PipelineData

h2ml.core.feature_store.PipelineData dataclass

Container that pairs a numpy array with its feature names.

Attributes:

Name	Type	Description
X	ndarray	Feature matrix as numpy array (n_samples, n_features).
feature_names	list[str]	Ordered list of feature names matching X columns.
y	ndarray	Target array (n_samples,).
y_true	Optional[ndarray]	Original-scale y when a y-transform was applied (optional).
y_transform	Optional[str]	Name of the y-transform applied, e.g. "log" or "sqrt" (optional).
coords	Optional[ndarray]	Spatial coordinates (n_samples, 2) — lat/lon or projected x/y (optional).
times	Optional[ndarray]	Sample dates as YYYY-MM-DD strings or numpy datetime64[D] (n_samples,). Used to build temporally-aware conformal calibration in LocalConformalCalibration. Optional — pipeline runs without it.

Example

store = PipelineData(X=X_arr, feature_names=df.columns.tolist(), y=y_arr) store.to_frame() # recover DataFrame locally when needed store.n_features # 10

n_features property

Number of feature columns in X.

n_samples property

Number of rows (observations) in X.

from_frame(df, y, y_true=None, y_transform=None, coords=None, times=None) classmethod

Build a PipelineData directly from a DataFrame.

Parameters:

Name	Type	Description	Default
df	DataFrame	Feature DataFrame; columns become feature_names.	required
y	ndarray	Target array (n_samples,).	required
y_true	Optional[ndarray]	Original-scale y before any transform (optional).	None
y_transform	Optional[str]	Name of the y-transform applied (optional).	None
coords	Optional[ndarray]	Spatial coordinates (n_samples, 2) — lat/lon or x/y (optional).	None
times	Optional[ndarray]	Sample dates as YYYY-MM-DD strings or datetime64[D] (optional).	None

select(features)

Return a new PipelineData with only the selected features.

Order follows the input list. coords and times are propagated unchanged.

to_frame()

Recover a DataFrame with correct column names.