props: tuple: Tuple containing SN redshift and MW E(B-V), which can be useful to have in memory when making plots

fit_model_globalRV(obs, weights, fix_tmax=False, fix_theta=False, theta_val=0, fix_AV=False, AV_val=0)

Numpyro model used for fitting latent SN properties with single global RV. Will fit for time of maximum as well as theta, epsilon, AV and distance modulus.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band-weights to calculate photometry
fix_tmax: Boolean, optional: If True, tmax will be fixed to fiducial value and will not be inferred. Defaults to False
fix_theta: Boolean, optional: If True, theta will be fixed to value specified by theta_val. Defaults to False.
theta_val: float or array-like, optional: Value to fix theta to, if fix_theta=True. Defaults to 0
fix_AV: Boolean, optional: If True, AV will be fixed to value specified by theta_AV. Defaults to False.
AV_val: float or array-like, optional: Value to fix AV to, if fix_AV=True. Defaults to 0

Returns:

fit_model_globalRV_vi(obs, weights)

Numpyro model used for fitting SN properties assuming fixed global properties from a trained model. Will fit for tmax as well as theta, epsilon, Av and distance modulus. This model is slightly modified for ZLTN VI.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band-weights to calculate photometry

fit_model_popRV(obs, weights, fix_tmax=False, fix_theta=False, theta_val=0, fix_AV=False, AV_val=0)

Numpyro model used for fitting latent SN properties with a truncated Gaussian prior on RV. Will fit for time of maximum as well as theta, epsilon, AV, RV and distance modulus.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band-weights to calculate photometry
fix_tmax: Boolean, optional: If True, tmax will be fixed to fiducial value and will not be inferred. Defaults to False
fix_theta: Boolean, optional: If True, theta will be fixed to value specified by theta_val. Defaults to False.
theta_val: float or array-like, optional: Value to fix theta to, if fix_theta=True. Defaults to 0
fix_AV: Boolean, optional: If True, AV will be fixed to value specified by theta_AV. Defaults to False.
AV_val: float or array-like, optional: Value to fix AV to, if fix_AV=True. Defaults to 0

fit_model_popRV_vi(obs, weights)

Numpyro model used for fitting latent SN properties with a truncated Gaussian prior on RV. Will fit for time of maximum as well as theta, epsilon, AV, RV and distance modulus. This model is slightly modified for ZLTN VI.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band-weights to calculate photometry

fit_model_uniformRV(obs, weights, fix_tmax=False, fix_theta=False, theta_val=0, fix_AV=False, AV_val=0)

Numpyro model used for fitting latent SN properties with single global RV. Will fit for time of maximum as well as theta, epsilon, AV and distance modulus.

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band-weights to calculate photometry
fix_tmax: Boolean, optional: If True, tmax will be fixed to fiducial value and will not be inferred. Defaults to False
fix_theta: Boolean, optional: If True, theta will be fixed to value specified by theta_val. Defaults to False.
theta_val: float or array-like, optional: Value to fix theta to, if fix_theta=True. Defaults to 0
fix_AV: Boolean, optional: If True, AV will be fixed to value specified by theta_AV. Defaults to False.
AV_val: float or array-like, optional: Value to fix AV to, if fix_AV=True. Defaults to 0

Returns:

get_flux_batch(M0, theta, AV, W0, W1, eps, Ds, RV, band_indices, mask, J_t, hsiao_interp, weights)

Calculates observer-frame fluxes for given parameter values

Parameters:

M0: float or array-like: Normalising constant to scale Hsiao template to correct order of magnitude. Typically fixed to -19.5 although can be inferred separately for different bins in a mass split analysis
theta: array-like: Set of theta values for each SN
AV: array-like: Set of host extinction values for each SN
W0: array-like: Global W0 matrix
W1: array-like: Global W1 matrix
eps: array-like: Set of epsilon values for each SN, describing residual colour variation
Ds: array-like: Set of distance moduli for each SN
RV: float: Global R_V value for host extinction (need to allow this to be variable in future)
band_indices: array-like: Array containing indices describing which filter each observation is in
mask: array-like: Array containing mask describing whether observations should contribute to the posterior
J_t: array-like: Matrix for cubic spline interpolation in time axis for each SN
hsiao_interp: array-like: Array containing Hsiao template spectra for each t value, comprising model for previous day, next day and t % 1 to allow for linear interpolation
weights: array_like: Array containing band weights to use for photometry

Returns:

model_flux: array-like: Matrix containing model fluxes for all SNe at all time-steps

get_flux_from_chains(t, bands, chains, zs, ebv_mws, mag=True, num_samples=None, mean=False)

Returns model photometry for posterior samples from BayeSN fits, which can be used to make light curve fit plots.

