Getting Started
Check the topic you are interested in.
Installation
Requirements
Pip
In the console type
pip install reddemcee
From Source
In the console type
git clone https://github.com/ReddTea/reddemcee
cd reddemcee
python -m pip install -e .
APT Sampler
View the object detail here.
FAQ
Why reddemcee?
- APT is really well suited for multimodal posteriors, as well as very large prior-volumes.
- APT provides a reliable estimation of the bayesian evidence (Z)
- Easy to adapt emcee codes to reddemcee
What is APT?
Adaptative Parallel Tempering. I highly recommend checking proper literature as the ones suggested in the homepage!
In very few sentences, Adaptive Parallel Tempering MCMC is a variant of Markov Chain Monte Carlo (MCMC) designed to improve sampling efficiency, especially for multimodal distributions. It runs multiple chains in parallel at different "temperatures". Higher temperatures explore the distribution more broadly, while lower temperatures focus on finer details. Periodically, the chains swap states, enabling better exploration of the distribution's modes.
The "adaptive" component refers to dynamically adjusting the temperature levels based on past performance, making the method more efficient over time. This combination enhances convergence and reduces the risk of getting stuck in local minima.
How many walkers?
[In construction!]
- At least double the dimensions.
- I recommend a multiple of the threads you are using to minimise idle-core time.
- I like to scale them as an exponential of ndim.
- More walkers means more concurrent memory usage.
How many temps?
[In construction!]
- Really problem dependant.
- At least 6, usually 12, sometimes 16, exceptionally 24.