Latent Variable Models for Distributional Features

This repo contains all the code necessary to produce the results in the corresponding paper. Both the simulation study and illustrative example have been implemented using the targets package. We only provide basic instructions on how to run our code here, consult the documentation for more information.

Requisites

The code has been tested to run in R 4.4.2. All required packages are listed at the top of the *_targets.R files. Not all packages are available from CRAN. Check the comments within the package loading blocks for more information.

Simulation study

The simulation study was run on TU Dortmund's LiDO3 HPC cluster and may require modifications to be run on a different computing environment. Postprocessing of fitted models was done locally.

To reproduce the simulation study, follow these steps:

1. Copy the simulation-study folder into the computing environment were you intend to run the code.

2. Rename hpc_targets.R to _targets.R. If using Slurm, call sbatch run.sh. Otherwise, you will need a way to launch a long-running R process that calls targets::tar_make() until all fits are done.

3. Transfer the estimates.RDS file to the in folder in your local copy of the simulation-study folder.

4. Rename local_targets.R to _targets.R, then run targets::tar_make().

The repo already contains a copy of estimates.RDS and the resulting outputs for convenience.

Illustrative example

To run our re-analysis of Mader et al. (2023), follow the instructions of the README within the in folder to obtain the necessary data. Then, run targets::tar_make() to start fitting the models. The entire pipeline takes a few hours to run.