April 10, 2018 Software Open Access

Fast and scalable non-parametric Bayesian inference for Poisson point processes

Gugushvili, Shota; van der Meulen, Frank; Schauer, Moritz; Spreij, Peter

Code and datasets accompanying the article "Gugushvili, van der Meulen, Schauer, Spreij (2018): Fast and scalable non-parametric Bayesian inference for Poisson point processes" (http://arxiv.org/abs/1804.03616)

Correspondence:

f.h.vandermeulen@tudelft.nl

Instruction:

- Put all files into one directory (R, jl and csv files); say the path is 'wd'.
- Make a directory named 'out' within this directory.
- Set the working directory in Julia to 'wd' and run 'include("ppp.jl").
- Within ppp.jl, the variable 'data_choice' can be set to analyse the datasets from the paper. For example 'data_choice="coal"' will analyse the coal-ming disaster data. Data can also be generated using 'data_choice="generated"'. In that case, in the file 'gen-extract-data.jl', the user has to provide the intensity function lambda, the right-end-point of the sampling interval T, the number of Poisson point process samples n, and an upper bound on lambda over the interval [0,T].
- Csv files containing simulation results are written to the directory ~\out 
- Run 'makeFigs.R' (adjust with setwd the working directory op top of this script) in R. 
- For the plot of the posterior mean, changing the value 'y_max' in the R-script may be necessary (it sets the upper bound on the vertical axis). 

Julia version 0.6.1. Julia dependencies:
- Package Distributions (version 0.15.0)
- Package Optim (version 0.14.0)

R version 3.4.3. R dependencies:
- ggplot2
- dyplyr
- gridExtra