Numerical LF builders

This article describes the set of numeric labeling function (LF) builders that are available for classification and information extraction applications.

Field length builder

Label data points based on the length of a specific field according to units (e.g., subject, body etc.)

Numeric builder

Label data points that match a numerical condition based on a numerical field. Using a non-numerical field will likely result in abstains for all data points. Selecting a field and label will enable the autotuner functionality, which will calculate the optimal numerical condition for the field and label based on labeled examples in the dev set. It will match or exceed the specified minimum precision (if possible) while maximizing recall.

note

For the email-spam application, if the email contains at least 3 images, it’s likely SPAM.

Bounding polygon builder

Label data points that are inside the polygon created by tracing the specified coordinates in the provided order, or on one of the vertices of the polygon. At least 3 coordinates must be provided.