Is a c++ std::priority_queue the right structure for the dijkstra queue

Question

I don't think a c++ priority queue is the right structure for the dijkstra queue, because it contains no functionality for an easy lookup or deletion of elements.

The right structure would be a fibonacci heap, but there is none in the std library.

Does anyone have suggestions for a better, c++ implemented structure?

There is an implementation of a Fibonacci heap in [Boost](http://www.boost.org/doc/libs/1_55_0/doc/html/boost/heap/fibonacci_heap.html). — D Drmmr, Jun 28 '14 at 09:52
I totally recommend the Boost Fibonacci heap or the D_ary heap with D=2 (binary heap). Actually, this last use to be the faster when the grid is not big enough (which is in most applications) — Javi, Jun 28 '14 at 10:57

score 0 · Answer 1 · answered Jun 28 '14 at 10:38

0

You can use std::set and store a pair<distance, vertex> in it. To lookup and delete elements, you can keep distance to each vertex in an array or in std::vector to obtain a pair<distance, vertex> for a given vertex quickly. The closest unvisited vertex is always in the first element of the set(and can obtained using set.begin()).

answered Jun 28 '14 at 10:38

kraskevich

18,368
4
33
45

wouldn't be a hashmap better than a set in this case? – thi gg Jun 28 '14 at 12:07
It is possible to find and delete minimum element in a set in O(log N) time but is not possible to do it in hashset or hashmap. – kraskevich Jun 28 '14 at 12:24

score 0 · Answer 2 · answered Jun 29 '14 at 19:28

For most practical purposes, the std::priority_queue based implementation is good enough for sparse graphs. Implementing Dijkstra this way has a runtime of O(E log V). If you have a dense enough graph, you could simply use the basic O(V*V) version of Dijkstra's algorithm. As the graph gets denser, the asymptotics of the Fib-heap version move closer to the vanilla implementation.

Is a c++ std::priority_queue the right structure for the dijkstra queue

2 Answers2