aip_sdk.from_dataframe

aip_sdk.from_dataframe(df: pd.DataFrame, schema: str, project: Project | None = None, mapping: dict[str, str] | None = None, task_type: str | None = None) -> pd.DataFrame

Map and validate a raw DataFrame to a GDI schema.

This function:

1. Applies column mapping (renames columns)
2. Injects a constant task_type column when one is provided and absent
3. Validates against the GDI schema using Pandera
4. Attaches the .aip accessor with schema metadata

Parameters

• df pd.DataFrame: Raw DataFrame
• schema str: GDI schema name (e.g., 'gdi_text_v1')
• project Project | None: Optional project for dimension metadata
• mapping dict[str, str] | None: Column mapping: {raw_column: gdi_column}
• task_type str | None: Task discriminator for task-scoped schemas (e.g. 'single_turn_llm' for gdi_text_v1). Injected as a constant column when the DataFrame does not already carry one.

Returns

• pd.DataFrame: Validated DataFrame with .aip accessor attached

Raises

• ValidationError: If required columns missing or validation fails

Examples

>>> df = pd.DataFrame(
...     {
...         "input_id": ["q1"],
...         "question": ["What is AI?"],
...         "answer": ["Artificial Intelligence"],
...         "ref": ["AI is..."],
...     }
... )
>>> mapped_df = from_dataframe(
...     df,
...     schema="gdi_text_v1",
...     task_type="single_turn_llm",
...     mapping={"question": "prompt", "answer": "sut_response", "ref": "expected_output"},
... )