MCMC has moved to Lora. Development in Lora will continue. MCMC is a placeholder for the future merge of various independent MCMC implementations in Julia, including Lora.
Furthermore, the Lora package has just gone through a major upgrade. For this reason, some aspects of the packages haven't been fully ported yet. Furthermore, the README package is not entirely up-to-date. The porting of the remaining code and the documentation will be completely ready in a few days' time. All the basic functionality of the package is already available as far as serial simulations are concerned.
The Julia Lora package provides a generic engine for Markov Chain Monte Carlo (MCMC) inference.
Briefly, Lora implements:
- imperative model specification,
- a range of Monte Carlo samplers,
- serial and sequential Monte Carlo methods,
- tuning of the samplers' parameters,
- various job managers for controlling the flow of simulations,
- descriptive statistics for MCMC and MCMC diagnostic tools,
- output managemement,
- resuming Monte Carlo simulations,
- Monte Carlo sampling with the help of automatic differentiation.
Jobs are the central input entities for handling MCMC simulations. A job is first instantiated to delineate the MCMC configuration. The main defining components of a job are the model, sampler and runner. Once set up, the job can be run or resumed.
Chains form the building block for managing the output of MCMC simulations. Jobs return chains. Descriptive statistics, MCMC diagnostics and output processing can be performed on chains.
- User Guide (PDF)
- Cheat Sheet (to appear soon)