frsutils 0.1.1
frsutils 0.1.1 strengthens the scientific documentation, reproducibility
artifacts, release validation, and repository hygiene of the fuzzy-rough
computation library while preserving the package-root API introduced in 0.1.0.
Highlights
- ITFRS, VQRS, and OWAFRS through a shared approximation result contract.
- Package-root functions for similarity matrices, lower and upper
approximations, signed boundary regions, and positive-region scores. - Dense NumPy reference execution and exact blockwise execution.
- Optional, model-specific CuPy-backed execution paths with conservative claims.
- Reusable registries and builders for similarities, T-norms, implicators,
fuzzy quantifiers, and OWA weights. - A reproducible real-dataset reference study with resolved configuration,
environment metadata, machine-readable outputs, figures, and checksums. - Continuous integration, package-build validation, documentation deployment,
and manuscript-validation workflows. - Machine-readable installed-package validation for wheel and source
distributions from read-only package trees and working directories. - Reviewable JSON scientific reference data protected by provenance metadata,
SHA-256 manifests, schema checks, and read-only NumPy reconstruction. - Corrected scientific concept pages with traceable primary references.
Compatibility
- Python 3.10 or newer is required; continuous integration covers Python 3.10,
3.11, and 3.12. - Core dependencies are NumPy and scikit-learn.
- NumPy remains the stable default backend.
- CuPy is optional and is not installed with the core package.
- The CUDA 12 optional environment is constrained to CuPy 14.x.
- Recorded GPU evidence represents a validated configuration, not an exhaustive
compatibility or performance guarantee.
Reproducibility and validation
The repository includes automated checks for default and exhaustive tests,
branch-aware API and core coverage, citation metadata, documentation builds,
wheel and source-distribution contents, isolated installed-package behavior, and
dense/blockwise numerical equivalence. The reference-study snapshot records its
configuration, software environment, source revision, outputs, and checksums.
Citation
Machine-readable software citation metadata is distributed in CITATION.cff.
An immutable archive DOI, when present in that file, identifies the corresponding
executable software release.