v1.2.5 - The Anisotropic Tree Update

Added

• Topological Split Strategies (TreeEffect): Formalized the split_strategy parameter. Users can now explicitly toggle between 'uniform' (creating mathematically orthogonal, shift-invariant Cartesian grids) and 'quantile' (applying the empirical Probability Integral Transform to create density-adaptive partitions that perfectly balance sample distributions across all leaves).

Changed

• Empirical Sparsity-Adaptive Penalty (Anisotropic Ridge): Upgraded the structural penalty of the Random Forest (t(...)) and Linear Tree (lt(...)) modules. By setting sp_alpha > 0, the initialization pass now accurately records the empirical data density of each terminal leaf ($C_i$).
• Drift & Singularity Prevention: Starved or empty edge-boundary leaves now receive geometrically massive penalties. This guarantees global matrix rank, eliminates the catastrophic test drift associated with hard Cartesian grids, and theoretically resolves the OLS singularities traditionally found in Model-Based Recursive Partitioning (MOB).

Release v1.2.4: Operational API Standardization

Added

• API Standardization: Introduced explicit fit() and predict() methods across all meta-models (AdaptiveTAM, OperaTAM, KalmanTAM). This establishes a unified, scikit-learn-like operational workflow (train on historical data, freeze state, predict out-of-sample) regardless of the underlying algorithm.
• AdaptiveTAM: Added fit() and predict() methods for production deployment. The fit() method efficiently extracts and solves the linear system strictly for the final available training window. The predict() method then applies this frozen state (last_state_dict_) to new data in $O(1)$ time with strict safety clipping, ensuring instant, deterministic inference without target leakage.
• KalmanTAM: Added fit() and predict() methods alongside end-of-training state extraction (last_state_dict_ and scale_dict_). This allows users to project the finalized Kalman drift weights forward as a stable, static rule on new data, with the internal normalization math handled automatically.
• OperaTAM: Added fit() and predict() methods to transition from continuous dynamic simulation to frozen-weight inference. fit() runs the historical simulation, while predict() cleanly extracts and applies the final expert aggregation weights to new out-of-sample data.

Fixed

• StaticTAM: Removed the target_col requirement from the required features check in decompose_prediction. This resolves a critical blocker for operational inference pipelines where the target variable is naturally unavailable.
• KalmanTAM: Patched _prepare_kalman_features to securely bypass target column extraction during out-of-sample inference, preventing crashes when the target variable is absent.

v1.2.3 - tam - Initial Public Release

TAM Framework v1.2.3 - Initial Public Release

This release marks the first official open-source release of the unified TAM framework. It aggregates all internal development milestones (v1.1.1 through v1.2.2), representing a complete architectural overhaul from the legacy weakl package (v0.0.6).

⚠️ Note to JOSS Reviewers: Features tagged with (BETA) (active research) and (EXP) (experimental) are strictly excluded from the stable JOSS evaluation scope.

🏗️ Major Architectural Overhaul

• Formula API: Transitioned from a legacy dictionary-based configuration to an intuitive, R-like Formula API (e.g., y ~ s(temp) + l(day)).
• Meta-Learner Ecosystem: Introduced a complete suite of object-oriented meta-learners, including StaticTAM (core fitting), AdaptiveTAM (online learning), OperaTAM (GPU-accelerated expert aggregation), KalmanTAM (BETA), AutoTAM (EXP) (evolutionary AutoML), SafetyTAM (EXP) (Conformal Prediction), and HierarchicalTAM (BETA).
• Math & Hardware Dispatchers: Implemented an intelligent routing layer that dynamically switches between chunked direct solvers and Matrix-Free Sparse Conjugate Gradient solvers based on topological complexity and available VRAM.

🌌 The "Spectrum" Core Effects Library

This release completes the "Spectrum" library, flattening diverse physical and statistical topologies into a single unified API:

• Standard & Continuous Bases: Added standard Linear terms (l), Real-valued Fourier harmonics (f) (transitioned from complex exponentials to real sine/cosine bases), Splines (s), and Chebyshev polynomials (p).
• Categorical & Signal Processing: Introduced Categorical effects (c) supporting nominal, ordinal, and fourier topological mappings, alongside Ricker Wavelets (w), and PID controller effects (pid) for dynamic derivative penalization.
• Advanced Functional Bases: Added RBF with both Gaussian and Matérn kernels (rbf), and Neural projections (n) (EXP) for embedding deep layers.
• Tree & Forest Integration: Integrated native GPU-based Random Forests (t), Flat N-ary Histograms to prevent matrix singularities, and varying-coefficient Linear Trees (lt).
• Interactions & Physics: Added Multivariate Tensor Products (te) for surface modeling and Universal Physics effects (phys) (BETA) for Physics-Informed Kernel Learning (PIKL) using ODEs/PDEs.

🚀 Performance & Mathematical Stability

• Native GPU Acceleration: Fully migrated intensive design matrix constructions (Splines, Wavelets, RBF) from CPU to GPU.
• Hardware Safeguards: Built a robust memory management module with smart chunking and dummy-pass VRAM footprint estimation to prevent CUDA Out-of-Memory crashes.
• OOD Extrapolation Wrapper: Introduced native Out-Of-Distribution extrapolation utilizing multidimensional directional derivatives, strictly bounding feature maps to the $[-1, 1]^F$ hypercube.
• GCV Auto-ML: Added Generalized Cross Validation (GCV) for automatic global regularization parameter selection, eliminating the need for validation sets.

🛠️ Ecosystem, MLOps, and Tooling

• Academic Benchmarks: Included official reproduction scripts for foundational load forecasting architectures (Pierrot-Goude 2011, Doumèche et al. 2025).
• MLOps Dashboard: Enhanced the plotting suite with chronological forecast smoothing, Test Set Vulnerability heatmaps, and unified model color mapping.
• Comprehensive Documentation: Published extensive markdown documentation bridging mathematical theory and framework architecture, alongside a full cheatsheet.py.