Reddemcee

An Adaptative Parallel Tempering wrapper for emcee 3 for personal use, which someone in the community might find useful on it's own.

Overview

Reddemcee is simply a wrapper for the excellent MCMC implementation emcee, that contains an adaptative parallel tempering version of the sampler, according to Vousden et al. implementation. It's coded in such a way that minimal differences in input are required, and it's fully compatible with emcee (v. 3.1.3).

Dependencies

This code makes use of:

Numpy
tqdm (https://pypi.python.org/pypi/tqdm)
emcee (https://github.com/dfm/emcee)

Most of them come with conda, if some are missing they can be easily installed with pip.

Installation

In the console type in your work folder

pip install reddemcee

Usage

Please refer to the test file in the tests folder.

import numpy as np
import reddemcee

def log_like(x, ivar):
    return -0.5 * np.sum(ivar * x ** 2)

def log_prior(x):
    return 0.0

ndim, nwalkers = 5, 100
ntemps = 5
ivar = 1. / np.random.rand(ndim)
p0 = list(np.random.randn(10, nwalkers, ndim))
sampler = reddemcee.PTSampler(nwalkers,
                             ndim,
                             log_like,
                             log_prior,
                             ntemps=ntemps,
                             adaptative=True,
                             logl_args=[ivar],
                             )
                             
sampler.run_mcmc(p0, 100, 2)  # starting pos, nsweeps, nsteps

Additional Options

ntemps betas pool adaptative config_adaptation_halflife rn: adaptations reduced by half at this time config_adaptation_rate rn: smaller, faster moves backend

Stored

ratios betas_history betas_history_bool ratios_history

Funcs

thermodynamic_integration(self, coef=3, sampler_dict = {'flat':False, 'discard':10})

get_Z(discard=1, coef=3, largo=100) get_attr(x) get_func(x)

ReddTea/reddemcee