diff --git a/src/aix_page/models/baselines.py b/src/aix_page/models/baselines.py index 1b64073..d7fdfe5 100644 --- a/src/aix_page/models/baselines.py +++ b/src/aix_page/models/baselines.py @@ -25,12 +25,24 @@ def require_string_columns(frame: pd.DataFrame) -> list[str]: def records(frame: pd.DataFrame) -> list[dict[str, float]]: columns = require_string_columns(frame) + if not frame.columns.is_unique: + raise ValueError("Hashed-linear feature frames require unique column labels") + if frame.empty: + return [] rows: list[dict[str, float]] = [] history: list[str] = [column for column in columns if column.startswith("hist_")] - for _, row in frame.iterrows(): - item: dict[str, float] = {f"{c}={int(row[c])}": 1.0 for c in FEATURE_COLUMNS} - item["normalized_position"] = float(row["normalized_position"]) - item.update({c: float(row[c]) for c in history}) + feature_count = len(FEATURE_COLUMNS) + required_columns = [*FEATURE_COLUMNS, "normalized_position", *history] + for values in frame[required_columns].itertuples(index=False, name=None): + item: dict[str, float] = { + f"{column}={int(value)}": 1.0 + for column, value in zip(FEATURE_COLUMNS, values[:feature_count], strict=True) + } + item["normalized_position"] = float(values[feature_count]) + item.update( + (column, float(value)) + for column, value in zip(history, values[feature_count + 1 :], strict=True) + ) rows.append(item) return rows diff --git a/tests/test_feature_records.py b/tests/test_feature_records.py new file mode 100644 index 0000000..5bebf35 --- /dev/null +++ b/tests/test_feature_records.py @@ -0,0 +1,25 @@ +from __future__ import annotations + +import pandas as pd +import pytest + +from aix_page.models.baselines import FEATURE_COLUMNS, records + + +def test_feature_records_preserve_integer_ids_and_column_order(): + large_id = 2**53 + 1 + frame = pd.DataFrame({column: [large_id] for column in FEATURE_COLUMNS}) + frame["normalized_position"] = 0.25 + frame["hist_ctr"] = 0.125 + expected = {f"{column}={large_id}": 1.0 for column in FEATURE_COLUMNS} + expected.update(normalized_position=0.25, hist_ctr=0.125) + assert records(frame[list(reversed(frame.columns))]) == [expected] + + +def test_empty_feature_records(): + assert records(pd.DataFrame()) == [] + + +def test_duplicate_feature_columns_are_rejected(): + with pytest.raises(ValueError, match="unique column"): + records(pd.DataFrame([[1, 2]], columns=["Position", "Position"]))