petgraph/algo/floyd_warshall.rs
1use alloc::{vec, vec::Vec};
2use core::hash::Hash;
3
4use hashbrown::HashMap;
5
6use crate::algo::{BoundedMeasure, NegativeCycle};
7use crate::visit::{
8 EdgeRef, GraphProp, IntoEdgeReferences, IntoNodeIdentifiers, NodeCompactIndexable,
9};
10
11#[allow(clippy::type_complexity, clippy::needless_range_loop)]
12/// \[Generic\] [Floyd–Warshall algorithm](https://en.wikipedia.org/wiki/Floyd%E2%80%93Warshall_algorithm) is an algorithm for all pairs shortest path problem
13///
14/// Compute the length of each shortest path in a weighted graph with positive or negative edge weights (but with no negative cycles).
15///
16/// # Arguments
17/// * `graph`: graph with no negative cycle.
18/// * `edge_cost`: closure that returns cost of a particular edge.
19///
20/// # Returns
21/// * `Ok`: (if graph contains no negative cycle) a [`struct@hashbrown::HashMap`] containing all pairs shortest paths.
22/// * `Err`: if graph contains negative cycle.
23///
24/// # Complexity
25/// * Time complexity: **O(|V|³)**.
26/// * Auxiliary space: **O(|V|²)**.
27///
28/// where **|V|** is the number of nodes.
29///
30/// **Note**: If the graph is sparse (the number of edges is quite small),
31/// consider using [`johnson`](fn@crate::algo::johnson)'s algorithm, which is likely to show better performance in such cases.
32///
33/// # Examples
34/// ```rust
35/// use petgraph::{prelude::*, Graph, Directed};
36/// use petgraph::algo::floyd_warshall;
37/// use hashbrown::HashMap;
38///
39/// let mut graph: Graph<(), (), Directed> = Graph::new();
40/// let a = graph.add_node(());
41/// let b = graph.add_node(());
42/// let c = graph.add_node(());
43/// let d = graph.add_node(());
44///
45/// graph.extend_with_edges(&[
46/// (a, b),
47/// (a, c),
48/// (a, d),
49/// (b, c),
50/// (b, d),
51/// (c, d)
52/// ]);
53///
54/// let weight_map: HashMap<(NodeIndex, NodeIndex), i32> = [
55/// ((a, a), 0), ((a, b), 1), ((a, c), 4), ((a, d), 10),
56/// ((b, b), 0), ((b, c), 2), ((b, d), 2),
57/// ((c, c), 0), ((c, d), 2)
58/// ].iter().cloned().collect();
59/// // ----- b --------
60/// // | ^ | 2
61/// // | 1 | 4 v
62/// // 2 | a ------> c
63/// // | 10 | | 2
64/// // | v v
65/// // ---> d <-------
66///
67/// let inf = core::i32::MAX;
68/// let expected_res: HashMap<(NodeIndex, NodeIndex), i32> = [
69/// ((a, a), 0), ((a, b), 1), ((a, c), 3), ((a, d), 3),
70/// ((b, a), inf), ((b, b), 0), ((b, c), 2), ((b, d), 2),
71/// ((c, a), inf), ((c, b), inf), ((c, c), 0), ((c, d), 2),
72/// ((d, a), inf), ((d, b), inf), ((d, c), inf), ((d, d), 0),
73/// ].iter().cloned().collect();
74///
75///
76/// let res = floyd_warshall(&graph, |edge| {
77/// if let Some(weight) = weight_map.get(&(edge.source(), edge.target())) {
78/// *weight
79/// } else {
80/// inf
81/// }
82/// }).unwrap();
83///
84/// let nodes = [a, b, c, d];
85/// for node1 in &nodes {
86/// for node2 in &nodes {
87/// assert_eq!(res.get(&(*node1, *node2)).unwrap(), expected_res.get(&(*node1, *node2)).unwrap());
88/// }
89/// }
90/// ```
91pub fn floyd_warshall<G, F, K>(
92 graph: G,
93 edge_cost: F,
94) -> Result<HashMap<(G::NodeId, G::NodeId), K>, NegativeCycle>
95where
96 G: NodeCompactIndexable + IntoEdgeReferences + IntoNodeIdentifiers + GraphProp,
97 G::NodeId: Eq + Hash,
98 F: FnMut(G::EdgeRef) -> K,
99 K: BoundedMeasure + Copy,
100{
101 let num_of_nodes = graph.node_count();
102
103 // |V|x|V| matrix
104 let mut m_dist = Some(vec![vec![K::max(); num_of_nodes]; num_of_nodes]);
105
106 _floyd_warshall_path(graph, edge_cost, &mut m_dist, &mut None)?;
107
108 let mut distance_map: HashMap<(G::NodeId, G::NodeId), K> =
109 HashMap::with_capacity(num_of_nodes * num_of_nodes);
110
111 if let Some(dist) = m_dist {
112 for i in 0..num_of_nodes {
113 for j in 0..num_of_nodes {
114 distance_map.insert((graph.from_index(i), graph.from_index(j)), dist[i][j]);
115 }
116 }
117 }
118
119 Ok(distance_map)
120}
121
122#[allow(clippy::type_complexity, clippy::needless_range_loop)]
123/// \[Generic\] [Floyd–Warshall algorithm](https://en.wikipedia.org/wiki/Floyd%E2%80%93Warshall_algorithm) is an algorithm for all pairs shortest path problem
124///
125/// Compute all pairs shortest paths in a weighted graph with positive or negative edge weights (but with no negative cycles).
