CLOISTER

Estimate empirical full and correlation-constrained instance-space bounds

Syntax

from instancespace.stages.cloister import CloisterStage

outputs = space.run_until_stage(CloisterStage)

Description

CLOISTER constructs feature-range corners, removes combinations contradicted by strong significant feature correlations, projects the permitted corners with the fitted PILOT model and computes convex hulls. Large feature counts use the observed-data fallback specified by max_features.

PLS bounds use the fitted feature mean. Full-dimensional 3D hulls include surface triangulation; coplanar 3D boundaries use a flat triangulated polygon. Collinear inputs cannot define a 2D/3D hull and raise an actionable error.

Parameters

p_val and c_thres control correlation significance and strength. max_features bounds corner enumeration. hull_dims="all" uses the projected dimensionality; an integer selects a leading-coordinate hull. See CLOISTER options.

Returns

Full and correlation-constrained boundary coordinates plus derived mesh information for persistence, plotting and CSV output. The low-level stage retains its two-array return contract. See the CLOISTER API.

Examples

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

boundary_options = replace(options, cloister=CloisterOptions.default(
    hull_dims="all", max_features=16,
))