modAL

Why we included this project

Getting labels for unlabeled data is often the slowest and most expensive part of a machine learning project. modAL takes the active learning approach to that problem: instead of labeling everything, you wrap a scikit-learn classifier in an ActiveLearner and let it pick the most informative examples for a human to label, so you reach a given accuracy with far less annotation work. The design is deliberately modular. Query strategies, uncertainty measures, and the underlying estimator are all swappable pieces, and writing your own strategy amounts to a small function. It also handles active regression, for example steering a Gaussian process toward uncertain regions. Teams whose bottleneck is manual labeling will find the clean API a fast path from experiment to a working labeling loop.