bayes_spec

bayes_spec is a framework for user-defined, cloud-based models of astrophysical systems (e.g., the interstellar medium) that enables spectral line simulation and statistical inference. Built in the pymc probabilistic programming library, bayes_spec uses Monte Carlo Markov Chain techniques to fit user-defined models to data. The user-defined models can be a simple line profile (e.g., a Gaussian profile) or a complicated physical model. The models are “cloud-based”, meaning there can be multiple “clouds” or “components” each with a unique set of the model parameters. bayes_spec includes algorithms to estimate the optimal number of components in a given dataset.

Useful information can be found in the Github repository and in the tutorials below.

Installation

conda create --name bayes_spec -c conda-forge pymc pip
conda activate bayes_spec
# Due to a bug in arviz, this fork is temporarily necessary
# See: https://github.com/arviz-devs/arviz/issues/2437
pip install git+https://github.com/tvwenger/arviz.git@plot_pair_reference_labels
pip install bayes_spec

Guides:

• Creating bayes_spec Models
  • Preliminary Steps
  • Model Structure
  • __init__()
  • add_priors()
  • add_likelihood()
  • Complete Model
  • Checking Model Specification
• Tips & Tricks
  • Checking model specification
  • Check prior distributions
  • Sampling with Variational Inference (VI)
  • Sampling with Monte Carlo Markov Chain (MCMC) methods
  • Convergence diagnostics
  • Optimization
  • Other Tips and Tricks

Tutorials:

• Basic Tutorial
  • Model Definition
  • Data Format
  • Simulating Data
  • Model
  • Posterior Sampling: Variational Inference
  • Posterior Sampling: MCMC
• Basic Tutorial: GaussNoiseModel
  • Data Format
  • Simulating Data
  • Model
  • Posterior Sampling: MCMC
• Optimization Tutorial
  • Model Definition and Simulated Data
  • Optimize
• Other Samplers
  • Data Format
  • Simulating Data
  • MCMC
  • nutpie
  • numpyro
  • blackjax
  • Sequential Monte Carlo

API:

• bayes_spec
  • bayes_spec