Module weak_supervision 

Weak Supervision and Label Model.

Combines noisy labels from multiple labeling functions to produce high-quality training labels without manual annotation.

Approach

Based on the Snorkel paradigm:

1. Define labeling functions (LFs) that provide noisy labels
2. Learn a label model that weighs LF outputs
3. Generate probabilistic labels for training

Reference

• Ratner, A., et al. (2017). Snorkel: Rapid Training Data Creation with Weak Supervision. VLDB.
• Ratner, A., et al. (2019). Training Complex Models with Multi-Task Weak Supervision. AAAI.

Structs

ConfidentLearning

Confident Learning for label noise detection.

LabelModel

Label Model for combining weak supervision sources.

Enums

LFOutput

Labeling function output.

Constants

ABSTAIN

Abstain value for labeling functions (no label provided).