snorkel-team/snorkel

What it solves

Snorkel addresses the bottleneck of manual data labeling in machine learning. Instead of having humans label thousands of individual examples by hand, it allows users to programmatically create and manage training data, reducing the time and effort required to build high-quality datasets.

How it works

It provides a framework for "weak supervision," where users can write labeling functions to programmatically assign labels to data. This approach brings mathematical and systems structure to the data creation process, allowing for iterative development and the ability to inject subject matter expertise into the training pipeline without manual labeling of every instance.

Who it’s for

It is designed for ML engineers, researchers, and subject matter experts who need to generate large amounts of labeled training data efficiently for machine learning projects.

Highlights

• Programmatic labeling of training data
• Framework for weak supervision modeling
• Support for data augmentation and multi-task learning
• Integration with various domains including medicine and industry