All Tools
S
DataFreemiumOpen Source
SNORKEL AI
Programmatic data labeling and weak supervision for AI
Apache-2.0
ABOUT
Creating labeled training datasets for supervised learning is the most expensive and time-consuming bottleneck in AI — manual labeling by domain experts does not scale, and rule-based labeling requires continuous maintenance. Snorkel introduced weak supervision, a paradigm that lets subject matter experts write noisy labeling functions (heuristics, keywords, knowledge bases, and external model outputs) and then uses a generative model to estimate each function's accuracy and combine their outputs into high-quality probabilistic training labels — dramatically reducing the cost of creating labeled data.
INSTALL
pip install snorkelINTEGRATION GUIDE
1. Generate training labels for text classification by combining keyword rules, regex patterns, and knowledge base lookups as labeling functions
2. Label large volumes of documents for information extraction using a mix of heuristics, pre-trained models, and human-validated slices
3. Automatically detect and resolve conflicts between noisy labeling sources with a trained label model that estimates each source's accuracy
4. Accelerate model development by quickly iterating on labeling strategies without waiting for human annotation cycles
5. Apply weak supervision to image, video, and multilingual data for domain-specific AI applications
TAGS
data-labelingweak-supervisiontraining-datadata-centric-ailabelingmlops