Inferencer

mllabs._inferencer.Inferencer

Applies trained processors to new data for inference.

Created by :meth:~mllabs._trainer.Trainer.to_inferencer. Self-contained — no dependency on Experimenter or Trainer at serve time.

Attributes:

Name	Type	Description
pipeline	Pipeline	Minimal pipeline (selected nodes only).
selected_stages	list[str]	Stage node names.
selected_heads	list[str]	Head node names.
n_splits	int	Number of cross-validation splits.
node_objs	dict	{node_name: [processor_split0, ...]}.
v		Output column filter applied to Head outputs.

process(data, agg='mean')

Run inference on new data and aggregate across splits.

Parameters:

Name	Type	Description	Default
data		Input dataset (pandas/polars DataFrame or numpy array).	required
agg	str | callable | None	Aggregation strategy across splits. 'mean' (default), 'mode', a callable receiving a list of per-split DataFrames, or None (returns list). Ignored when n_splits == 1.	'mean'

Returns:

Type	Description
	DataFrame | list: Aggregated predictions, or a list of per-split
	predictions when agg=None.

save(path)

Serialize the Inferencer to a single file.

Parameters:

Name	Type	Description	Default
path	str | Path	Directory to save into. Creates {path}/__inferencer.pkl.	required

load(path) classmethod

Load a saved Inferencer from disk.

Parameters:

Name	Type	Description	Default
path	str | Path	Directory containing __inferencer.pkl.	required

Returns:

Name	Type	Description
Inferencer		Restored inferencer.