README.md

In tlamadon/mopt: Parallel Derivative-free Moment Estimation

mopt

Parallel Derivative Free Moment Optimization in R

This is a package intended for people who want to estimate a model based on some distance criteria. The library is built around the square minimization of the distance between a set of moments. The value of the objective function as well as the value of the moments at each evaluations are stored for later analysis.

The parallelisation overhead is relatively costly, and so I would not recommend using this library if solving your model for a parameter set is under 1 second. My experience has been that a solving time of around 30 seconds is a good place to start.

Features:

• works with MPI and OpenMP, but also in searial mode
• offers serial optimization see serial optim example using the minqa package.
• offers several MCMC implementations (see list of algorithms)
• see bgp example for Likelihood-Free Parallel Tempering by Baragatti Grimaud and Pommeret
• offers several reporting tools, but also exports to classic mcmc types to use in R
• offers a function to compute slices of the objective function in orthogonal directions, see example.

References:

• Approximate Bayesian computation
• Monte Carlo Markov Chain
• "Determining the density of states for classical statistical models: A random walk algorithm to produce a flat histogram " Wang, F. and Landau, D. Physical Review E  64  056101  (2001)
• "An MCMC approach to classical estimation" Chernozhukov and Hong.

Install

install.packages('devtools');require(devtools);install_github('mopt',user='tlamadon')

Use MPI on UCL econ HPC

• Install Rmpi on HPC
• copy .Rprofile into your working directoy

See also

A Julia version of mopt exists as MOpt.jl on github.

Contributors

• Ran Gu
• Thibaut Lamadon
• Florian Oswald

tlamadon/mopt documentation built on May 31, 2019, 3:48 p.m.