Above gifs: Different algorithms searching for a path. The red line shows the discovered path.
This repo compares different search-algorithms in random 2d-maze. The maze is generated by starting from empty square and then adding walls of random length to random locations.
The goal is to find a path from corner to corner. Possible moves are up, down, left or right. Distance or length is measured with manhattan distance which is the sum of vertical and horizontal lengths.
The compared algorithms are:
- A-star with a manhattan-distance heuristic which ignores walls
- BFS: breadth first search
- DFS: depth first search
Notes:
- BFS and A-star are guaranteed to find the shortest possible path if any path exists.
- in this case BFS is same as Dijkstra because the distance between all connected nodes is uniform
| finished | algo | seconds | n_explored | length_path | count |
|---|---|---|---|---|---|
| False | A-star | 0.0124813 | 2397.05 | nan | 429 |
| False | BFS | 0.00818549 | 2397.05 | nan | 429 |
| False | DFS | 0.00808539 | 2397.05 | nan | 429 |
| True | A-star | 0.0159034 | 2987.72 | 211.52 | 571 |
| True | BFS | 0.0244835 | 7104.2 | 211.52 | 571 |
| True | DFS | 0.0131783 | 3811.35 | 822.445 | 571 |
In this maze-setting DFS finds one path faster than the other two algorithms measured in wall clock time. However, that path is usually not the shortest. BFS and A-star always find the optimal path. A-star runs considerably faster than BFS both in wall clock time and in terms of explored nodes.
A-star explores nodes slightly slower than the other two algorithms because it has to maintain priority-queue to decide the next node to explore.
All results can be repoduced with runner.py.
Doesn't work in Windows and might require some dependencies.