refactor: model class hierarchy into Forecaster/StepPredictor layers

.github/workflows/ci-pypi-deploy.yml

@@ -16,14 +16,9 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: actions/setup-python@v6
-        with:
-          python-version: '3.14'
-      - name: Install uv
-        uses: astral-sh/setup-uv@v7
-      - name: Build with uv
-        run: uv build
+      - uses: actions/setup-python@v5
       - uses: casperdcl/deploy-pypi@v2
         with:
           password: ${{ secrets.PYPI_TOKEN }}
+          pip: wheel -w dist/ --no-deps .
           upload: ${{ github.event_name == 'release' && github.event.action == 'published' }}

.gitignore

@@ -81,14 +81,10 @@ tags
 .DS_Store
 __MACOSX
 
-# Virtual environments
+# pdm (https://pdm-project.org/en/stable/)
+.pdm-python
 .venv
-venv
-activate
-
-# Build artifacts
-dist/
-build/
-*.egg-info/
 
+# exclude pdm.lock file so that both cpu and gpu versions of torch will be accepted by pdm
+pdm.lock
 tests/test_outputs/

CHANGELOG.md

@@ -7,60 +7,14 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [unreleased](https://github.com/mllam/neural-lam/compare/v0.5.0...HEAD)
 
-### Added
-
-- Add `AGENTS.md` file to the repo to give agents more information about the codebase and the contribution culture.[\#416](https://github.com/mllam/neural-lam/pull/416) @sadamov
-
-- 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
-
-### Changed
-
-- 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))
-
 ### Fixed
 
-- Initialize `da_forcing_mean` and `da_forcing_std` to `None` when forcing data is absent, fixing `AttributeError` in `WeatherDataset` with `standardize=True` [\#369](https://github.com/mllam/neural-lam/issues/369) @Sir-Sloth-The-Lazy
-
-- Ensure proper sorting of `analysis_time` in `NpyFilesDatastoreMEPS._get_analysis_times` independent of the order in which files are processed with glob [\#386](https://github.com/mllam/neural-lam/pull/386) @Gopisokk
-
-- Switch to lat/lon-based plotting with `pcolormesh` and `cartopy` for accurate spatial visualisation regardless of underlying projection. [\#168](https://github.com/mllam/neural-lam/pull/168) @sadamov
-
-- Replace `shell=True` subprocess call in `compute_standardization_stats.py` with a safe argument list and Python-side hostname parsing to prevent command injection via `SLURM_JOB_NODELIST` [\#264](https://github.com/mllam/neural-lam/pull/264) @ashum9
-
-- Avoid NaN when standardizing fields with zero std [#189](https://github.com/mllam/neural-lam/pull/189) @varunsiravuri
-- Replaces multiple `assert` statements used for runtime input validation with explicit `ValueError` [\#279](https://github.com/mllam/neural-lam/pull/279) @Sir-Sloth-The-Lazy
-
 - Fix README image paths to use absolute GitHub URLs so images display correctly on PyPI [\#188](https://github.com/mllam/neural-lam/pull/188) @bk-simon
 
-- Fix typo in `ar_model.py` that causes `AttributeError` during evaluation [\#204](https://github.com/mllam/neural-lam/pull/204) @ritinikhil
-
 - Changed the hardcoded True to a conditional check "persistent_workers=self.num_workers > 0" [\#235](https://github.com/mllam/neural-lam/pull/235) @santhil-cyber
 
-- Avoid eager download of the MEPS example dataset during pytest collection by lazily initializing it in `tests/conftest.py`, allowing tests to run without triggering a dataset download at import time. [#391](https://github.com/mllam/neural-lam/pull/391) @Saptami191
-
-- `fractional_plot_bundle` now correctly multiplies by fraction instead of dividing
-[\#222](https://github.com/mllam/neural-lam/pull/222) @santhil-cyber
-
-- Fix `all_gather_cat` producing wrong shapes on single-device runs by only flattening when `all_gather` actually introduces a new leading dimension [\#424](https://github.com/mllam/neural-lam/pull/424) @RajdeepKushwaha5
-
-### Added
-
-- Expose `--wandb_id` CLI argument to allow resuming an existing W&B run by
-  ID. When provided, `resume="allow"` is set automatically so the same job
-  script works for both the initial submission and all resubmissions, making
-  it suitable for HPC systems with limited job runtimes or that may crash.
-  [\#197](https://github.com/mllam/neural-lam/pull/197) @Mani212005
-
-
-
-- Fix Slack domain link [\#288](https://github.com/mllam/neural-lam/pull/288) @sadamov
-
-### Fixed
-
-- Infer spatial coordinate names for MDPDatastore (rather than assuming names `x` and `y`), allows for e.g. lat/lon regular grids [\#169](https://github.com/mllam/neural-lam/pull/169) @leifdenby
-
+- `fractional_plot_bundle` now correctly multiplies by fraction instead of dividing [
+\#222](https://github.com/mllam/neural-lam/pull/222) @santhil-cyber
 
 ### Maintenance
 
@@ -74,10 +28,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 - Fix caching of MEPS example data in CI/CD [\#181](https://github.com/mllam/neural-lam/pull/181) @observingClouds
 
-- Migrated build backend from PDM to Hatchling with hatch-vcs and added uv build in deploy CI
-
-- Warn when running with `--eval` without `--load` to avoid accidentally evaluating randomly initialized weights [#190](https://github.com/mllam/neural-lam/pull/190) @varunsiravuri
-
 ## [v0.5.0](https://github.com/mllam/neural-lam/releases/tag/v0.5.0)
 
 This release contains maintenance and fixes, preventing some unexpected crashes and improving CICD and testing.

README.md

@@ -1,4 +1,4 @@
-[![slack](https://img.shields.io/badge/slack-join-brightgreen.svg?logo=slack)](https://kutt.to/mllam)
+[![slack](https://img.shields.io/badge/slack-join-brightgreen.svg?logo=slack)](https://kutt.it/mllam)
 [![Linting](https://github.com/mllam/neural-lam/actions/workflows/pre-commit.yml/badge.svg?branch=main)](https://github.com/mllam/neural-lam/actions/workflows/pre-commit.yml)
 [![CPU+GPU testing](https://github.com/mllam/neural-lam/actions/workflows/install-and-test.yml/badge.svg?branch=main)](https://github.com/mllam/neural-lam/actions/workflows/install-and-test.yml)
 

neural_lam/create_graph.py

@@ -11,7 +11,6 @@
 import scipy.spatial
 import torch
 import torch_geometric as pyg
-from loguru import logger
 from torch_geometric.utils.convert import from_networkx
 
 # Local
@@ -231,7 +230,7 @@ def create_graph(
     """
     os.makedirs(graph_dir_path, exist_ok=True)
 
-    logger.info(f"Writing graph components to {graph_dir_path}")
+    print(f"Writing graph components to {graph_dir_path}")
 
     grid_xy = torch.tensor(xy)
     pos_max = torch.max(torch.abs(grid_xy))
@@ -594,8 +593,9 @@ def cli(input_args=None):
     )
     args = parser.parse_args(input_args)
 
-    if args.config_path is None:
-        raise ValueError("Specify your config with --config_path")
+    assert (
+        args.config_path is not None
+    ), "Specify your config with --config_path"
 
     # Load neural-lam configuration and datastore to use
     _, datastore = load_config_and_datastore(config_path=args.config_path)