README.md

In kkdey/tmcmcR: tmcmcR : a R/C++ package for Transformation based Markov Chain Monte Carlo

tmcmcR

A R/C++ package for fast implementataion of Transformation based Markov Chain Monte Carlo (TMCMC) Methods, which has higher acceptance rate and faster convergence in many scenarios, compared to the standard Random Walk Metropolis Hastings (RWMH) approach. The package implements R/C++ implementtaions for both the standard TMCMC and the RWMH algorithms, but can also be used for adaptive versions of both methods (SCAM, RAMA and Atchade methods). The package also implements Metropolis coupling variations of the TMCMC and the RWMH approaches (applicable for simulating from multimodal densities), together with a inverse temperature selection scheme with both fixed or randomized step designs.

Description

• Version: 0.1.2
• Authors: Kushal K Dey, Sourabh Bhattacharya.
• Maintainer: Kushal K. Dey
• License: GPL (>=2)

Installation

To install the tmcmcR package,

library(devtools)
install_github('kkdey/tmcmcR')

To load the package,

require(tmcmcR)

Functions

The chains we implement in this package, together with the exact functionality, include

• tmcmcR:::rwmh_metrop( ) : Standard Random Walk Metropolis Hastings (RWMH)
• tmcmcR:::tmcmc_metrop( ): Standard Transformation based Markov Chain Monte Carlo (TMCMC)
• tmcmcR:::adapt_rwmh_metrop( ) : Adaptive Random Walk Metropolis Hastings, method: SCAM, RAMA, Atchade
• tmcmcR:::adapt_tmcmc_metrop( ): Adaptive Transformation based Markov Chain Monte Carlo, method: SCAM, RAMA, Atchade
• tmcmcR:::rmc3( ): Metropolis Coupling Markov Chain Monte Carlo with RWMH update
• tmcmcR:::rtmc3( ): Metropolis Coupling Markov Chain Monte Carlo with TMCMC update (TMC3)

The last two methods need selection of inverse temperatures to run multiple chains and use these parallel chains for swapping at regular intervals. For target densities with product of known iid components, we have a selection scheme for inverse temperatures which may be of two types, depending on if the step size is fixed or randomized.

• tmcmcR:::select_inverse_temp( ): Inverse temperature selection scheme, method: TMCMC, RWMH, scheme: fixed, randomized.

As of now, the functions tmcmcR:::rmc3( ) and tmcmcR:::rtmc3( ) uses parallel::mclapply( ) for parallel implementation of chains at different inverse temperatures. However, this function is not compatible with Windows framework. For Windows users, we recommend replacing the mclapply( ) by mclapply.hack( ), a nice hack implemented by Nathan VanHoudnos, for which one needs to source this [script](source('http://www.stat.cmu.edu/~nmv/setup/mclapply.hack.R'). Check this blog post for more details on this functionality. I may switch to this hack for later versions of this package, so watch this space.

Vignettes

For checking example usage of functions under the tmcmcR package, check out the vignette. Also check the Github folder test for simulation examples R scripts on each of the princiapl functions of the package (we use these codes to generate the figures in the vignette).

Citation

If you are using the tmcmcR R package, please cite

KK Dey, S Bhattacharya. tmcmcR: R package for MCMC with improved acceptance rate and coverage. .

For the TMCMC methods and materials, please check

S Dutta, S Bhattacharya. Markov Chain Monte Carlo Based on Deterministic Transformations. Statistical Methodology

KK Dey, S Bhattacharya. On Optimal Scaling of Additive Transformation Based Markov Chain Monte Carlo. Submitted Brazilian Journal of Probability and Statistics.

KK Dey, S Bhattacharya. On Geometric Ergodicity of Additive and Multiplicative Transformation Based Markov Chain Monte Carlo in High Dimensions. Brazilian Journal of Probability and Statistics.

KK Dey, S Bhattacharya. On Single Variable Transformation Approach to Markov Chain Monte Carlo. JSM Proceedings. 2014.

Contact

For any queries, contact [e-mail]

Acknowledgements

• Gao Wang, University of Chicago
• Nan Xiao, University of Chicago

kkdey/tmcmcR documentation built on May 20, 2019, 10:39 a.m.