WS-Lite

A tiny weak-supervision toolkit for text classification.

Quickstart

python -m venv .venv
. .venv/bin/activate   # Windows: .\.venv\Scripts\activate
pip install -r requirements.txt
make reproduce
make ablation
make test

What it does

• Applies label functions to text

• Builds a label matrix with abstentions

• Majority vote and one-step weighted vote

• Trains a logistic regression on denoised labels

• Reports test accuracy and leave-one-LF-out ablation

NOTE : First run tries to download 20 Newsgroups; if SSL blocks it, the code uses a synthetic local dataset so everything still works.