slice 13: to_rows(properties) returns pd.DataFrame

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

packages/domain/pyproject.toml

@@ -3,7 +3,10 @@ name = "domna-domain"
 version = "0.1.0"
 description = "Shared domain types for the Ara modelling pipeline and sibling Domna services."
 requires-python = ">=3.11"
-dependencies = []
+dependencies = [
+    "pandas>=2.0",
+    "pandas-stubs",
+]
 
 [build-system]
 requires = ["hatchling"]

packages/domain/src/domain/ml/tests/test_transform.py

@@ -1,5 +1,6 @@
 """Tests for EpcMlTransform v0.1.0 — schema-contract surface and target extraction."""
 
+import pandas as pd
 import pytest
 
 from datatypes.epc.domain.epc_property_data import SapRoomInRoof, WindowTransmissionDetails
@@ -1087,6 +1088,66 @@ def test_to_row_extracts_ventilation_features() -> None:
     assert row["pressure_test"] == 4
 
 
+def test_to_rows_returns_dataframe_with_one_row_per_property() -> None:
+    # Arrange — two properties with different floor areas + SAP scores
+    epcs = [
+        make_minimal_sap10_epc(energy_rating_current=82, total_floor_area_m2=70.0),
+        make_minimal_sap10_epc(energy_rating_current=45, total_floor_area_m2=120.0),
+    ]
+    transform = EpcMlTransform()
+
+    # Act
+    df = transform.to_rows(epcs)
+
+    # Assert
+    assert isinstance(df, pd.DataFrame)
+    assert len(df) == 2
+    assert df.loc[0, "sap_score"] == 82
+    assert df.loc[1, "sap_score"] == 45
+    assert df.loc[0, "total_floor_area_m2"] == 70.0
+    assert df.loc[1, "total_floor_area_m2"] == 120.0
+
+
+def test_to_rows_returns_empty_dataframe_for_empty_input() -> None:
+    # Arrange
+    transform = EpcMlTransform()
+
+    # Act
+    df = transform.to_rows([])
+
+    # Assert
+    assert isinstance(df, pd.DataFrame)
+    assert len(df) == 0
+    # Every advertised column appears as an output column even for empty input.
+    schema = transform.schema()
+    for name in schema.feature_columns:
+        assert name in df.columns
+    for name in schema.target_columns:
+        assert name in df.columns
+
+
+def test_to_rows_casts_categorical_columns_to_pd_categorical_dtype() -> None:
+    # Arrange — minimal property with a categorical feature populated
+    epcs = [
+        make_minimal_sap10_epc(
+            energy_rating_current=82, dwelling_type="Mid-terrace house"
+        ),
+        make_minimal_sap10_epc(
+            energy_rating_current=45, dwelling_type="Detached house"
+        ),
+    ]
+    transform = EpcMlTransform()
+
+    # Act
+    df = transform.to_rows(epcs)
+
+    # Assert — every column flagged ColumnSpec.categorical=True is a pd.Categorical
+    schema = transform.schema()
+    for name, spec in schema.feature_columns.items():
+        if spec.categorical:
+            assert isinstance(df[name].dtype, pd.CategoricalDtype), name
+
+
 def test_to_row_area_weights_window_u_value_and_solar_transmittance() -> None:
     # Arrange — two windows with transmission details; one without.
     sap_windows = [

packages/domain/src/domain/ml/transform.py

@@ -10,7 +10,9 @@ are added in subsequent slices.
 See docs/adr/0007-kwh-as-ml-target.md for the target set and rationale.
 """
 
-from typing import Any, Optional
+from typing import Any, Iterable, Optional
+
+import pandas as pd
 
 from datatypes.epc.domain.epc import Epc
 from datatypes.epc.domain.epc_property_data import (
@@ -502,6 +504,28 @@ class EpcMlTransform:
             target_columns=dict(_TARGET_COLUMNS),
         )
 
+    def to_rows(self, properties: Iterable[EpcPropertyData]) -> pd.DataFrame:
+        """Apply `to_row` across many properties and return a typed DataFrame.
+
+        Columns flagged `categorical=True` in the schema are cast to
+        `pd.Categorical`; everything else is left at pandas-inferred dtype.
+        The DataFrame always carries every advertised column, even when the
+        input is empty.
+        """
+        schema = self.schema()
+        all_columns = list(schema.feature_columns.keys()) + list(
+            schema.target_columns.keys()
+        )
+        rows = [self.to_row(epc) for epc in properties]
+        df = pd.DataFrame(rows, columns=all_columns)
+        for name, spec in schema.feature_columns.items():
+            if spec.categorical:
+                df[name] = df[name].astype("category")
+        for name, spec in schema.target_columns.items():
+            if spec.categorical:
+                df[name] = df[name].astype("category")
+        return df
+
     def to_row(self, epc: EpcPropertyData) -> dict[str, Any]:
         """Map an EpcPropertyData to a single row of features + targets.