Fpa fod wildfire

docs/source/datasets/fpa_fod_tabular.rst

@@ -0,0 +1,38 @@
+FPA FOD — Tabular Classification Dataset
+========================================
+
+What it is
+----------
+
+This dataset provides a tabular supervised-learning view derived from FPA FOD wildfire records.
+One sample corresponds to one incident (or one processed record).
+
+Data source and licensing
+-------------------------
+
+The raw FPA FOD dataset is large and must be obtained by the user. PyHazards does not ship the raw data
+due to licensing/size constraints.
+
+How to provide data
+-------------------
+
+1. Obtain the FPA FOD sqlite (user-provided).
+2. Place it at the path expected by the dataset builder (see project README/scripts).
+3. Run the preprocess/build step to produce processed dataset artifacts.
+
+Returned tensors (contract)
+---------------------------
+
+A single sample returns:
+- ``x``: float tensor of shape ``[input_dim]``
+- ``y``: integer class id (scalar)
+
+A batch returns:
+- ``x``: ``[B, input_dim]``
+- ``y``: ``[B]``
+
+Micro dataset for CI
+--------------------
+
+For CI/testing: set ``micro=True`` to use deterministic synthetic data that matches the schema and shapes,
+so tests run without requiring the raw FPA FOD sqlite.

docs/source/datasets/fpa_fod_weekly.rst

@@ -0,0 +1,31 @@
+FPA FOD — Weekly Forecasting Dataset
+====================================
+
+What it is
+----------
+
+This dataset provides weekly sequences for forecasting wildfire activity.
+One sample corresponds to a sequence of weekly feature vectors.
+
+Data source and licensing
+-------------------------
+
+The raw FPA FOD dataset must be obtained by the user. PyHazards does not ship the raw data due to
+licensing/size constraints.
+
+Returned tensors (contract)
+---------------------------
+
+A single sample returns:
+- ``x``: float tensor of shape ``[T, input_dim]``
+- ``y``: float tensor of shape ``[out_dim]`` (or scalar)
+
+A batch returns:
+- ``x``: ``[B, T, input_dim]``
+- ``y``: ``[B, out_dim]``
+
+Micro dataset for CI
+--------------------
+
+For CI/testing: set ``micro=True`` to use deterministic synthetic data (small B and T) to validate shapes,
+dtypes, and a trainer smoke run, without requiring the raw FPA FOD sqlite.

pyhazards/datasets/__init__.py

@@ -14,3 +14,15 @@
     "GraphTemporalDataset",
     "graph_collate",
 ]
+from .wildfire_fpa_fod import FPAFODWildfireTabular, FPAFODWildfireWeekly
+
+# Register datasets so framework load_dataset(...) can find them
+register_dataset(
+    name="wildfire_fpa_fod_tabular",
+    builder=lambda **kwargs: FPAFODWildfireTabular(**kwargs),
+)
+
+register_dataset(
+    name="wildfire_fpa_fod_weekly",
+    builder=lambda **kwargs: FPAFODWildfireWeekly(**kwargs),
+)