|
Download README.md from LKForge/maze-solver-benchmark: direct link, hf CLI and curl.
- Browser
- Download file 2.3 kB
-
https://huggingface.co/datasets/LKForge/maze-solver-benchmark/resolve/main/README.md
- Command line
-
hf download hf://datasets/LKForge/maze-solver-benchmark/README.md
-
curl -L -o README.md https://huggingface.co/datasets/LKForge/maze-solver-benchmark/resolve/main/README.md
2.3 kB
| license: mit | |
| pretty_name: "Maze Solver Benchmark (BFS shortest path by maze size)" | |
| language: | |
| - en | |
| tags: | |
| - maze | |
| - bfs | |
| - shortest-path | |
| - recursive-backtracker | |
| - pathfinding | |
| - benchmark | |
| task_categories: | |
| - other | |
| size_categories: | |
| - n<1K | |
| configs: | |
| - config_name: bfs_by_size | |
| data_files: data/maze-benchmark.json | |
| # Maze Solver Benchmark — BFS Shortest Path by Maze Size | |
| By **[LK Forge](https://lkforge.com)**: privacy-first browser tools & games, nothing to install, nothing uploaded. | |
| Search effort and path length for the [LK Forge Maze Solver](https://lkforge.com/tools/puzzles/maze-solver/), | |
| which generates a **perfect maze** with a recursive-backtracker and finds the **unique shortest | |
| path** with breadth-first search. 200 seeded mazes per size. | |
| - **How the solver works:** https://lkforge.com/tools/puzzles/maze-solver/how-the-maze-solver-works.html | |
| - **Try it:** https://lkforge.com/tools/puzzles/maze-solver/ | |
| - **Producer:** [LK Forge](https://lkforge.com) — client-side AI games, solvers, and tools. | |
| ## Headline numbers (200 mazes each) | |
| | Maze | Open cells | Mean shortest path | Mean cells visited by BFS | % of open cells explored | | |
| |---|---|---|---|---| | |
| | 12×12 | 287 | 131 | 185 | 64.5% | | |
| | 20×20 | 799 | 308 | 487 | 60.9% | | |
| | 30×30 | 1,799 | 607 | 1,071 | 59.5% | | |
| | 40×40 | 3,199 | 972 | 1,784 | 55.8% | | |
| In a perfect maze there is exactly one path between any two cells, so BFS returns the unique | |
| shortest route. It explores a **fairly stable ~56–65% of all open cells** before reaching the | |
| exit, and both path length and search effort scale roughly linearly with the number of open cells. | |
| ## Files | |
| - `data/maze-benchmark.json` — per size: open cells, mean shortest-path length, mean cells visited, % explored. | |
| ## Method | |
| - **Engine:** the recursive-backtracker generator and BFS solver copied verbatim from the shipped | |
| `maze-solver.js` (DOM stripped, `Math.random` → seeded RNG, size parametrized). | |
| - **Cells visited** = BFS dequeue count (machine-independent search effort). | |
| ## Reproduce | |
| ```bash | |
| node scripts/benchmark-maze.mjs | |
| ``` | |
| ## Citation | |
| > LK Forge (2026). *Maze Solver Benchmark — BFS Shortest Path by Maze Size.* | |
| > https://lkforge.com/tools/puzzles/maze-solver/ | |
| Everything runs client-side, no login, no tracking. | |