NEWS

SmoothPLS 0.1.4

Implemented dynamic load balancing for parallel computing in evaluate_lambda functions.
Numerical integration steps are now up to 8x faster on multi-core machines while preventing RAM overhead on massive datasets (e.g. >100.000 evaluations).

Documentation:
- Launched the official pkgdown website (hosted on GitHub Pages) including detailed vignettes and function references.
- Added the complete compiled PDF manual to inst/doc/.
README & Branding:
- Added a comprehensive quick-start example (One-State Categorical PLS) with visual outputs.
- Fixed LaTeX equations rendering for cross-compatibility between GitHub and Pandoc/pkgdown.
- Added institutional links (Decathlon SportsLab, Inria).
Continuous Integration:
- Configured GitHub Actions workflows for automated R CMD check and pkgdown site deployment.

Numerical Precision: Optimized categorical integration in evaluate_id_func_integral with stricter relative tolerance (rel.tol) and increased subdivisions (1000) for high-order B-splines.
Analytic Prediction: Implemented analytic L2 inner product for Scalar Functional Data (SFD) using fda::inprod, replacing discrete trapezoidal integration for near-perfect precision.
Safety Checks: Added time-range assertions in smoothPLS_predict to prevent silent errors when predicting on data outside the basis domain.

Tidyselect Compatibility: Fixed deprecation warnings by implementing all_of() in data pivoting functions.
Multivariate Support: Corrected logical assertions in smoothPLS to properly handle mixed lists of categorical and numerical predictors.
Integration Robustness: Added stop.on.error = FALSE in segment integration to handle micro-intervals without crashing the full model.

Core Test Suite: Added 70 unit tests covering Theorems (univariate equivalence), score orthogonality, and prediction consistency.
Edge Cases: Added tests for time-mismatch handling and multi-state categorical transitions.

Code Refactoring: Modularization of internal functions for Lambda matrix evaluation.
Synthetic Data: Improved generate_X_df and generate_Y_df for more realistic categorical state transitions.
S3 Structure Prep: Initial work on internal objects to support future S3 methods (print, plot, predict).

Thesis Milestone: First functional version used for the initial examples in the doctoral thesis.
Core Algorithms: Implementation of Smooth PLS for Hybrid Functional Data (CFD and SFD).
Basis Expansion: Support for B-spline basis representation of functional predictors.
Categorical Handling: Implementation of the "active area" integration concept for state-based predictors.