API reference · grader/inspection.py · 176 lines

`grader.inspection`

Host-side reasoning over the probe payload.

Students do not use the solution's variable names, so grading works by finding objects that look like the expected result and reporting how sure we are (design.md §21 and §25).

Summary

Name	Kind	Description
`confidence_for_candidates()`	function
`dataframes()`	function
`scalars()`	function
`collections()`	function
`hidden_test()`	function
`match_column()`	function	Best column in `profile` matching any of `hints` (substring, case-insensitive).
`has_datelike_column()`	function	A date role can be satisfied by a column name hint or a datetime dtype.
`numeric_columns()`	function
`zips_in_profile()`	function	All ZIP-looking values visible in a dataframe profile.
`zips_in_collection()`	function
`find_dataframe_candidates()`	function	Return profiles satisfying `predicate`, best first.
`find_scalar_close_to()`	function	Scalars in the namespace within `tolerance` of the expected answer.
`describe_dataframe()`	function	Compact, TA-readable description used as rubric evidence (design.md §12).

Constants

Name	Value	Notes
`CONFIDENCE_UNIQUE`	`0.98`	Confidence ladder from design.md §25.
`CONFIDENCE_TWO`	`0.75`
`CONFIDENCE_MANY`	`0.4`
`CONFIDENCE_NONE`	`0.0`

Functions

confidence_for_candidates(count: int) -> float #[source]: No description.

dataframes(probe: dict[str, Any] | None) -> list[dict[str, Any]] #[source]: No description.

scalars(probe: dict[str, Any] | None) -> list[dict[str, Any]] #[source]: No description.

collections(probe: dict[str, Any] | None) -> list[dict[str, Any]] #[source]: No description.

hidden_test(probe: dict[str, Any] | None, test_id: str) -> dict[str, Any] | None #[source]: No description.

match_column(profile: dict[str, Any], hints: Iterable[str]) -> str | None #[source]: Best column in profile matching any of hints (substring, case-insensitive).

has_datelike_column(profile: dict[str, Any], hints: Iterable[str]) -> str | None #[source]: A date role can be satisfied by a column name hint or a datetime dtype.

numeric_columns(profile: dict[str, Any]) -> list[str] #[source]: No description.

zips_in_profile(profile: dict[str, Any], id_hints: Iterable[str] = ()) -> set[str] #[source]: All ZIP-looking values visible in a dataframe profile.

zips_in_collection(entry: dict[str, Any]) -> set[str] #[source]: No description.

find_dataframe_candidates(probe: dict[str, Any] | None, predicate: Callable[[dict[str, Any]], bool], score: Callable[[dict[str, Any]], float] | None = None) -> list[dict[str, Any]] #[source]

Return profiles satisfying predicate, best first.

This is the generic form of design.md §21's find_dataframe_with_columns: matching may use columns, row counts, value ranges or expected geography.

find_scalar_close_to(probe: dict[str, Any] | None, expected: float, tolerance: float) -> list[dict[str, Any]] #[source]: Scalars in the namespace within tolerance of the expected answer.

describe_dataframe(profile: dict[str, Any], max_columns: int = 12) -> dict[str, Any] #[source]: Compact, TA-readable description used as rubric evidence (design.md §12).