Skip to content

CHANGELOG.md

Latest commit

 

History

History
126 lines (99 loc) · 5.72 KB

CHANGELOG.md

File metadata and controls

126 lines (99 loc) · 5.72 KB
Raw

Changelog

All notable changes to SIROM are documented here. The format follows Keep a Changelog, and this project adheres to Semantic Versioning.

0.4.4 — 2026-06-07

Documentation patch. No library code or API change.

Added

  • A "Release history" table in README.md linking each version to its GitHub release, its milestone (with PR count), and CHANGELOG.md.

0.4.3 — 2026-06-07

Tooling / process patch. No library code or API change.

Added

  • scripts/refresh_release_index.py — regenerates the cross-release "Changelog index" footer on every GitHub release (versions from gh release list, blurbs from this file); idempotent, with a --dry-run preview.
  • A "Cutting a release" section in CONTRIBUTING.md documenting the release flow end to end.

0.4.2 — 2026-06-07

Tooling / process patch. No library code or API change.

Added

  • This CHANGELOG.md, documenting every release.
  • A three-layer guard ensuring a version bump always ships its changelog entry: a CI changelog-guard job (the un-bypassable merge gate), a git pre-commit hook (.githooks/, enabled per clone with git config core.hooksPath .githooks), and a Claude Code PreToolUse hook (.claude/). See CONTRIBUTING.md.

0.4.1 — 2026-06-07

Documentation patch over 0.4.0. No code or API change.

Documentation

  • The robust-CVRP demo README records that it runs on SIROM v0.4.0 and clarifies that the v0.4.0 refactors don't change the POST /solve → GET /jobs/{id} contract the demo consumes.
  • CONTEXT.md gains the rest of the v0.4.0 vocabulary — the cluster tree (Cluster tree, build, replicate) and validation status (Status, has_errors).

0.4.0 — 2026-06-06

Architecture-deepening release: make modules honest and concentrate scattered logic. No behavioural change to the solver or the HTTP API contract.

Changed

  • Honest two-stage Solution contract — the solver produces an UnscoredSolution (decision vector only on optimal solves); the scoring stage adds feasibility_probability via an explicit score() transform, yielding a ScoredSolution. is_optimal() / feasibility() accessors replace the hand-rolled key-presence guards, and solve_status is typed int.
  • ClusterTree.build() now owns the KMeans / WCSS / selection / subdivision algorithm that previously lived in the orchestrator; the replicate gate no longer leaks. batch_solver shrinks ~70 lines.
  • One core-level has_errors gate (sirom/status_checks.py) replaces the any("[ERROR]" in s …) prose-match repeated at eight sites.
  • The API drops a brittle results slice and counts cluster nodes from the tree.

Added

  • CONTEXT.md — domain glossary (including the feasibility probability vs robustness naming split).
  • docs/adr/ — first Architecture Decision Records, capturing the decisions not to make: keep OptimizationProblem + Coefficient (0001), keep the portfolio and CVRP request builders separate (0002), and count cluster nodes from the tree rather than add a stage-tag seam (0003).

0.3.0 — 2026-06-05

Added

  • Robust CVRP demo over VRP-REP (demo/vrp/) — pick a real CVRP instance from VRP-REP and get a Pareto frontier of candidate routings trading distance against the probability the plan stays feasible. A "SIROM-as-a-service" demo: the backend builds a SolveRequest, POSTs it to the SIROM HTTP API, and polls the job — it never imports SIROM.
    • Two uncertainty modes (demand → capacity; travel time → a synthesized shift limit), toggled in the UI with α / customers / scenarios / shift sliders.
    • Every coordinate-based CVRP dataset on VRP-REP (Augerat A/B/P, Fisher F, Christofides CMT/Set M, Uchoa 2014 X-instances, Golden, Li, VeRoLog); instances listed lazily from each zip; large instances subsampled to the depot + K nearest customers.
    • Route-based robustness (Monte-Carlo over decoded routes), an SVG route map, and a recharts distance-vs-robustness frontier, all in Docker Compose.

0.2.0 — 2026-06-01

First substantial release of SIROM as a deployable, documented service.

Added

  • HTTP API — async job API (POST /solve → poll GET /jobs/{id}) with a self-documenting Swagger /docs and a ready-to-run /example.
  • Docker — multi-stage image + docker compose up, with a Redis-backed job store for multi-worker scale-out.
  • MILP support — integer variables, with solver auto-selection (GLOP for LPs, SCIP for MILPs) and overrides for CLP, PDLP, and commercial GUROBI/CPLEX/XPRESS.
  • A polished robust investment-portfolio web demo (demo/).

Changed

  • ~9–29× faster — GLOP auto-selection, vectorized quality measure and scenario generation, and threaded scenario solves (see benchmarks/).

Fixed

  • Corrected the constraint-slack used as a clustering feature (it previously computed A_i·x - b_i·sum(x) instead of A_i·x - b_i).

[0.1.0]

Initial implementation of the Simulation-based Robust Optimization Method (unreleased; superseded by 0.2.0). Implements the method from Butkeraites, de Salles Neto & Gendreau, Expert Systems with Applications 203:117337 (2022).