Adds `PropagationNet` GNN layer and makes it optionally usable in existing deterministic models#507

Open

Sir-Sloth-The-Lazy wants to merge 22 commits intomllam:mainfrom

Sir-Sloth-The-Lazy:refactor/batch-fold-ensemble-prep

Commits on Feb 21, 2026

refactor: add Forecaster and StepPredictor abstract base classes
Sir-Sloth-The-Lazy
committed
refactor: add ARForecaster implementing auto-regressive unrolling
Sir-Sloth-The-Lazy
committed
refactor: add ForecasterModule as pl.LightningModule wrapper
Sir-Sloth-The-Lazy
committed
refactor: change BaseGraphModel base class to StepPredictor
Sir-Sloth-The-Lazy
committed
refactor: update train_model.py and __init__.py for new hierarchy
Sir-Sloth-The-Lazy
committed
test: add tests for refactored model hierarchy and checkpoint loading
Sir-Sloth-The-Lazy
committed
refactor: archive legacy monolithic ar_model.py
Sir-Sloth-The-Lazy
committed
fix: update tests for new model hierarchy and fix pre-existing bugs
Sir-Sloth-The-Lazy
committed

Commits on Feb 24, 2026

Removed ar_model.py
Sir-Sloth-The-Lazy
committed

Commits on Feb 27, 2026

Merge branch 'main' into refactor/model-class-hierarchy-issue-49
Sir-Sloth-The-Lazy
authored

Commits on Mar 3, 2026

Commits on Mar 6, 2026

Commits on Mar 7, 2026

Broad namespace remap and added regression test
Sir-Sloth-The-Lazy
committed

Commits on Mar 18, 2026

Commits on Mar 19, 2026

Commits on Mar 22, 2026

Add PropagationNet GNN layer with optional integration into deterministic models
Sir-Sloth-The-Lazy
committed

Commits on Mar 24, 2026

integration test for the propagation_net for deterministic models
Sir-Sloth-The-Lazy
committed