cpf-diff-init

Source code and materials related to the article Conditional particle filters with diffuse initial distributions (Karppinen and Vihola, 2020). The code is written in Julia (version 1.3.1).

Getting started

1. Install Julia. For compatibility it is best to use version 1.3.1.
2. Clone the project with git clone https://github.com/skarppinen/cpf-diff-init.git.
3. Inside the project folder cpf-diff-init, run julia install_dependencies.jl. This script will install all Julia packages that are required by the project.

Descriptions of the source code files in the project

The relevant source code files are found in:

/data/covid/

Data used in the COVID-19 stochastic SEIR example.

/src/julia/lib/

Functionality related to particle filtering and implementations of the methods described in the paper.

/src/julia/models/

Source code for the noisy random walk, stochastic volatility, multivariate normal and SEIR models.

/src/julia/scripts/

Scripts for running the experiments. The full simulation experiments are computationally intensive, and the script download-simulation-summaries.jl can be used to download the (postprocessed) simulation data visible in the article from the data repository. After downloading, the scripts beginning with analyse can be run to reproduce the results of the article.

If needed, the individual experiments can also be run with the scripts beginning with run. Type

julia run-*experiment-name*.jl --help

for usage instructions. These scripts produce raw simulation output, which can be postprocessed with the script postprocess-simulations.jl. Run julia postprocess-simulations.jl --help for further details.

Running the scripts in this folder will produce a new folder output containing the generated data.