126/// Returns HashMap of shortest path lengths. Additionally, returns HashMap of intermediate nodes along shortest path for indicated edges.
127///
128/// # Arguments
129/// * `graph`: graph with no negative cycle
130/// * `edge_cost`: closure that returns cost of a particular edge
131///
132/// # Returns
133/// * `Ok`: (if graph contains no negative cycle) a hashmap containing all pairs shortest path distances and a hashmap for all pairs shortest paths
134/// * `Err`: if graph contains negative cycle.
135///
136/// # Complexity
137/// * Time complexity: **O(|V|³)**
138/// * Auxiliary space: **O(|V|²)**
139///
140/// # Examples
141/// ```rust
142/// use petgraph::{prelude::*, Graph, Directed};
143/// use petgraph::algo::floyd_warshall::floyd_warshall_path;
144/// use std::collections::HashMap;
145///
146/// let mut graph: Graph<(), (), Directed> = Graph::new();
147/// let a = graph.add_node(());
148/// let b = graph.add_node(());
149/// let c = graph.add_node(());
150/// let d = graph.add_node(());
151///
152/// graph.extend_with_edges(&[
153/// (a, b),
154/// (a, c),
155/// (a, d),
156/// (b, c),
157/// (b, d),
158/// (c, d)
159/// ]);
160///
161/// let weight_map: HashMap<(NodeIndex, NodeIndex), i32> = [
162/// ((a, a), 0), ((a, b), 1), ((a, c), 4), ((a, d), 10),
163/// ((b, b), 0), ((b, c), 2), ((b, d), 2),
164/// ((c, c), 0), ((c, d), 2)
165/// ].iter().cloned().collect();
166/// // ----- b --------
167/// // | ^ | 2
168/// // | 1 | 4 v
169/// // 2 | a ------> c
170/// // | 10 | | 2
171/// // | v v
172/// // ---> d <-------
173///
174/// let inf = std::i32::MAX;
175/// let expected_res: HashMap<(NodeIndex, NodeIndex), i32> = [
176/// ((a, a), 0), ((a, b), 1), ((a, c), 3), ((a, d), 3),
177/// ((b, a), inf), ((b, b), 0), ((b, c), 2), ((b, d), 2),
178/// ((c, a), inf), ((c, b), inf), ((c, c), 0), ((c, d), 2),
179/// ((d, a), inf), ((d, b), inf), ((d, c), inf), ((d, d), 0),
180/// ].iter().cloned().collect();
181///
182///
183/// let (res, prev) = floyd_warshall_path(&graph, |edge| {
184/// if let Some(weight) = weight_map.get(&(edge.source(), edge.target())) {
185/// *weight
186/// } else {
187/// inf
188/// }
189/// }).unwrap();
190///
191/// assert_eq!(prev[0][2], Some(1));
192///
193/// let nodes = [a, b, c, d];
194/// for node1 in &nodes {
195/// for node2 in &nodes {
196/// assert_eq!(res.get(&(*node1, *node2)).unwrap(), expected_res.get(&(*node1, *node2)).unwrap());
197/// }
198/// }
199///
200/// ```
201pub fn floyd_warshall_path<G, F, K>(
202 graph: G,
203 edge_cost: F,
204) -> Result<(HashMap<(G::NodeId, G::NodeId), K>, Vec<Vec<Option<usize>>>), NegativeCycle>
205where
206 G: NodeCompactIndexable + IntoEdgeReferences + IntoNodeIdentifiers + GraphProp,
207 G::NodeId: Eq + Hash,
208 F: FnMut(G::EdgeRef) -> K,
209 K: BoundedMeasure + Copy,
210{
211 let num_of_nodes = graph.node_count();
212
213 // |V|x|V| matrix
214 let mut m_dist = Some(vec![vec![K::max(); num_of_nodes]; num_of_nodes]);
215 // `prev[source][target]` holds the penultimate vertex on path from `source` to `target`, except `prev[source][source]`, which always stores `source`.
