Skip to content

API Reference

Welcome to the highFIS API Reference. This section contains the complete reference documentation for all public modules, classes, and helper functions in the library.

The highFIS codebase is structured to decouple PyTorch model definitions from high-level scikit-learn estimators and optimization drivers.


1. Top-Level Estimators

For most machine learning tasks, you will interact with the scikit-learn compatible estimator wrappers. These wrappers handle the fit-predict cycle, metric evaluations, validation splits, and persistence.

  • highfis.estimators — Estimators Reference: Scikit-learn compatible classifiers (e.g., HTSKClassifier, DGTSKClassifier) and regressors (e.g., HTSKRegressor, DGTSKRegressor).

2. Core Architectures

If you are developing custom neural architectures or need direct access to the raw PyTorch models, use the core model definitions:

  • highfis.models — Model Architectures: PyTorch nn.Module classes representing TSK model variants (e.g., HTSKModel, LogTSKModel, DGTSKModel).
  • highfis.models.BaseTSK — Base TSK: Unified PyTorch training hooks and shared base classes.

3. Layer Primitives

highFIS builds neuro-fuzzy systems by combining custom PyTorch layer modules:

  • highfis.layers — Layer Modules: Core layers including FuzzificationLayer, RuleLayer, and consequent layers.
  • highfis.memberships — Membership Functions: Individual membership functions (e.g., GaussianMF, GaussianPiMF, CompositeGMF).
  • highfis.t_norms — T-Norms: Logical antecedent aggregation (e.g., Product, Minimum, Dombi, Yager, and adaptive softmin T-norms).
  • highfis.defuzzifiers — Defuzzifiers: Normalization engines that map rule activation weights to output weights.
  • highfis.gates — Gating Modules: Structural pruning gates used for feature selection and rule extraction.

4. Utilities and Protocols

  • highfis.optim — Optimizers & Trainers: Decoupled training drivers like GradientTrainer, DGTrainer, and FSRETrainer.
  • highfis.clustering — Clustering & Initialization: Algorithms for initializing fuzzy membership centers (e.g. k-means, grid, P-FRB).
  • highfis.persistence — Checkpoint Serialization: Versioned, secure checkpoint saving and loading utilities.
  • highfis.metrics — Evaluation Metrics: Custom statistical metrics for regression and classification validation.
  • highfis.protocols — Type Protocols: Structural typing interfaces ensuring API compatibility.