Weak Supervision Vote Matrix: Combining Noisy Labeling Functions

Before any model can train on data, someone has to label it — and hand-labeling every example is slow and expensive. Weak supervision replaces (or supplements) manual labels with a handful of cheap, imperfect heuristic rules called labeling functions, each of which fires on some fraction of the data and gets it right only some of the time. This simulator renders every labeling function as one row of a live vote matrix — one column per data point, sorted by true class — lets you tune how many labeling functions vote, how accurate and how thorough they are on average, and compares a naive majority vote against a generative label model that learns to trust the more reliable rows, the same idea used in real weak-supervision systems like Snorkel.