Implementation for issue #445: Global forecasting support in test infrastructure

CHANGELOG.md

@@ -13,11 +13,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Enable `pin_memory` in DataLoaders when GPU is available for faster async CPU-to-GPU data transfers [\#236](https://github.com/mllam/neural-lam/pull/236) @abhaygoudannavar
 
+- Add `GlobalDummyDatastore` test variant for global forecasting support with all-zero boundary mask [\#445](https://github.com/mllam/neural-lam/issues/445)
+
 ### Changed
 
+- Refactor model class hierarchy: Extract `NormalizationManager`, `MetricTracker`, and `ModelVisualizer` from `ARModel` to improve separation of concerns and code maintainability ([#49](https://github.com/mllam/neural-lam/issues/49))
 - Change the default ensemble-loading behavior in `WeatherDataset` / `WeatherDataModule` to use all ensemble members as independent samples for ensemble datastores (with matching ensemble-member selection for forcing when available); single-member behavior now requires explicitly opting in via `--load_single_member` [\#332](https://github.com/mllam/neural-lam/pull/332) @kshirajahere
 - Refactor graph loading: move zero-indexing out of the model and update plotting to prepare using the research-branch graph I/O [\#184](https://github.com/mllam/neural-lam/pull/184) @zweihuehner
 - Replace `print()`-based `rank_zero_print` with `loguru` `logger.info()` for structured log-level control ([#33](https://github.com/mllam/neural-lam/issues/33))
+- Relax `test_boundary_mask` to accept all-zero boundary masks for global domains in support of global forecasting capabilities [\#445](https://github.com/mllam/neural-lam/issues/45)]
 
 ### Fixed
 

neural_lam/models/__init__.py

@@ -4,3 +4,6 @@
 from .graph_lam import GraphLAM
 from .hi_lam import HiLAM
 from .hi_lam_parallel import HiLAMParallel
+from .metric_tracker import MetricTracker
+from .model_visualizer import ModelVisualizer
+from .normalization_manager import NormalizationManager