pomp I have some contribution to pomp, a tool for working with partially observed Markov processes developed mainly by Aaron King.
It provides facilities for implementing POMP models, simulating them,
and fitting them to time series data by a variety of frequentist and
Bayesian methods. It is also a platform for the implementation of new
inference methods. I am actively working on an R package called iterated smoothing is2. With my adviser Edward Ionides, I have written an R package that implement three second-order iterated smoothing algorithms, exploiting efficiently estimation of the observed information matrix to increase convergence rate. Inherited from R package pomp, is2 can virtually be applied to any model developed by pomp. Since it is on Rforge, it is straightforward to install from R using install.packages("is2", repos="http://R-Forge.R-project.org"). pis2 Motivated by the favorable theoretical properties of particle Markov Chain Monte Carlo methods, I implement second-order
particle iterated smoothing, which can improve the estimations in term
of i) shortening the burn-in period, (ii) accelerating the mixing of the
Markov chain at the stationary period, and (iii) simplifying tuning. It
is in my github repository so you can download and install it from the source. RcppSQMC Sequential
Quasi-Monte Carlo was originally developed by Mathew Gerber and Nicolas
Chopin. Its theory sound promising. Currently, I find no R package that
do quasi-Monte Carlo sampling and using C and R codes of Mathew Gerber
and Nicolas Chopin is not straightforward. Therefore, I wrap the
original code by Rcpp and make the interface with R easier. I intend to
develop some iterated filtering and smoothing algorithms based on
quasi-Monte Carlo sampling instead of Monte Carlo sampling. You can find
the code on my github repository. |