sfs-da 1.0.4

A package for Statistical Inference for Feature Selection after OT-based Domain Adaptation

Statistical Inference for Feature Selection after Optimal Transport-based Domain Adaptation (AISTATS 2025)

This package implements a statistical inference for feature selection (FS) after optimal transport-based domain adaptation (OT-based DA). The main idea is to leverages the SI framework and employs a divide-and conquer approach to efficiently compute the $p$-value. By providing valid $p$-values for the selected features, our proposed method not only controls the false positive rate (FPR) in FS under DA but also maximizes the true positive rate (TPR), i.e., reducing the false negative rate (FNR). We believe this study represents a significant step toward controllable machine learning in the context of DA.

See the paper https://arxiv.org/abs/2410.15022 for more details.

Requirements

This package has the following requirements:

• numpy
• mpmath
• scipy
• scikit-learn

We recommend to install or update anaconda to the latest version and use Python 3 (We used Python 3.12.3).

NOTE: We use scipy package (version 1.13.1) to solve the linear program (simplex method). However, the default package does not return the set of basic variables. Therefore, we slightly modified the package so that it can return the set of basic variables by replacing the two files '_linprog.py' and '_linprog_simplex.py' in scipy.optimize module with our modified files in the folder 'files_to_replace' at files_to_replace.

How to Automatically Replace the Files

If Using Anaconda

• First, initialize and activate the target environment (if you want to use your conda base environment, you should replace 'your-env-name' by 'base'):

$ conda init
$ conda activate your-env-name

• Then, run the following command to replace the necessary files (file 'replace_scipy_linprog.py' is available at: replace_scipy_linprog.py):

python replace_scipy_linprog.py --env anaconda --dir files_to_replace

If Using System Python (Non-Anaconda) - Run the following command:

$ python replace_scipy_linprog.py --env python --dir files_to_replace

Installation

This package can be installed using pip:

$ pip install sfs_da

Usage

We provide several Jupyter notebooks demonstrating how to use the sfs-da package in our examples directory.

• Example for computing $p$-value for Lasso after DA

>> ex0_p_value_lasso_DA.ipynb

• Example for computing $p$-value for Elastic Net after DA

>> ex1_p_value_elasticnet_DA.ipynb

• Check the uniformity of the pivot

>> ex2_validity_of_p_value.ipynb