PREPROCESSING

Select metadata columns and prepare viable training inputs

Syntax

from instancespace.stages.preprocessing import PreprocessingStage

outputs = space.run_until_stage(PreprocessingStage)

Description

PREPROCESSING selects requested features and algorithms, filters missing input data and records retained feature order. Metadata column roles and names are validated before subsequent scientific stages run.

Names in selvars.feats and selvars.algos can include their CSV prefixes. Selections are case-insensitive. If some names match, unknown names warn and are ignored; a request with no matches raises an error. At least three input features and one training algorithm must remain viable.

Parameters

Inputs include feature and performance matrices, instance/feature/algorithm labels, optional source labels, selvars settings and the composed PRELIM configuration. All rows and labels must align.

Returns

Raw and working feature/performance matrices, filtered labels, sources and a retained-feature record. Full typed contracts are in PreprocessingInput and PreprocessingOutput.

Examples

Prepare data as in Getting Started. To restrict the analysis before building:

from dataclasses import replace
from instancespace.data.options import SelvarsOptions

selected_options = replace(options, selvars=SelvarsOptions.default(
    feats=["size", "shape", "density"], algos=["left", "right"],
))
selected = InstanceSpace(metadata, selected_options)
outputs = selected.run_until_stage(PreprocessingStage)