Contains an optimised implementation of FBNs in Cython for learning applications. Experiments are specified in yaml files and can be given as a list of configurations or a cartesian product thereof (be careful of combinatorial explosion in the number of experiment instances).
Includes three base optimisers, which all allow restarts:
- Hill Climbing
- Late-Acceptance Hill Climbing
- Simulated Annealing
Several multi-target methods can also be selected from including:
- Individual Classifiers (aka Binary Relevance)
- Classifier Chains (with and without target curricula)
- Adaptive Learning Via Iterated Selection and Scheduling
- Ensemble of Classifier Chains Also includes wrappers for various a priori curriculum generation methods.
This currently includes scripts for configuring and running experiments on a PBS cluster, however these will be excised into another repository in the future.