DisSModel v0.6.2

Fixed

• SyncRasterModel.synchronize(): state variables loaded as temporal
  (time, y, x) arrays (e.g. when CubeClient.to_lucc_data() finds
  temporal catalog entries) now correctly use the first slice as the
  initial _past snapshot instead of copying the full 3D array, which
  caused shift2d to crash with ValueError: too many values to unpack.
• _write_geotiff: casting a float array containing NaN to an integer
  dtype (e.g. int16) now fills NaN/inf with the band's nodata value
  before the cast, eliminating a RuntimeWarning: invalid value encountered in cast.

DisSModel v0.6.1

Fixed

• GeoTIFF write with mixed-dtype band_spec (e.g. int32 categorical +
  float32 continuous) silently truncated float bands to the first band's
  dtype. Bands are now promoted to the common NumPy result type
  (np.result_type) before writing.

Added

• Test suites for dissmodel.io (utils, dispatch, storage, raster, vector,
  convert, xarray) and dissmodel.visualization (chart, map, env detection).
  Coverage: 55% → 79% (319 → 441 tests).

Notes

• GeoTIFFs containing mixed-dtype bands saved with v0.6.0 may have truncated
  float bands. Re-exporting those files is recommended.

DisSModel v0.6.0

This release lands three coordinated pieces of work: a breaking fix to the
scheduler's time semantics, a new temporal raster API that underpins the
DisSCube integration, and a leaner, more verifiable installation/CI setup.
It also closes out the housekeeping items raised ahead of the JOSS review.

⚠️ Breaking Changes

• Environment.run(): end_time is now inclusive (TerraME-style). The
  simulation executes every scheduled tick t with
  start_time <= t <= end_time; a model's own end_time is inclusive too.

  Migration: if your code used end_time=N_STEPS expecting N_STEPS
  executions starting at 0, change it to end_time=N_STEPS - 1.

✨ Added

• Temporal raster variables in RasterBackend: set(name, arr, time=coords) / get(name, time=t) (ceiling lookup via searchsorted,
  clamped at the boundaries), plus is_temporal, time_axis,
  temporal_band_names() / static_band_names(). from_xarray /
  to_xarray now round-trip temporal variables with their time coordinates
  preserved. This is the integration substrate for DisSCube's
  CubeClient.to_lucc_data().
• ExperimentRecord.period: optional (start, end) window recording
  which cube time slices fed a simulation run — provenance for
  temporally-aware experiments.

🐛 Fixed

• ExperimentRecord.created_at no longer uses the deprecated
  datetime.utcnow(); timestamps are now timezone-aware
  (+00:00 in ISO output).
• RasterBackend.from_xarray: fixed Affine transform reconstruction.
• Scheduler no longer stalls when models finish before the environment's
  end_time.

🔧 Changed

• Package version is now single-sourced from package metadata — no more
  drift between dissmodel.__version__, pyproject.toml, CITATION.cff,
  and the README citation.
• README ecosystem table now points to existing repositories and uses
  pip install "git+https://…" for satellite libraries not yet on PyPI.
• README/CITATION.cff citation aligned with the JOSS paper author list.
• Core dependencies are leaner: streamlit removed, ipywidgets moved to
  the optional viz extra (pip install dissmodel[viz]), and the
  numpy<2.0 pin relaxed to support NumPy 2.
• from dissmodel.geo.raster import RasterBackend and the equivalent
  vector shortcut now work as documented.

🏗 Internal

• mypy dissmodel --ignore-missing-imports is clean (0 errors).
• CI now runs across Python 3.10–3.12, with Ruff linting, doctest checks,
  and test coverage reporting.

---

Full Changelog: v0.5.0...v0.6.0

v0.5.0

Remove salabim dependency — lightweight pure-Python scheduler

DisSModel 0.5.0 replaces salabim with a minimal time-stepped scheduler
implemented entirely in Python, eliminating the salabim and greenlet
dependencies with zero changes to the public API.

Models continue to implement setup() / execute() as before. Two new
lifecycle hooks — pre_execute() and post_execute() — replace the salabim
process() loop for models that need per-step snapshotting (SyncRasterModel,
SyncSpatialModel).

v0.4.1

Merge pull request #125 from DisSModel/develop

paper update

v0.4.0 — Load as first-class lifecycle phase + Xarray interoperability

What's new in 0.4.0

Breaking change — ModelExecutor.run signature

run(self, record) → run(self, data, record)

data is the direct return value of load(), injected by the platform.
Executors that called self.load(record) inside run() must remove that call
and receive the loaded data as the first argument instead.

Executor lifecycle orchestration

• New dissmodel/executor/runner.py with execute_lifecycle — the canonical
  validate → load → run → save loop, shared by the CLI and the platform
  job_runner. Each phase is timed independently and written to
  record.metrics (time_load_sec now visible separately from time_run_sec).
• execute_lifecycle exported from dissmodel.executor.
• ExecutorTestHarness updated to call load → run(data, record) in sequence
  and to reject the old one-parameter run signature with a clear error message.