Parameters:

t: array-like: Array of phases to evaluate model photometry at
bands: array-like: List of bandpasses to evaluate model photometry in. Photometry will be
chain_path: str: Path to file containing BayeSN fitting posterior samples you wish to obtain photometry for
zs: array-like: Array of heliocentric redshifts corresponding to the SNe you are obtaining model fit light curves for.
ebv_mws: array-like: Array containing Milky Way extincion values corresponding to the SNe you are obtaining model fit light curves for.
mag: Bool, optional: Boolean to specify whether you want magnitude or flux data. If True, magnitudes will be returned. If False, flux densities (f_lambda) will be returned. Default to True i.e. mag data.
num_samples: int, optional: An optional keyword argument to specify the number of posterior samples you wish to obtain photometry for. Might be useful in testing if you are looking at lots of SNe, as otherwise this function will take a while to generate e.g. photometry for 1000 posterior samples across 1000 SNe. Default to None, meaning that photometry will be calculated for all posterior samples in chains provided.

Returns:

flux_grid: jax.numpy.array: Array of shape (number of SNe, number of posterior samples, number of bands, number of phases to evaluate), containing photometry across all SNe, all posterior samples, all bands and at all phases requested.

get_mag_batch(M0, theta, AV, W0, W1, eps, Ds, RV, band_indices, mask, J_t, hsiao_interp, weights)

Calculates observer-frame magnitudes for given parameter values

Parameters:

M0: float or array-like: Normalising constant to scale Hsiao template to correct order of magnitude. Typically fixed to -19.5 although can be inferred separately for different bins in a mass split analysis
theta: array-like: Set of theta values for each SN
AV: array-like: Set of host extinction values for each SN
W0: array-like: Global W0 matrix
W1: array-like: Global W1 matrix
eps: array-like: Set of epsilon values for each SN, describing residual colour variation
Ds: array-like: Set of distance moduli for each SN
RV: float: Global R_V value for host extinction (need to allow this to be variable in future)
band_indices: array-like: Array containing indices describing which filter each observation is in
mask: array-like: Array containing mask describing whether observations should contribute to the posterior
J_t: array-like: Matrix for cubic spline interpolation in time axis for each SN
hsiao_interp: array-like: Array containing Hsiao template spectra for each t value, comprising model for previous day, next day and t % 1 to allow for linear interpolation
weights: array_like: Array containing band weights to use for photometry

Returns:

model_mag: array-like: Matrix containing model magnitudes for all SNe at all time-steps

get_spectra(theta, AV, W0, W1, eps, RV, J_t, hsiao_interp)

Calculates rest-frame spectra for given parameter values

Parameters:

theta: array-like: Set of theta values for each SN
AV: array-like: Set of host extinction values for each SN
W0: array-like: Global W0 matrix
W1: array-like: Global W1 matrix
eps: array-like: Set of epsilon values for each SN, describing residual colour variation
RV: float: Global R_V value for host extinction (need to allow this to be variable in future)
J_t: array-like: Matrix for cubic spline interpolation in time axis for each SN
hsiao_interp: array-like: Array containing Hsiao template spectra for each t value, comprising model for previous day, next day and t % 1 to allow for linear interpolation

Returns:

model_spectra: array-like: Matrix containing model spectra for all SNe at all time-steps

initial_guess(args, reference_model='M20_model')

Sets initialisation for training chains, using some global parameter values from previous models. W0 and W1 matrices are interpolated to match wavelength knots of new model, and set to zero beyond the time range that the reference model is defined for. Note that unlike Stan, in numpyro we cannot set each chain’s initialisation separately.

Parameters:

reference_model: str, optional: Previously-trained model to be used to set initialisation, defaults to T21.

Returns:

param_init: dict: Dictionary containing initial values to be used

parse_yaml_input(args, cmd_args)

Method to parse the input yaml file and process data-set

Parameters:

args: dict: dictionary of arguments to define model based on input yaml file
cmd_args: dict: dictionary of command line arguments, which will override yaml file if specified
——-

postprocess(samples, args)

Function to postprocess BayeSN output. Applies transformations to some parameters e.g. ensuring consistency for W1 and theta, as flipping the sign in front of W1 and theta will lead to an identical result. Saves output chains and calculated a fit summary

Parameters:

samples: dict: Output of MCMC, dictionary containing posterior samples for each parameter with parameter names as keys
args: dict: dictionary of arguments to define model based on input yaml file and command line arguments

Returns:

process_dataset(args)

Processes a data set to be used by the numpyro model.

