T1FF 0.1.0
- Initial public development release.
- Added Type-1 Fuzzy Function models for binary classification and
numeric regression using fuzzy C-means and membership-weighted local
predictions.
- Added formula and column interfaces, categorical predictor encoding,
feature scaling, and explicit missing-value handling.
- Added automatic separation detection, dependency-free ridge-logistic
fallback, and configurable probability clipping.
- Added validation, K-fold, and stratified K-fold hyperparameter
tuning with task-appropriate metrics.
- Added MAPE and SMAPE regression metrics to tuning, evaluation, and
nested benchmarking, with explicit handling of zero actual values.
- Added efficient joint tuning of cluster count, fuzziness, and
classification threshold for threshold-dependent metrics.
- Added model evaluation and repeated nested cross-validation with
matched logistic or linear regression baselines.
- Added optional probabilistic SVM and epsilon-SVR learners within
fuzzy clusters via
local_model = "svm".
- Tightened public-input validation for integer resampling controls,
logical scaling flags, classification labels, and validation split
sizes.
- Ensured unresolved final-model GLM instability warnings remain
visible after tuning while routine candidate-fit warnings stay
contained.
- Added a comprehensive vignette and PDF reference manual.