Deep Compress-Then-Test (Deep CTT) accelerates deep kernel two-sample testing using high-fidelity compression.
For a detailed description of the Deep CTT algorithm and its power guarantees, see Low-Rank Thinning.
@inproceedings{carrell2025lowrank,
title={Low-Rank Thinning},
author={Annabelle Michael Carrell and Albert Gong and Abhishek Shetty and Raaz Dwivedi and Lester Mackey},
booktitle={Forty-second International Conference on Machine Learning},
year={2025},
url={https://openreview.net/forum?id=iAkg2nVmvN}
}To install the deepctt package, use the following pip command:
pip install git+https://github.com/microsoft/deepctt.gitTo test whether two samples, X and Y, are drawn from a common distribution, please follow these steps:
from deepctt import ctt
from deepctt.utils import train_deep_kernel
import torch
# Assumes the samples X and Y are numpy arrays of shape (n1,d) and (n2,d), respectively
n1, _ = X.shape
n2, _ = Y.shape
X_train, Y_train = X[:n1//2], Y[:n2//2]
X_test, Y_test = X[n1//2:], Y[n2//2:]
# Fit the deep kernel
model, sigma0, sigma, ep = train_deep_kernel(
X_train, Y_train, N_epoch, device, dtype, input_dim, learning_rate=5e-5, hidden_dim=20, embedding_dim=20
)
rejects, threshold_values, statistic_values = ctt(
torch.cat((model(X_test), X_test), dim=1),
torch.cat((model(Y_test), Y_test), dim=1),
g=0, # oversampling parameter
B=100, # number of permutations
alpha=0.05, # nominal level
sigma0=sigma0,
sigma=sigma,
ep=ep,
d_embd=embedding_dim,
)For an example usage, see our Higgs experiments.
This package has been tested with the following operating system, Python, and PyTorch combintations:
- Ubuntu 20.04, Python 3.12.9, Torch 2.6.0
- Ubuntu 20.04, Python 3.12.9, Torch 2.4.0
This project welcomes contributions and suggestions. Most contributions require you to agree to a Contributor License Agreement (CLA) declaring that you have the right to, and actually do, grant us the rights to use your contribution. For details, visit https://cla.opensource.microsoft.com.
When you submit a pull request, a CLA bot will automatically determine whether you need to provide a CLA and decorate the PR appropriately (e.g., status check, comment). Simply follow the instructions provided by the bot. You will only need to do this once across all repos using our CLA.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.
This project may contain trademarks or logos for projects, products, or services. Authorized use of Microsoft trademarks or logos is subject to and must follow Microsoft's Trademark & Brand Guidelines. Use of Microsoft trademarks or logos in modified versions of this project must not cause confusion or imply Microsoft sponsorship. Any use of third-party trademarks or logos are subject to those third-party's policies.