This will read in SNANA-format files, either in text or FITS format. This will read through all light curves and work out the maximum number of data points for a single object - all others will then be padded to match this size. This is required because to benefit from the GPU, we need to have a fixed array structure allowing us to calculate flux integrals from parameter values across the whole sample in a single tensor operation. A mask is applied in the model to ensure that these padded values do not contribute to the likelihood.

Generated data set is saved to the SEDmodel.data attribute, while the J_t matrices used to interpolate the W0, W1 and epsilon matrices are also calculated and saved to the SEDmodel.J_t attribute. Observer-frame band weights, including the effect of Milky Way extinction, are also calculated for the data set and saved to the SEDmodel.band_weights attribute.

Parameters:

args: dict: Combination of arguments from input yaml file and command line overrides, defines model wavelength range and data set to load

run(args, cmd_args)

Main method to run BayeSN. Can be used for either model training, fitting or dust inference with fixed population SN parameters, depending on input yaml file.

Parameters:

args: dict: dictionary of arguments to define model based on input yaml file
cmd_args: dict: dictionary of command line arguments, which will override yaml file if specified
——-

sample_AV(N)

Samples AV from model prior

Parameters:

N: int: Number of objects to sample for

Returns:

AV: array-like: Sampled AV values

sample_del_M(N)

Samples grey offset del_M from model prior

Parameters:

N: int: Number of objects to sample for

Returns:

del_M: array-like: Sampled del_M values

sample_epsilon(N)

Samples epsilon from model prior

Parameters:

N: int: Number of objects to sample for

Returns:

eps_full: array-like: Sampled epsilon values

sample_theta(N)

Samples theta from model prior

Parameters:

N: int: Number of objects to sample for

Returns:

theta: array-like: Sampled theta values

simulate_light_curve(t, N, bands, yerr=0, err_type='mag', z=0, zerr=0.0001, mu=0, ebv_mw=0, RV=None, logM=None, tmax=0, del_M=None, AV=None, theta=None, eps=None, mag=True, write_to_files=False, output_dir=None)

Simulates light curves from the BayeSN model in either mag or flux space. and saves them to SNANA-format text files if requested

Parameters:

t: array-like: Set of t values to simulate spectra at. If len(t) == len(bands), will assume that the t values correspond to the bands. Otherwise, will simulate photometry at each value of t for each band.
N: int: Number of separate objects to simulate spectra for
bands: array-like: List of bands in which to simulate photometry. If len(t) == len(bands), will assume that the t values correspond to the bands. Otherwise, will simulate photometry at each value of t for each band.
yerr: float or array-like, optional: Uncertainties for each data point, simulated light curves will be randomised assuming a Gaussian uncertainty around the true values. Can be either a float, meaning that the same value will be used for each data point, a 1d array of length equal to each light curve, meaning that these values will be used for each simulated light curve, or a 2d array of shape (N, light curve length) allowing you to specify each individual error. Defaults to 0, meaning that exact model photometry will be returned.
err_type: str: Specifies which type of error you are passing, either ‘mag’ or ‘flux’. Defaults to ‘mag’, meaning that this is in mag units. If you want to simulate fluxes and pass a mag error, it will be converted to a flux error.
z: float or array-like, optional: Redshift to simulate spectra at, affecting observer-frame wavelengths and reducing spectra by factor of (1+z). Defaults to 0. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same redshift will be used for all objects.
zerr: float, optional: Error on spectroscopic redshifts, only needed when saving to SNANA-format light curve files
mu: float, array-like or str, optional: Distance modulus to simulate spectra at. Defaults to 0. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. If set to ‘z’, distance moduli corresponding to the redshift values passed in the default model cosmology will be used. Technically these are heliocentric redshifts rather than Hubble diagram redshifts so won’t be perfect, but can be useful sometimes.
ebv_mw: float or array-like, optional: Milky Way E(B-V) values for simulated spectra. Defaults to 0. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects.
RV: float or array-like, optional: RV values for host extinction curves for simulated spectra. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the global RV value for the BayeSN model loaded when initialising SEDmodel will be used.
logM: float or array-like, optional: Currently unused, will be implemented when split models are included
tmax: float or array-like, optional: Time of maximum in rest-frame days, useful for plotting light curve fits with free tmax. Defaults to 0, i.e. the simulated time of maximum will be at 0 days. If a float is passed, the same value will be used for all objects.
del_M: float or array-like, optional: Grey offset del_M value to be used for each SN. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the prior distribution will be sampled for each object.
AV: float or array-like, optional: Host extinction RV value to be used for each SN. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the prior distribution will be sampled for each object.
theta: float or array-like, optional: Theta value to be used for each SN. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the prior distribution will be sampled for each object.
eps: array-like or int, optional: Epsilon values to be used for each SN. If passing a 2d array, this must be of shape (l_knots, tau_knots) and will be used for each SN generated. If passing a 3d array, this must be of shape (N, l_knots, tau_knots) and provide an epsilon value for each generated SN. You can also pass 0, in which case an array of zeros of shape (N, l_knots, tau_knots) will be used and epsilon is effectively turned off. Defaults to None, in which case the prior distribution will be sampled for each object.
mag: Bool, optional: Determines whether returned values are mags or fluxes
write_to_files: Bool, optional: Determines whether to save simulated light curves to SNANA-format light curve files, defaults to False
output_dir: str, optional: Path to output directory to save simulated SNANA-format files, onl required if write_to_files=True

