Baal
github.com
- Free plan
- Windows
- macOS
- Linux
- Self-hosted
6.8 Club score Visit github.com
What the maker publishes
- Active learning methods
- The framework supports Monte-Carlo Dropout, MCDropConnect, deep ensembles, and semi-supervised learning. source
- Dataset handling
- ActiveLearningDataset splits a dataset into training and unlabeled pool sets. source
- Deployment
- The repository provides Docker commands to build and run an experiment. source
- Framework components
- The framework has four main parts: ActiveLearningDataset, Heuristics, ModelWrapper, and ActiveLearningLoop. source
- Installation
- The README gives pip and Poetry installation methods. source
- Integrations
- The README links a seminar with Label Studio and describes training with Hugging Face or ModelWrapper. source
- License
- GitHub lists the project license as Apache-2.0. source
- Model preparation
- Baal provides modules for preparing models for active learning, including an MCDropoutModule wrapper. source
- Purpose
- Baal is an active learning library for industrial applications and research. source
- Python requirement
- Baal requires Python 3.10 or later. source
- Uncertainty-based selection
- ActiveLearningLoop computes uncertainty and labels the most uncertain items in the pool. source
Checked against the maker's pages on 14 Nov 2026.