Welcome to the documentation for baal (bayesian active learning)¶
baal aims to implement active learning using metrics of uncertainty derived from approximations of bayesian posteriors in neural networks.
For support, please see if your question is already answered in our FAQ. If it is not there, please file an issue or join our Gitter.
Baal has several components:
- Active learning dataset and training loop classes
- Methods for approximating bayesian posteriors
- API Index
- How to predict uncertainty per sample in a dataset
- How to know if my model is calibrated
- What to do if my models/datasets don’t fit in memory?
- How can I specify that a label is missing and how to label it.
- Tips & Trick for a successful active learning experiment