Returns:

data: array-like: Array containing simulated flux or mag values
yerr: array-like: Aray containing corresponding errors for each data point
param_dict: dict: Dictionary of corresponding parameter values for each simulated object

simulate_spectrum(t, N, dl=10, z=0, mu=0, ebv_mw=0, RV=None, logM=None, del_M=None, AV=None, theta=None, eps=None)

Simulates spectra for given parameter values in the observer-frame. If parameter values are not set, model priors will be sampled.

Parameters:

t: array-like: Set of t values to simulate spectra at
N: int: Number of separate objects to simulate spectra for
dl: float, optional: Wavelength spacing for simulated spectra in rest-frame. Default is 10 AA
z: float or array-like, optional: Redshift to simulate spectra at, affecting observer-frame wavelengths and reducing spectra by factor of (1+z). Defaults to 0. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same redshift will be used for all objects.
mu: float, array-like or str, optional: Distance modulus to simulate spectra at. Defaults to 0. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. If set to ‘z’, distance moduli corresponding to the redshift values passed in the default model cosmology will be used.
ebv_mw: float or array-like, optional: Milky Way E(B-V) values for simulated spectra. Defaults to 0. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects.
RV: float or array-like, optional: RV values for host extinction curves for simulated spectra. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the global RV value for the BayeSN model loaded when initialising SEDmodel will be used.
logM: float or array-like, optional: Currently unused, will be implemented when split models are included
del_M: float or array-like, optional: Grey offset del_M value to be used for each SN. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the prior distribution will be sampled for each object.
AV: float or array-like, optional: Host extinction RV value to be used for each SN. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the prior distribution will be sampled for each object.
theta: float or array-like, optional: Theta value to be used for each SN. If passing an array-like object, there must be a corresponding value for each of the N simulated objects. If a float is passed, the same value will be used for all objects. Defaults to None, in which case the prior distribution will be sampled for each object.
eps: array-like or int, optional: Epsilon values to be used for each SN. If passing a 2d array, this must be of shape (l_knots, tau_knots) and will be used for each SN generated. If passing a 3d array, this must be of shape (N, l_knots, tau_knots) and provide an epsilon value for each generated SN. You can also pass 0, in which case an array of zeros of shape (N, l_knots, tau_knots) will be used and epsilon is effectively turned off. Defaults to None, in which case the prior distribution will be sampled for each object.

Returns:

l_o: array-like: Array of observer-frame wavelength values
spectra: array-like: Array of simulated spectra
param_dict: dict: Dictionary of corresponding parameter values for each simulated object

static spline_coeffs_irr_step(x_now, x, invkd)

Vectorized version of cubic spline coefficient calculator found in spline_utils

Parameters:

x_now: array-like: Current x location to calculate spline knots for
x: array-like: Numpy array containing the locations of the spline knots.
invkd: array-like: Precomputed matrix for generating second derivatives. Can be obtained from the output of spline_utils.invKD_irr.