216 let mut m_prev = Some(vec![vec![None; num_of_nodes]; num_of_nodes]);
217
218 _floyd_warshall_path(graph, edge_cost, &mut m_dist, &mut m_prev)?;
219
220 let mut distance_map = HashMap::with_capacity(num_of_nodes * num_of_nodes);
221
222 if let Some(dist) = m_dist {
223 for i in 0..num_of_nodes {
224 for j in 0..num_of_nodes {
225 distance_map.insert((graph.from_index(i), graph.from_index(j)), dist[i][j]);
226 }
227 }
228 }
229
230 Ok((distance_map, m_prev.unwrap()))
231}
232
233/// Helper function to copy a value to a 2D array
234fn set_object<K: Clone>(m_dist: &mut Option<Vec<Vec<K>>>, i: usize, j: usize, value: K) {
235 if let Some(dist) = m_dist {
236 dist[i][j] = value;
237 }
238}
239
240/// Helper to check if the distance map is greater then a specific value
241fn is_greater<K: PartialOrd>(
242 m_dist: &mut Option<Vec<Vec<K>>>,
243 i: usize,
244 j: usize,
245 value: K,
246) -> bool {
247 if let Some(dist) = m_dist {
248 return dist[i][j] > value;
249 }
250 false
251}
252
253/// Helper that implements the floyd warshall routine, but paths are optional for memory overhead.
254fn _floyd_warshall_path<G, F, K>(
255 graph: G,
256 mut edge_cost: F,
257 m_dist: &mut Option<Vec<Vec<K>>>,
258 m_prev: &mut Option<Vec<Vec<Option<usize>>>>,
259) -> Result<(), NegativeCycle>
260where
261 G: NodeCompactIndexable + IntoEdgeReferences + IntoNodeIdentifiers + GraphProp,
262 G::NodeId: Eq + Hash,
263 F: FnMut(G::EdgeRef) -> K,
264 K: BoundedMeasure + Copy,
265{
266 let num_of_nodes = graph.node_count();
267
268 // Initialize distances and predecessors for edges
269 for edge in graph.edge_references() {
270 let source = graph.to_index(edge.source());
271 let target = graph.to_index(edge.target());
272 let cost = edge_cost(edge);
273 if is_greater(m_dist, source, target, cost) {
274 set_object(m_dist, source, target, cost);
275 set_object(m_prev, source, target, Some(source));
276
277 if !graph.is_directed() {
278 set_object(m_dist, target, source, cost);
279 set_object(m_prev, target, source, Some(target));
280 }
281 }
282 }
283
284 // Distance of each node to itself is the default value
285 for node in graph.node_identifiers() {
286 let index = graph.to_index(node);
287 set_object(m_dist, index, index, K::default());
288 set_object(m_prev, index, index, Some(index));
289 }
290
291 // Perform the Floyd-Warshall algorithm
292 for k in 0..num_of_nodes {
293 for i in 0..num_of_nodes {
294 for j in 0..num_of_nodes {
295 if let Some(dist) = m_dist {
296 let (result, overflow) = dist[i][k].overflowing_add(dist[k][j]);
297 if !overflow && dist[i][j] > result {
298 dist[i][j] = result;
299 if let Some(prev) = m_prev {
300 prev[i][j] = prev[k][j];
301 }
302 }
303 }
304 }
305 }
306 }
307
308 // value less than 0(default value) indicates a negative cycle
309 for i in 0..num_of_nodes {
310 if let Some(dist) = m_dist {
311 if dist[i][i] < K::default() {
312 return Err(NegativeCycle(()));
313 }
314 }
315 }
316 Ok(())
317}