This repository contains code from the paper A Bias-Variance Decomposition for Ensembles over Multiple Synthetic Datasets.

Installing Dependencies

We use the Poetry package manager to manage Python dependencies. After installing Poetry according to their instructions, run

poetry install

to install the python dependencies.

R Dependencies

We use the R library Synthpop as one synthetic data generation algorithm. It can be installed by starting R with

R

and running

install.packages("synthpop")

Datasets

The datasets are included in this submission. The code we used to download the ACS 2018 dataset is in the notebook download_ACS_2018.ipynb.

Running the Code

We use Snakemake to run the experiments. The command

poetry run snakemake -j 5

will run all of the experiments with 5 jobs in parallel. This can take several days if run a single computer.

After the code has run, the notebooks in the plotting directory can be run to plot the figures, which will appear in the figures directory and its subdirectories.

The code for the random forest experiment is in the notebook random-forest-mse-prediction.ipynb.