Returns:

X: Set of spline coefficients for each x knot

train_model_globalRV(obs, weights)

Numpyro model used for training to learn global parameters, assuming a single global RV

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band weights based on filter responses and MW extinction curves for numerical flux integrals

train_model_popRV(obs, weights)

Numpyro model used for training to learn global parameters with a truncated Gaussian RV distribution

Parameters:

obs: array-like: Data to fit, from output of process_dataset
weights: array-like: Band weights based on filter responses and MW extinction curves for numerical flux integrals

bayesn.spline_utils module

BayeSN Spline Utilities. Defines a set of functions which carry out the 2D spline operations essential to BayeSN.

bayesn.spline_utils.cartesian_prod(x, y)

Compute cartesian product of two vectors.

Parameters:

xnumpy.array: First vector.
xnumpy.array: Second vector.

Returns:

znumpy.array: Cartesian product of x and y.

bayesn.spline_utils.invKD_irr(x)

Compute K^{-1}D for a set of spline knots.

For knots y at locations x, the vector, y’’ of non-zero second derivatives is constructed from y’’ = K^{-1}Dy, where K^{-1}D is independent of y, meaning it can be precomputed and reused for arbitrary y to compute the second derivatives of y.

Parameters:

xnumpy.array: Numpy array containing the locations of the cubic spline knots.

Returns:

KDnumpy.array: y independednt matrix whose product can be taken with y to obtain a vector of second derivatives of y.

bayesn.spline_utils.spline_coeffs_irr(x_int, x, invkd, allow_extrap=True)

Compute a matrix of spline coefficients.

Given a set of knots at x, with values y, compute a matrix, J, which can be multiplied into y to evaluate the cubic spline at points x_int.

Parameters:

x_intnumpy.array: Numpy array containing the locations which the output matrix will interpolate the spline to.
xnumpy.array: Numpy array containing the locations of the spline knots.
invkdnumpy.array: Precomputed matrix for generating second derivatives. Can be obtained from the output of invKD_irr.
allow_extrapbool: Flag permitting extrapolation. If True, the returned matrix will be configured to extrapolate linearly beyond the outer knots. If False, values which fall out of bounds will raise ValueError.

Returns:

Jnumpy.array: y independednt matrix whose product can be taken with y to evaluate the spline at x_int.

bayesn.spline_utils.spline_coeffs_irr_step(x_now, x, invkd)

bayesn.bayesn_io module

BayeSN I/O Utilities. Currently defines function for writing output SNANA-format light curves for simulated light curves

bayesn.bayesn_io.write_snana_lcfile(output_dir, snname, mjd, flt, mag, magerr, tmax, z_helio, z_cmb, z_cmb_err, ebv_mw, ra=None, dec=None, author='anonymous', survey=None, paper=None, filename=None)

Write user data to an SNANA-like light curve file

Parameters:

output_dirstr: Path to a directory where the file will be written. A default filename will be used, but you can specify your own with the filename argument. Default name format is snname[_survey][_paper].snana.dat, with the survey and/or paper being appended to the name if provided.
snnamestr: Name of the supernova
mjdlist or numpy.array: Modified Julian Dates of observations
fltlist or numpy.array of str: Filter idenitifiers of observations
maglist or numpy.array: Magnitudes of observations
magerrlist or numpy.array: Magnitude errors of observations
tmaxfloat: Estimated time of maximum
z_heliofloat: Heliocentric redshift
z_cmbfloat: CMB-frame redshift
z_cmb_errfloat: Error on CMB-frame redshift (excluding peculiar velocity uncertainty contribution)
ebv_mwfloat: E(B-V) reddening due to the Milky Way
rafloat, optional: Right Ascension, to be writen to the header if desired
decfloat, optional: Declination, to be written into the header if desired
authorstr, optional: Who is creating this file? Will be printed into the header’s preamble, if desired
surveystr, optional: Optional argumanet specifying the survey the data came from. Will be written into the header and filename if provided.
paperstr, optional: Optional argument specifying the paper the data came from. Will be written into the filename if provided.
filenamestr, optional: Custom filename to save as within output_dir. If not provided, a default format will be used. Do not provide an extension, as this will be added automatically.

Returns:

pathstr: Full path to the generated light curve file.

Notes

This will write a user’s data to the SNANA-like file format readable by out I/O routines. It will write the provided metadata into the file header, so this will be read in and used correctly by BayeSN. All vital metadata are required as inputs to this function.