Graph Algorithms

Graphs.jl implements a collection of classic graph algorithms:

  • graph traversal with visitor support: BFS, DFS, MAS
  • cycle detection
  • connected components
  • topological sorting
  • shortest paths: Dijkstra, Floyd-Warshall, A*
  • minimum spanning trees: Prim, Kruskal
  • flow: Minimum Cut
  • random graph generation
  • more algorithms are being implemented

Graph Traversal

Graph traversal refers to a process that traverses vertices of a graph following certain order (starting from user-input sources). This package implements three traversal schemes: breadth-first, depth-first, and Maximum-Adjacency.

During traveral, each vertex maintains a status (also called color), which is an integer value defined as below:

  • color = 0: the vertex has not been encountered (i.e. discovered)
  • color = 1: the vertex has been discovered and remains open
  • color = 2: the vertex has been closed (i.e. all its neighbors have been examined)
traverse_graph(graph, alg, source, visitor[, colormap])
Parameters:
  • graph – The input graph, which must implement vertex_map and adjacency_list.
  • alg – The algorithm of traveral, which can be either BreadthFirst() or DepthFirst().
  • source – The source vertex (or vertices). The traversal starts from here.
  • visitor – The visitor which performs certain operations along the traversal.
  • colormap – An integer vector that indicates the status of each vertex. If this is input by the user, the status will be written to the input vector, otherwise an internal color vector will be created.

Here, visitor must be an instance of a sub-type of AbstractGraphVisitor. A specific graph visitor type can choose to implement some or all of the following methods.

discover_vertex!(visitor, v)

invoked when a vertex v is encountered for the first time. This function should return whether to continue traversal.

open_vertex!(visitor, v)

invoked when a vertex v is about to examine v‘s neighbors.

examine_neighbor!(visitor, u, v, color, ecolor)

invoked when a neighbor/out-going edge is examined. Here color is the status of v, and ecolor is the status of the outgoing edge. Edge statuses are currently only considered by depth-first search.

close_vertex!(visitor, v)

invoked when all neighbors of v has been examined.

If a method of these is not implemented, it will automatically fallback to no-op. The package provides some pre-defined visitor types:

  • TrivialGraphVisitor: all methods are no-op.
  • VertexListVisitor: it has a field vertices, which is a vector comprised of vertices in the order of being discovered.
  • LogGraphVisitor: it prints message to show the progress of the traversal.

Many graph algorithms can be implemented based on graph traversal through certain visitors or by using the colormap in certain ways. For example, in this package, topological sorting, connected components, and cycle detection are all implemented using traverse_graph with specifically designed visitors.

Cycle detection

In graph theory, a cycle is defined to be a path that starts from some vertex v and ends up at v.

test_cyclic_by_dfs(g)

Tests whether a graph contains a cycle through depth-first search. It returns true when it finds a cycle, otherwise false. Here, g must implement vertex_list, vertex_map, and adjacency_list.

Connected components

In graph theory, a connected component (in an undirected graph) refers to a subset of vertices such that there exists a path between any pair of them.

connected_components(g)

Returns a vector of components, where each component is represented by a vector of vertices. Here, g must be an undirected graph, and implement vertex_list, vertex_map, and adjacency_list.

Cliques

In graph theory, a clique in an undirected graph is a subset of its vertices such that every two vertices in the subset are connected by an edge. A maximal clique is the largest clique containing a given node.

maximal_cliques(g)

Returns a vector of maximal cliques, where each maximal clique is represented by a vector of vertices. Here, g must be an undirected graph, and implement vertex_list and adjacency_list.

Topological Sorting

Topological sorting of an acyclic directed graph is a linear ordering of vertices, such that for each directed edge (u, v), u always comes before v in the ordering.

topological_sort_by_dfs(g)

Returns a topological sorting of the vertices in g in the form of a vector of vertices. Here, g may be directed or undirected, and implement vertex_list, vertex_map, and adjacency_list.

Shortest Paths

This package implements three classic algorithms for finding shortest paths: Dijkstra’s algorithm, the Floyd-Warshall algorithm, and the A* algorithm. We plan to implement the Bellman-Ford algorithm and Johnson’s algorithm in the near future.

Dijkstra’s Algorithm

dijkstra_shortest_paths(graph, edge_dists, source[, visitor])

Performs Dijkstra’s algorithm to find shortest paths to all vertices from input sources.

Parameters:
  • graph – The input graph
  • edge_dists – The vector of edge distances or an edge property inspector.
  • source – The source vertex (or vertices)
  • visitor – An visitor instance
Returns:

An instance of DijkstraStates that encapsulates the results.

