Skip to content

RELEASE_NOTES.md

Latest commit

 

History

History
100 lines (65 loc) · 5.7 KB

RELEASE_NOTES.md

File metadata and controls

100 lines (65 loc) · 5.7 KB
Raw

Znum 4.3.0

Configurable comparison weighting + Znum.config()

  • Znum.config() context manager — Unified configuration for Z-number computation. Replaces Znum.fast() as the primary configuration entry point. Accepts fast_triangle, min_b, and the new sort_a_weight parameter.
  • sort_a_weight parameter — Controls how much the A (restriction/value) component weighs against B (reliability) in Sort.solver_main comparisons. Default 0.5 preserves the original equal weighting. Higher values (e.g., 0.7) make the actual value dominate, preventing B mismatches from masking real value differences in constraint checks.
  • Znum.fast() preserved — Now a convenience alias for Znum.config(fast_triangle=True, ...). Fully backward compatible.

Motivation

When comparing Z-numbers with mismatched B values (e.g., a summed total with degraded B vs a crisp limit with B=1.0), the original equal weighting allowed B differences to override clear A differences. A value 33 minutes over a time budget could be ranked as "smaller" simply because its B was lower. sort_a_weight=0.7 fixes this by weighting A at 70%, ensuring that actual value dominance is preserved while B still contributes to the comparison.

Example

with Znum.config(fast_triangle=True, min_b=True, sort_a_weight=0.7):
    result = z1 + z2        # fast analytical + min_b
    is_over = total > limit  # A weighted 70%, B weighted 30%

Znum 4.2.0

min_b option for Znum.fast()

  • min_b parameter on Znum.fast() — When min_b=True, arithmetic operations use min(B1, B2) element-wise instead of convolution/LP-based B computation. Prevents B reliability degradation through repeated operations (Aliev et al. 2017).

Znum 4.0.0

Li et al. analytical engine

  • Znum.fast() now uses Li et al. 2023 analytical method — For triangular Z-numbers, B computation uses extended triangular distribution convolutions instead of LP. ~5x faster, deterministic, and produces narrower B spread than LP for high-B inputs. Non-triangular Z-numbers fall back to LP automatically.
  • New module tri_math.py — Implements the full analytical pipeline: extended triangular PDF, convolution, and closed-form B computation.
  • New properties — is_triangular, A_tri, B_tri for detecting and decomposing triangular Z-numbers.
  • Removed old fast_b mode — The v3 B = min(B1, B2) heuristic is replaced by the analytically grounded Li et al. method. test_fast_b.py deleted, replaced by test_tri_analytical.py (68 tests).

Reference

Li, Y. et al. (2023). The arithmetic of triangular Z-numbers with reduced calculation complexity using an extension of triangular distribution. Information Sciences, 647, 119477. doi:10.1016/j.ins.2023.119477

Znum 3.0.0

Refactored arithmetic pipeline

  • Separated A and B computation — math_ops.py rewritten with clear phases: _compute_a_pairs, _merge_rows, _extract_trapezoid for A; _compute_b_columns, _compute_prob_pos, _compute_result_B_lp for B. Old internal methods (get_matrix_main, get_minimized_matrix, get_Q_from_matrix, get_prob_pos) removed.
  • Added Znum.fast() context manager — Thread-safe opt-in for fast B computation via threading.local(). Supports nesting and exception safety.

Znum 2.1.0

Performance

  • Replaced scipy with highspy — Calls the HiGHS LP solver directly instead of going through scipy's Python wrapper. Same solver, same results, ~3x faster arithmetic operations. Dependency shrinks from 46MB (scipy) to 2.5MB (highspy).
  • LP model reuse — Builds the HiGHS model once per get_matrix call and re-solves by changing only the RHS, avoiding redundant model construction.

Dependency change

- scipy>=1.10.0
+ highspy>=1.7.0

Znum 1.0.0

Breaking Changes

  • Beast renamed to MCDMUtils — All references to Beast must be updated.
  • shouldNormalizeWeight renamed to normalize_weights — Affects Topsis and Promethee constructors.
  • distanceType renamed to distance_type — Affects Topsis constructor.
  • Removed Znum.Topsis, Znum.Sort, Znum.Promethee, Znum.Beast, Znum.Math, Znum.Dist — Import these directly from znum instead of accessing them as class attributes.
  • Removed Vikor module — It had incorrect return values and no test coverage.
  • Removed IncompatibleABPartsException — It was never raised anywhere.
  • Removed Type class (znum/ztype.py) — Shape properties (is_trapezoid, is_triangle, is_even) are now directly on Znum.

Bug Fixes

  • Fixed Topsis.ordered_indices sorting in wrong direction — Was sorting worst-first instead of best-first. The core solver_main() was unaffected.
  • Fixed division-by-zero risks in sort.py, topsis.py, and utils.py with proper guards.
  • Fixed silent B-value mutation — Now emits a UserWarning when near-zero B values are adjusted to avoid degenerate LP solutions.
  • Fixed decorator wrappers — Added @functools.wraps and switched to isinstance() checks in validators.

Performance

  • Constant Z-number LP shortcut — Ideal Z-numbers (used in Hellinger distance) now solve 1 LP instead of 10, reducing TOPSIS Hellinger runtime by ~27%.
  • Promethee symmetry exploitation — Preference table computation now uses upper-triangle iteration, cutting Sort comparisons in half (~22% faster).

Improvements

  • Full type hints on all public APIs.
  • Docstrings on all public classes and methods.
  • Consistent snake_case naming throughout.
  • Magic numbers replaced with named constants.
  • Promethee API aligned with Topsis (added result property and guards on all accessor properties).