snorkel
A system for quickly generating training data with weak supervision
What is snorkel?
A system for quickly generating training data with weak supervision
Overview of snorkel
What it is
Snorkel is a system for quickly generating training data with weak supervision. It was started at Stanford in 2015 and is now maintained as an open-source project. The core idea is that users can programmatically label, build, and manage training data, bringing mathematical and systems structure to the process of training data creation.
Key features
Programmatic labeling of training data using weak supervisionSupport for data augmentation, multi-task learning, and data slicingIntegration with labeling functions and other user-defined heuristicsCompatibility with Python 3.11 or laterInstallable via pip or conda
Use cases
Snorkel is designed for machine learning projects where training data is the key bottleneck. It is suitable for tasks such as text classification, information extraction, and other domains where labeled data is scarce or expensive to obtain. It has been used in production deployments at organizations like Google, Intel, and Stanford Medicine.
Getting started
To install Snorkel, use pip:pip install snorkelOr conda:conda install snorkel -c conda-forgeFor detailed tutorials and documentation, see the Snorkel website and tutorials repository.
License and pricing
Snorkel is open-source under the Apache-2.0 license.
Based on the project's public README and repository metadata.