Here, graph can be directed or undirected. It must implement vertex_map, edge_map and incidence_list. edge_dists is optional; if not specified, default distances of 1 are used for each edge.

The following is an example that shows how to use this function:

# construct a graph and the edge distance vector

g = simple_inclist(5)

inputs = [       # each element is (u, v, dist)
    (1, 2, 10.),
    (1, 3, 5.),
    (2, 3, 2.),
    (3, 2, 3.),
    (2, 4, 1.),
    (3, 5, 2.),
    (4, 5, 4.),
    (5, 4, 6.),
    (5, 1, 7.),
    (3, 4, 9.) ]

ne = length(inputs)
dists = zeros(ne)

for i = 1 : ne
    a = inputs[i]
    add_edge!(g, a[1], a[2])   # add edge
    dists[i] = a[3]             # set distance
end

r = dijkstra_shortest_paths(g, dists, 1)

@assert r.parents == [1, 3, 1, 2, 3]
@assert r.dists == [0., 8., 5., 9., 7.]

The result has several fields, among which the following are most useful:

  • parents[i]: the parent vertex of the i-th vertex. The parent of each source vertex is itself.
  • hasparent[i]: true if the i-th vertex has a parent, and false otherwise. When hasparent[i] == false, it means that the vertex at index i isn’t reachable from any source. Note that hasparent[i] == true for all source vertices.
  • dists[i]: the minimum distance from the i-th vertex to source.

The user can (optionally) provide a visitor that perform operations along with the algorithm. The visitor must be an instance of a sub type of AbstractDijkstraVisitor, which may implement part of all of the following methods.

discover_vertex!(visitor, u, v, d)

Invoked when a new vertex v is first discovered (from the parent u). d is the initial distance from v to source.

include_vertex!(visitor, u, v, d)

Invoked when the distance of a vertex is determined (at the point v is popped from the heap). This function should return whether to continue the procedure. One can use a visitor to terminate the algorithm earlier by letting this function return false under certain conditions.

update_vertex!(visitor, u, v, d)

Invoked when the distance to a vertex is updated (relaxed).

close_vertex!(visitor, u, v, d)

Invoked when a vertex is closed (all its neighbors have been examined).

enumerate_paths(vertices, parent_indices[, dest])

Returns an array of vectors (containing vertices), whose i-th element corresponds to the path from a source to vertex dest[i]. Empty vectors indicate vertices that are unreachable from the source. dest can be a subset of indices, or left unspecified (in which case, all the indices will be considered). If dest is a single index, then the result is just an array of vertices, corresponding to the path from a source to dest.

enumerate_indices(parent_indices[, dest])

Returns an array of indices corresponding to the vertices returned by enumerate_paths(vertices, parent_indices[, dest])

The following is an example that shows how to use this function:

julia> g4 = Graphs.inclist([4,5,6,7],is_directed=true)
julia> add_edge!(g4,4,5); add_edge!(g4,4,6); add_edge!(g4,5,6); add_edge!(g4,6,7)
julia> s4 = dijkstra_shortest_paths(g4,5)
julia> sps = enumerate_indices(s4.parent_indices) # dest: all indices
4-element Array{Array{Int64,1},1}:
 []
 [2]
 [2,3]
 [2,3,4]

julia> enumerate_indices(s4.parent_indices, [2,4]) # dest: subset of indices
2-element Array{Array{Int64,1},1}:
 [2]
 [2,3,4]

julia> enumerate_indices(s4.parent_indices, 4) # dest: single index
3-element Array{Int64,1}:
 2
 3
 4

julia> enumerate_paths(vertices(g4), s4.parent_indices) # dest: all vertices
4-element Array{Array{Int64,1},1}:
 []
 [5]
 [5,6]
 [5,6,7]

julia> enumerate_paths(vertices(g4), s4.parent_indices, [2,4]) # dest: subset of vertices
2-element Array{Array{Int64,1},1}:
 [5]
 [5,6,7]

julia> enumerate_paths(vertices(g4), s4.parent_indices, 4) # dest: single vertex
3-element Array{Int64,1}:
 5
 6
 7

Remark: enumerate_paths and enumerate_indices are applicable to the results from both dijkstra_shortest_paths and bellman_ford_shortest_paths.

Bellman Ford Algorithm

bellman_ford_shortest_paths(graph, edge_dists, source)

Performs Bellman Ford algorithm to find shortest paths to all vertices from input sources.

Parameters:
  • graph – The input graph
  • edge_dists – The vector of edge distances or an edge property inspector.
  • source – The source vertex (or vertices)
Returns:

An instance of BellmanFordStates that encapsulates the results.

