Learning the Threshold of Labeling Functions in Snorke

In this project, we develop a method to automatically pick the threshold inside of labeling functionsof Snorkel framework. Our method uses a data-driven method to choose a ”good” threshold forlabeling functions. This paper aims to reduce human efforts when users choose and define labelingfunctions. We have shown in both simulated data and a simple image classification task, our methodoutperforms the baseline method in terms of accuracy and F1 score.

For more details, please read project report inside the repo (project_report.pdf).