tlienart / ConstrainedPDMP

Notebook to reproduce the results presented in the paper "PDMP for Scalable Monte Carlo on Restricted Domains"

Constrained PDMP

In this repository, two notebooks are available:

• ConstrainedLogReg-demo.ipynb: shows how to declare a model using the PDMP library, it is similar though more specialised than the examples found in the documentation of PDSampler.jl.
• ConstrainedLogReg-exps.ipynb: which reproduces the experiment results presented in the paper Piecewise Deterministic Markov Processes for Scalable Monte Carlo on Restricted Domains by Bierkens et al.

Executing the second notebook completely may take a significant amount of time for all the experiments to be computed.

Installation and Requirements

For the demo

• Julia 0.6 (Julia 0.5 will also work)

In a Julia REPL:

Pkg.update()
Pkg.add("IJulia")

This demo uses the package PDSampler.jl which you can install by doing

Pkg.clone("https://github.com/alan-turing-institute/PDSampler.jl")
Pkg.build("PDSampler")

Next,

• quit Julia
• start Jupyter jupyter notebook
• open the notebook *-demo.ipynb and go through it.

Note: when using Julia 0.6, you may get warning boxes. This is because the package relies on Klara.jl, an external package that does not yet meet the new syntax requirements. If you'd like to not see such warnings, run this in a Julia REPL:

using IJulia
IJulia.installkernel("Julia nodeps", "--depwarn=no")

then restart the notebook and change the kernel to Julia nodeps.

For the experiments

• To display figures, you will need the PyPlot package.