RealPatch

The supplementary matherial file is written in RealPatch_supplementary.pdf.

1. Create Environment with necessary dependencies

Install Poetry (if not already installed)

curl -sSL https://raw.githubusercontent.com/python-poetry/poetry/master/get-poetry.py | python -
# To configure your current shell run
source $HOME/.poetry/env

poetry --version
# Update poetry to the latest version
poetry self update

Install Dependencies with Poetry

# Enter code directory
cd code 
# Install dependecies
poetry install

Activating poetry environment

# Check the virtual environment created and activated
poetry env list
<env_name>

# If not activated, activate the virtual environment (entering the output of previous command)
source `poetry config virtualenvs.path`/<env_name>/bin/activate

2. Data

The data are hosted in a google drive folder.

To run the code you can either manually download the data and place them in the data folder here. Be sure all the required files are present:

features_file.npz
labels_train.csv
labels_val.csv
labels_test.csv

Or as an alternative, you can run the following:

python download_data.py

which will download the four files and place them in the data folder for you.

3. Run RealPatch

python realpatch.py data=celeba

4. Results

The script will produce multiple outputs in the outputs folder for evaluating the matching results. The Evaluation folder will contain 3 .csv files. One saving the number of training examples before and after matching (similar to Table 10 in Appendix), and two to assess the achieved covariate balanced in terms of SMD and VR (consistent to Table 3).

An example of results for a single run of CelebA with selected RealPatch hyperparameters is in the folder output/2022-03-14/16-13-25.

wearepal / RealPatch