Snorkel
Definition
Snorkel helps label training data using rules and other weak signals.
Instead of labeling each dataset individually, a team formulates labeling functions. Snorkel can combine their sometimes conflicting results to obtain probabilistic training labels.
Simply explained
Multiple imperfect cues together produce a suggestion for a label.
Why it matters in practice
Existing domain knowledge can support the creation of large training datasets.
Example from business
Hypothetical example: A team labels support tickets based on terms, senders, and existing rules.
Limits and pitfalls
- Faulty or interdependent rules can amplify biases; real test data remains necessary.
Origin & history
Snorkel originates from research on weak supervision of learning methods.
Use in mid-sized companies
- Preparation of training data for text classification and document routing.
Frequently asked questions
Does Snorkel replace every manual review?
No, especially the quality of the rules and independent test data must be checked.
Sources
- API-Dokumentation · Snorkel: API-Dokumentation –
Editorially reviewed on: