v1.0.0

[1.0.0] — 2026-05-07 (package) | Model: v1.1.0

First release of ise-py on PyPI. This version represents a full rewrite of the
original ise package, introducing ISEFlow as the primary model, GrIS support,
HuggingFace Hub weight distribution, validated input dataclasses, anomaly conversion,
and a complete test suite and CI/CD pipeline.

Changed

• Breaking: Package renamed from ise to ise-py on PyPI; import name stays ise.
• Breaking: from_raw_values() renamed to from_absolute_forcings() on both
  ISEFlowAISInputs and ISEFlowGrISInputs; old name kept as a deprecated alias
  (emits DeprecationWarning) until further versions.
• Breaking: Model classes restructured — ise.models.ISEFlow.ISEFlow,
  ise.models.predictors.deep_ensemble, and ise.models.density_estimators.normalizing_flow
  replaced by top-level ISEFlow, DeepEnsemble, NormalizingFlow, and LSTM in
  ise.models.
• Pretrained weights moved to HuggingFace Hub (pvankatwyk/ISEFlow); downloaded
  automatically on first use via huggingface_hub. Falls back to bundled local weights
  when HuggingFace is unavailable (air-gapped HPC / local dev).
• Build system switched from setup.py + setuptools to Hatchling (pyproject.toml);
  runtime dependencies cleaned up — removed transitive pins, moved dev/docs/GPU deps to
  optional extras. Removed heavy deps from core: seaborn, cartopy, geopandas,
  pyproj, xesmf, clisops, owslib, statsmodels.
• Replaced flake8 + isort + black with ruff; updated pyproject.toml and
  pre-commit config.
• mypy fixes: replaced = None defaults with | None unions across forcings.py,
  grids.py, inputs.py, process.py, feature_engineer.py, training.py,
  functions.py, and anomaly.py; explicit type annotations on ForcingFile and
  GridFile instance attributes.
• ISEFlow_AIS.predict / ISEFlow_GrIS.predict marked # type: ignore[override] to
  satisfy mypy while preserving the intentional signature narrowing.
• Variable shadowing fixed in ProjectionProcessor and get_model_densities.
• backfill_outliers docstring corrected (bfill, not ffill).
• AnomalyConverter updated to support GrIS MAR integration; GrIS anomaly variables
  (aSMB, aST) pass through unchanged (already anomalies in ISMIP6).
• NaN handling added in NormalizingFlow sampling.
• Introduced separate package vs. model versioning (see above).

Added

• ISEFlow_AIS and ISEFlow_GrIS — convenience subclasses of ISEFlow for
  AIS (18 sectors) and GrIS (6 basins) with pretrained weight loading.
• ISEFlowAISInputs / ISEFlowGrISInputs — validated input dataclasses for
  model.predict(), with from_absolute_forcings() classmethod for raw forcing values.
• AnomalyConverter — converts absolute sector-averaged forcings to ISMIP6
  anomalies using bundled climatology CSVs; called internally by from_absolute_forcings().
• ForcingFile (ise/data/forcings.py) — loads and formats climate forcing NetCDF
  files to 86-step ISMIP6 time series, with depth aggregation and sector assignment.
• GridFile (ise/data/grids.py) — loads sector boundary grids.
• NormalizingFlow (ise/models/normalizing_flow.py) — standalone autoregressive
  masked affine flow replacing the old density_estimators/normalizing_flow.py.
• LSTM (ise/models/lstm.py) — standalone LSTM replacing predictors/lstm.py,
  with variable sequence length and save/load support.
• DeepEnsemble (ise/models/deep_ensemble.py) — rewritten ensemble, trained on
  [X, z] where z is the NF latent; captures epistemic uncertainty.
• ise/data/utils.py — convert_and_subset_times() for xarray time handling.
• ise/utils/io.py — check_type() runtime type validation helper.
• ise/data/data_files/ismip6_model_configs.json and
  ise/data/data_files/GrIS_ismip6_model_configs.json — ISM configuration lookup tables.
• ise/data/data_files/AIS_atmos_climatologies.csv and
  ise/data/data_files/GrIS_atmos_climatologies.csv** — bundled ISMIP6 climatology
  baselines used by AnomalyConverter.
• ISEFlow model weights (v1.1.0) for both AIS and GrIS on HuggingFace Hub.
• ise.__version__ attribute (read from package metadata via importlib.metadata).
• ise/py.typed marker for PEP 561 type-checking support.
• [tool.mypy], [tool.pytest.ini_options], [tool.coverage.run] sections in
  pyproject.toml.
• .pre-commit-config.yaml, .editorconfig, Makefile.
• CHANGELOG.md, CONTRIBUTING.md, CITATION.cff.
• GitHub Actions CI workflow (lint + test on Python 3.11 & 3.12), mypy CI job, and
  release workflow (OIDC trusted publisher).
• GitHub Actions issue templates.
• Slow and GPU pytest markers; skip_if_no_gpu helper in conftest.py.
• __all__ exports defined in all subpackage __init__.py files.
• Full test suite covering anomaly conversion, input dataclasses, all model components,
  scalers, loss functions, training utilities, and metrics (244 tests).
• Example scripts: example_ais.py, example_gris.py, example_absolute_forcings.py,
  process_training_data.py.
• wandb added as a core dependency (used in NF and LSTM training).

Removed

• ise/models/ISEFlow/ directory (old monolithic ISEFlow.py, de.py, nf.py).
• ise/models/predictors/ directory (lstm.py, deep_ensemble.py).
• ise/models/density_estimators/ directory (normalizing_flow.py).
• setup.py (replaced by pyproject.toml + Hatchling).
• variational_lstm_emulator.pt legacy weight file (unused).
• Stray ISEFlow_GrIS_v1-1-0 copy/ duplicate weights directory.
• Old example scripts ISEFlow_from_NC.py and ISEFlow_predict.py.

Full Changelog: https://github.com/Brown-SciML/ise/commits/v1.0.0