Here, graph can be directed or undirected. Weights can be negative for a directed graph. It must implement vertex_map, edge_map and incidence_list. If there is a negative weight cycle an exception of NegativeCycleError is thrown.

The result has several fields, among which the following are most useful:

  • parents[i]: the parent vertex of the i-th vertex. The parent of each source vertex is itself.
  • dists[i]: the minimum distance from the i-th vertex to source.
has_negative_edge_cycle(graph, edge_dists)
Tests if the graph has a negative weight cycle.
Parameters:
  • graph – The input graph
  • edge_dists – The vector of edge distances or an edge property inspector.
Returns:

true if there is a negative weight cycle, false otherwise.e

Floyd-Warshall’s algorithm

floyd_warshall(dists)

Performs Floyd-Warshall algorithm to compute shortest path lengths between each pair of vertices.

Parameters:dists – The edge distance matrix.
Returns:The matrix of shortest path lengths.
floyd_warshall!(dists)

Performs Floyd-Warshall algorithm inplace, updating an edge distance matrix into a matrix of shortest path lengths.

floyd_warshall!(dists, nexts)

Performs Floyd-Warshall algorithm inplace, and writes the next-hop matrix. When this function finishes, nexts[i,j] is the next hop of i along the shortest path from i to j. One can reconstruct the shortest path based on this matrix.

A*

shortest_path(graph, dists, s, t[, heuristic])

Find the shortest path between vertices s and t of graph using Hart, Nilsson and Raphael’s A* algorithm.

Parameters:
  • graph – the input graph
  • dists – the edge distance matrix or an edge property inspector
  • s – the start vertex
  • t – the end vertex
  • heuristic – a function underestimating the distance from its input node to t.
Returns:

an array of edges representing the shortest path.

Minimum Spanning Trees

This package implements two algorithm to find a minimum spanning tree of a graph: Prim’s algorithm and Kruskal’s algorithm.

Prim’s algorithm

Prim’s algorithm finds a minimum spanning tree by growing from a root vertex, adding one edge at each iteration.

prim_minimum_spantree(graph, eweights, root)

Perform Prim’s algorithm to find a minimum spanning tree.

Parameters:
  • graph – the input graph
  • eweights – the edge weights (a vector or an edge property inspector)
  • root – the root vertex
Returns:

(re, rw), where re is a vector of edges that constitute the resultant tree, and rw is the vector of corresponding edge weights.

Kruskal’s algorithm

Kruskal’s algorithm finds a minimum spanning tree (or forest) by gradually uniting disjoint trees.

kruskal_minimum_spantree(graph, eweights[, K=1])
Parameters:
  • graph – the input graph
  • eweights – the edge weights (a vector or an edge property inspector)
  • K – the number of trees in the resultant forest. If K = 1, it ends up with a tree. This argument is optional. By default, it is set to 1.
Returns:

(re, rw), where re is a vector of edges that constitute the resultant tree, and rw is the vector of corresponding edge weights.

Flow

This package implements Simple Minimum Cut

Simple Minimum Cut

Stoer’s simple minimum cut gets the minimum cut of an undirected graph.

min_cut(graph[, eweights])
Parameters:
  • graph – the input graph
  • eweights – the edge weights (a vector or an edge property inspector). This argument is optional. If not given edges are weight “1”
Returns:

(parity, bestcut), where parity is a vector of boolean values that determines the partition and bestcut is the weight of the cut that makes this partition.

Random Graphs

Erdős–Rényi graphs

The Erdős–Rényi model sets an edge between each pair of vertices with equal probability, independently of the other edges.

erdos_renyi_graph(g, n, p[; has_self_loops=false])

Add edges between vertices 1:n of graph g randomly, adding each possible edge with probability p independently of all others.

Parameters:
  • g – the input graph
  • n – the number of vertices between which to add edges
  • p – the probability with which to add each edge
  • has_self_loops – whether to consider edges v -> v.
Returns:

the graph g.

erdos_renyi_graph(n, p[, has_self_loops=false])

Convenience function to construct an n-vertex Erdős–Rényi graph as an incidence list.

Watts-Strogatz graphs

The Watts–Strogatz model is a random graph generation model that produces graphs with small-world properties, including short average path lengths and high clustering.

watts_strogatz_graph(g, n, k, beta)

Adjust the edges between vertices 1:n of the graph g in accordance with the Watts-Strogatz model.

Parameters:
  • g – the input graph
  • n – the number of vertices between which to adjust edges
  • k – the base degree of each vertex (n > k, k >= 2, k must be even.)
  • beta – the probability of each edge being “rewired”.
Returns:

the graph g.

watts_strogatz_graph(n, k, beta)

Convenience function to construct an n-vertex Watts-Strogatz graph as an incidence list.