ConDo Adapter performs Confounded Domain Adaptation, which corrects for batch effects while conditioning on confounding variables. We hope it sparks joy as you clean up your data!
You can install the toolbox through PyPI with:
pip install condoNote: If you have issues with importing torchmin, you may need to install from source, as shown below. Or you can try re-installing pytorch-minimize from source.
After cloning this repo, install the dependencies on the command-line via:
pip install -r requirements.txtIn this directory, run
pip install -e .Import ConDo and create the adapter:
from condo import ConDoAdapterKLD
condoer = ConDoAdapterKLD()Try using it:
import numpy as np
X_T = np.sort(np.random.uniform(0, 8, size=(100, 1)))
X_S = np.sort(np.random.uniform(4, 8, size=(100, 1)))
Y_T = np.random.normal(4 * X_T + 1, 1 * X_T + 1)
Y_Strue = np.random.normal(4 * X_S + 1, 1 * X_S + 1)
Y_S = 5 * Y_Strue + 2
condoer.fit(Y_S, Y_T, X_S, X_T)
Y_S2T = condoer.transform(Y_S)
print(f"before ConDo: {np.mean((Y_S - Y_Strue) ** 2):.3f}")
print(f"after ConDo: {np.mean((Y_S2T - Y_Strue) ** 2):.3f}")More thorough examples are provided in the examples directory, and all code for experiments are in the papers directory.
In this directory run
pytestIf you use this toolbox in your research and find it useful, please cite ConDo using the following reference to the TMLR paper:
In Bibtex format:
@article{
mccarter2024towards,
title={Towards Backwards-Compatible Data with Confounded Domain Adaptation},
author={Calvin McCarter},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2024},
url={https://openreview.net/forum?id=GSp2WC7q0r},
note={}
}This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License.
