| commit | 6bca319df4bae83c90beac09e6ad26b1705d0566 | |
| author | Eric Blake <eblake@redhat.com> | |
| Thu, 1 Feb 2024 13:19:57 +0000 (1 07:19 -0600) | ||
| committer | Eric Blake <eblake@redhat.com> | |
| Thu, 1 Feb 2024 15:12:02 +0000 (1 09:12 -0600) | ||
| tree | c80fe89cd78d9073933f34c6a74c849aa0f5ed48 | treesnapshot (tar.gz zip) |
| parent | f30bc45268a74930d07cb9d752bea56932cb782f | commitdiff |
day 17 fix bug
The heuristic for A* had two bugs: it was not properly lowering the
cost, and it did not account for the fact that the cost for the final
node was hacked into the visit of the first node. These two things
together meant that the heuristic was not admissible, leading to some
puzzles being off by up to the weight of the goal node (even my own
input was off by 2, and I failed to notice that); the Dijkstra
algorithm, while slower, did not have the bug.
The heuristic for A* had two bugs: it was not properly lowering the
cost, and it did not account for the fact that the cost for the final
node was hacked into the visit of the first node. These two things
together meant that the heuristic was not admissible, leading to some
puzzles being off by up to the weight of the goal node (even my own
input was off by 2, and I failed to notice that); the Dijkstra
algorithm, while slower, did not have the bug.
| 2023/day15.eli5m4 | diffblobblamehistory | |
| 2023/day17.m4 | diffblobblamehistory |