PSI-QPOT 0.1.1

PSI-QPOT: Post-Selection Inference for Sparse Estimators Cast as Quadratic Programs after Domain Adaptation

PSI-QPOT: Post-Selection Inference for Sparse Estimators Cast as Quadratic Programs after Domain Adaptation

PSI-QPOT is a Python package that implements a selective inference (SI) framework for conducting valid statistical inference after Quadratic Programming (QP)-cast Feature Selection algorithms in the presence of Optimal Transport (OT)-based domain adaptation (DA). The main idea is to leverages the SI framework and employs a divide-and conquer approach to efficiently compute the $p$ -value. Our proposed methods provides valid $p$-value for FS-DA results, by keeping the false positive rate (FPR) under control, while also maximizing the true positive rate (TPR), i.e., lowering the false negative rate (FNR).

Requirements

This package has the following requirements:

cvxpy
mpmath
numpy
POT
scikit-learn
scipy

Installation

Package Installation

This package can be installed using pip:

$ pip install psi_qpot

Usage

We provide several Jupyter notebooks demonstrating how to use the psi_qpot package in action.

• Examples for conducting inference for QP-cast Feature Selection after OT-based DA

>> ex1_PSI-QPOT.py

• Check the uniformity of the pivot

>> ex2_validity_of_p_value.py