RasterBackend — Xarray interoperability

• RasterBackend.to_xarray(time=None) — converts to xr.Dataset with (y, x)
  dimensions, spatial coordinates from rasterio Affine transform, and CRS as
  spatial_ref (CF-1.8 / rioxarray convention).
• RasterBackend.from_xarray(ds) — builds a backend from an xr.Dataset or
  xr.DataArray, recovering transform and CRS when available.
• RasterBackend.__init__ now accepts transform and crs keyword arguments.

Tests

Full executor test suite added under tests/executor/:
test_runner, test_registry, test_cli, test_testing — with
_isolate_registry autouse fixture preventing cross-test registry pollution.

Removed

• make_raster_grid() alias (deprecated since 0.3.0)
• regular_grid() alias (deprecated since 0.2.1)

---

Migration guide for executor authors:

# Before (0.3.x)
def run(self, record: ExperimentRecord):
    data = self.load(record)
    ...

# After (0.4.0)
def run(self, data, record: ExperimentRecord):
    # data injected by execute_lifecycle — no I/O here
    ...

Full changelog: CHANGELOG.md

v0.3.0

What's Changed

Added

Sync models — automatic .past snapshot semantics (equivalent to TerraME's cs:synchronize()):

• SyncSpatialModel — vector substrate, manages <col>_past columns automatically
• SyncRasterModel — raster substrate, manages <n>_past arrays automatically
• Both expose a public synchronize() method

I/O — dissmodel.geo.raster.io

• shapefile_to_raster_backend — load any vector file (Shapefile, GeoJSON, GPKG, ZIP) directly into a RasterBackend, with automatic "mask" band
• save_raster_backend — write arrays to GeoTIFF without a band_spec
• load_geotiff / save_geotiff updated with .zip support

RasterMap — new visualization options: scheme (equal_interval, quantiles), legend, auto_mask, save_frames

Map — headless fallback (saves PNGs), new figsize, interval, save_frames parameters

Fixed

• RasterMap: multiple instances now maintain independent windows — plt.close("all") no longer closes other figures
• Map: AttributeError: 'Map' object has no attribute 'fig' with Streamlit
• Map: no longer raises RuntimeError when no display is available
• make_raster_grid() deprecated → raster_grid() (alias kept until v0.4.0)

Full changelog: CHANGELOG.md

v0.2.1

What's Changed

API improvements

• parse_idx now returns GridPos(row, col) — a namedtuple that eliminates the
  (col, row) vs (row, col) ambiguity. Tuple unpacking remains fully compatible,
  so existing code continues to work without changes.
• regular_grid() renamed to vector_grid() for symmetry with the raster substrate.
  The old name is kept as a DeprecationWarning alias and will be removed in v0.3.0.
• GridPos is now exported from dissmodel.geo.vector.regular_grid.

Bug fixes

• Fixed TypeError: parameter 'kwargs' not allowed in RasterModel.setup() and
  RasterCellularAutomaton.setup() caused by a salabim incompatibility with **kwargs.

Documentation

• API reference split into api/geo/vector.md and api/geo/raster.md.
• getting_started.md expanded with execution modes and substrate comparison table.
• visualization.md updated to cover RasterMap.
• core.md expanded with OOP modeling section and per-model time control example.
• CHANGELOG.md updated.

Upgrade notes

# before
from dissmodel.geo import regular_grid
gdf = regular_grid(dimension=(30, 30), resolution=1)

# after
from dissmodel.geo.vector.regular_grid import vector_grid
gdf = vector_grid(dimension=(30, 30), resolution=1)

# parse_idx
from dissmodel.geo.vector.regular_grid import parse_idx
pos = parse_idx("3-4")
pos.row  # 3
pos.col  # 4

Full changelog: [CHANGELOG.md](https://github.com/lambdageo/dissmodel/blob/main/CHANGELOG.md)

Add Raster/NumPy Backend for Cellular Automata Models

This release introduces a RasterBackend that allows cellular automata models in DisSModel to run using NumPy arrays instead of iterating over GeoDataFrame rows.

The existing vector backend (GeoPandas + libpysal) remains unchanged. The raster backend is an optional execution engine behind the same API.

0.1.5

• Description:

## What's new in v0.1.5

- Consolidated all dependencies in `pyproject.toml`
- NumPy docstrings and type annotations across all modules
- All CA and SysDyn models refactored to English
- Fixed PATTERN fill ID convention (`row-col`)
- Performance: `neighbor_values()` in all CA rules
- Fixed `Environment` salabim animation crash in headless environments
- Added CLI, Streamlit and Jupyter notebook examples for all models
- MkDocs documentation with mkdocstrings
- CI workflow with pytest and mypy