pyscreeningfs 0.1.1

This is a Python implementation by the authors of the paper 'Online Feature Screening for Data Streams With Concept Drift' from Dr. Mingyuan Wang and Dr. Adrian Barbu. Contain various feature selection methods.

Online-Feature-Screening-for-Datastream-with-Sparsity-Concept-Drifting

This is a Python implementation by the authors of the paper "Online Feature Screening for Data Streams With Concept Drift" from Dr. Mingyuan Wang and Dr. Adrian Barbu.

Please cite this paper if you use or build on our method. doi.org/10.1109/TKDE.2022.3232752

This project enabled well-known feature screening methods, including gini index, chi-square score, mutual information, fisher-score, T-score to handle streaming data, batch data, data with drifting, and sparse data. It currently only works on binary classification data.

Installation

Prerequisites

• Python 3.10 or newer
• pip
• numpy 2.2.4 or newer

Note

Although the package is designed OS independent, it was only tested on Windows. You might need to use methods listed below other than pip install pyscreeningfs.

For users installing from source (e.g., if no pre-built wheels are available for your system): You will need a C++ compiler compatible with your Python installation:

• Windows: Microsoft Visual C++ Build Tools (part of Visual Studio, or standalone).
• Linux: gcc and g++ (usually included or easily installed via your package manager, e.g., sudo apt-get install build-essential).
• macOS: Xcode Command Line Tools (install with xcode-select --install).

Install via git clone

1. Clone repository

git clone https://github.com/yourusername/repo_name.git

2. Navigate into the cloned repository directory

cd repo_name 

3. Install

pip install .

Install via download

1. Download the repository
2. Unpack to your own folder your_folder/repo_name
3. Navigate into the unpacked repository directory

cd repo_name  

4. Install

pip install .

Install via pip (Currently unavailable)

If pre-built wheels are available for your system on PyPI (coming soon!), you can install directly:

pip install pyscreeningfs

Data

For .svm sparse data, visit https://www.sysnet.ucsd.edu/projects/url/
Download and put into data/url_svmlight/

For any input data/data files, the Y/label/class vector can only contain numeric value and one of the label must be 1.

Demo

For a demo, see testing.py in the root directory.