Bayesclumpy v2.0

Introduction

BayesClumpy is a computer program that can be used for the fast synthesis of spectral energy distributions (SED) emerging from clumpy dusty torus models. For the moment, only the models developed by the Kentucky group are included. The fundamental advantage of the code is that these fast synthesis capabilities are used in a Bayesian scheme for carrying out inference over the model parameters for observed SED. The code is written in standard Fortran 90 and Python.

Installation

There are different ways to install Bayesclumpy v2.0, but the best is to install it into a virtual environment either with pip or conda, which makes everything much more safer, plus making sure that all packages are installed for the code. For example, once you have installed Miniconda, you can generate a new environment and install the dependencies (you can install whatever version of Python 3 you desire). For consistency, it is better to use the gfortran compilers from Anaconda. In the following we show how install all packages with Anaconda. Anyway, feel free to use the gfortran compiler from your system but you might have some issues.

For a Linux OS

Within an Anaconda environment, standard packages can be installed with

conda create -n bayesclumpy python=3.8
conda activate bayesclumpy
conda install -c conda-forge cython numpy astropy tqdm scipy gfortran_linux-64 gcc_linux-64 nestle matplotlib configobj pysimplegui corner

For a Mac OS

Within an Anaconda environment, standard packages can be installed with

conda create -n bayesclumpy python=3.8
conda activate bayesclumpy
conda install -c conda-forge cython numpy astropy tqdm scipy gfortran_osx-64 gcc_osx-64 nestle matplotlib configobj pysimplegui corner

Now install the package into the environment by typing:

pip install -vv -e .

Usage

There is an example of usage in the directory example. The sampling is done with the following three lines. First import bayesclumpy, instantiate the Bayesclumpy class and call the sample() method. This generates a FITS file with the MCMC sampling.

import bayesclumpy as bc
bc = bayesclumpy.Bayesclumpy('conf.ini')
bc.sample()

You can then do posterior checks and plot the posterior by using

bc.corner('output.fits', pdf='corner.pdf')
bc.posterior_check('output.fits', pdf='posterior_check.pdf')

Everything is controlled from a human-readable configuration file.

[General]
Models = Nenkova
Sampler = nestle 
Use AGN = False
Extinction law = chiar06
Output file = circinus_bc.fits

[Observations]
File = observations/circinus.dat

[Priors]
Y          = uniform(5,100)
sigma      = uniform(15,70)
N          = uniform(1,15)
q          = uniform(0,3)
tauv       = uniform(10,150)
i          = uniform(0,90)
shift      = uniform(-3,3)
extinction = uniform(0,5)
redshift   = uniform(0.0,0.01)

The General section defines:

Models: set of models to use during inference. Only Nenkova is admitted now but we will include more models in the near future.
Sampler: specific posterior sampling algorithm [nestle/dynesty/ultranest]. At least one of them should be installed. In our experience, the nested sampler nestle works nicely.
Use AGN: include AGN spectra or not
Extiction law: select one among [no-extinction/allen76/seaton79/fitzpatric86/prevot84/calzetti00/chiar06/chiar_galcen06]
Output file: FITS file used for output.

The Observations section defines:

File: file with the observations. See the examples for guidance.

The Priors section defines the prior for each parameter of each specific model. Priors can be [uniform/normal/dirac]

Observations

The observations are entered in a file like the following one:

'Circinus'
8     0
'nacoJ'         <1.60   0.95    
'F160W'         4.77    0.7     
'nacoK'         19      1.9     
'naco2p42'      31      3.1     
'nacoLp'        380     38      
'nacoMp'        1900    190     
'trecsSi2'      5939    297     
'trecsQa'       14078   3520

The first line contains the name of the source. The next line contains the number of filters and spectropcopic data. For filters, one can provide measured values or upper limits. If measured values are provided, you must give the measured value and a estimation of the uncertainty (assumed normal). If an upper limit is given, you indicate it with a < and then provide the confidence value.

Dependencies

cython
numpy
astropy
tqdm
scipy
corner
nestle    
matplotlib
configobj
pysimplegui
corner

Optional dependencies

dynesty
ultranest

leontavares/bayesclumpy2