Orginated from my research "A_r*: a Bounded Suboptimal Search algorithm using Conditional Node Re-expansion Policy"

A clean, research-ready package for comparing Weighted A* variants:

• CR: Weighted A* with Conditional Re-expansion (threshold r, multiple modes).
• NR: Never Re-open.
• AR: Always Re-open.
• ARA*: Anytime Repairing A* (baseline with reuse).

Includes diverse scenarios (grids, inconsistent heuristics, geometric graphs, 8-puzzle, terrain, mazes), a benchmark suite, and optional plotting.

pip install -e .[dev,plot]
pre-commit install

# Compare on a wall+gap grid (CSV to stdout)
reexpastar-compare --width 60 --height 60 --weight 1.5 --r_cr 0.2

# Ablation of r (NR/AR/CR)
reexpastar-ablate --width 80 --height 80 --weight 1.5 --r_values "inf,1.0,0.5,0.2,0.0" --r_mode rel_edge

# Full benchmark suite (writes CSV -> results/)
reexpastar-benchmark --weight 1.5 --r_values "inf,0.5,0.2,0.0" --r_mode abs --seed 42 --include_ara
# Then produce annotated scatter plots for each scenario
reexpastar-plot results/benchmark_YYYYMMDD_HHMMSS.csv

PYTHONPATH=. python -m unittest discover -s tests -t .
# or with pytest if you prefer
# pytest -q --cov=reexpastar

reexpastar/
  src/reexpastar/        # package code (CR/NR/AR/ARA*, grids, terrain, scenarios)
  tests/                 # unit tests
  examples/              # runnable samples
  docs/                  # MkDocs (optional)
  .github/workflows/     # CI config

See CONTRIBUTING.md. By participating you agree to the Code of Conduct.

See CITATION.cff.

• Best Paper, KICS – ICTC 2023 (selected from ~600 papers). This repository accompanies my research on re-expansion strategies in heuristic search.

• Generics & Protocols for type-safe state definitions (reexpastar.core.types).
• Tie-breaking: tie_break='g_low'|'g_high' to control OPEN ordering when f ties.
• Guards & telemetry: max_expansions, max_runtime_ms, log_every with reexpastar.logging.get_logger() (JSON-capable).
• Parallel benchmarks: reexpastar.cli.benchmark_parallel to run multiple seeds in parallel.
• Profiling: reexpastar.cli.profile_search to get cProfile snapshots on heavy instances.