def run(self):
"""Finding all shortest paths."""
# graph copy
size = self.graph.v()
self._new_graph = self.graph.__class__(size + 1, directed=True)
for node in self.graph.iternodes(): # O(V) time
self._new_graph.add_node(node)
for edge in self.graph.iteredges(): # O(E) time
self._new_graph.add_edge(edge)
self._new_node = size
self._new_graph.add_node(self._new_node)
for node in self.graph.iternodes(): # O(V) time
self._new_graph.add_edge(Edge(self._new_node, node, 0))
self._bf = BellmanFord(self._new_graph)
# If this step detects a negative cycle, the algorithm is terminated.
self._bf.run(self._new_node) # O(V*E) time
# Edges are reweighted.
for edge in list(self._new_graph.iteredges()): # O(E) time
edge.weight = (edge.weight + self._bf.distance[edge.source] -
self._bf.distance[edge.target])
self._new_graph.del_edge(edge)
self._new_graph.add_edge(edge)
# Remove _new_node with edges.
self._new_graph.del_node(self._new_node)
# Weights are modified!
self.distance = dict()
for source in self.graph.iternodes():
self.distance[source] = dict()
algorithm = Dijkstra(self._new_graph) # O(V*E*log(V)) total time
algorithm.run(source)
for target in self.graph.iternodes(): # O(V^2) total time
self.distance[source][target] = (algorithm.distance[target] -
self._bf.distance[source] +
self._bf.distance[target])
def test_shortest_path_cormen(self):
source = 0
target = 4
algorithm = BellmanFord(self.G)
algorithm.run(source)
distance_expected = {3: 7, 2: 4, 0: 0, 4: -2, 1: 2}
self.assertEqual(algorithm.distance, distance_expected)
parent_expected = {3: 0, 2: 3, 0: None, 4: 1, 1: 2}
self.assertEqual(algorithm.parent, parent_expected)
path_expected = [0, 3, 2, 1, 4]
self.assertEqual(algorithm.path(target), path_expected)
def test_shortest_path(self):
source = 0
target = 3
algorithm = BellmanFord(self.G)
algorithm.run(source)
distance_expected = {0: 0, 2: 2, 1: 1, 3: 3}
self.assertEqual(algorithm.distance, distance_expected)
parent_expected = {0: None, 2: 1, 1: 0, 3: 2}
self.assertEqual(algorithm.parent, parent_expected)
path_expected = [0, 1, 2, 3]
self.assertEqual(algorithm.path(target), path_expected)
def run(self):
"""Finding all shortest paths."""
if self.positive_weights:
self._new_graph = self.graph
else:
# graph copy
self._new_graph = self.graph.__class__(self.graph.v()+1, directed=True)
for node in self.graph.iternodes(): # O(V) time
self._new_graph.add_node(node)
for edge in self.graph.iteredges(): # O(E) time
self._new_graph.add_edge(edge)
self._new_node = self.graph.v()
self._new_graph.add_node(self._new_node)
for node in self.graph.iternodes(): # O(V) time
self._new_graph.add_edge(Edge(self._new_node, node, 0))
self._bf = BellmanFord(self._new_graph)
# If this step detects a negative cycle,
# the algorithm is terminated.
self._bf.run(self._new_node) # O(V*E) time
# Edges are reweighted.
for edge in list(self._new_graph.iteredges()): # O(E) time
edge.weight = (edge.weight
+ self._bf.distance[edge.source]
- self._bf.distance[edge.target])
self._new_graph.del_edge(edge)
self._new_graph.add_edge(edge)
# Remove _new_node with edges.
self._new_graph.del_node(self._new_node)
self.distance = dict()
for source in self.graph.iternodes():
self.distance[source] = dict()
algorithm = Dijkstra(self._new_graph) # O(V*E*log(V)) total time
algorithm.run(source)
for target in self.graph.iternodes(): # O(V**2) total time
if self.positive_weights:
self.distance[source][target] = algorithm.distance[target]
else:
self.distance[source][target] = (
algorithm.distance[target]
- self._bf.distance[source]
+ self._bf.distance[target])
class JohnsonFaster:
"""The Johnson algorithm for the shortest path problem.
Attributes
----------
graph : input directed weighted graph
distance : dict-of-dict
positive_weights : bool
_new_node : node, private
_new_graph : graph, private
_bf : BellmanFord instance, private
Examples
--------
>>> from graphtheory.structures.edges import Edge
>>> from graphtheory.structures.graphs import Graph
>>> from graphtheory.shortestpaths.johnson import JohnsonFaster
>>> G = Graph(n=10, True) # an exemplary directed graph
# Add nodes and edges here.
>>> algorithm = JohnsonFaster(G) # initialization
>>> algorithm.run() # calculations
>>> algorithm.distance[source][target] # distance from source to target
Notes
-----
Based on:
Cormen, T. H., Leiserson, C. E., Rivest, R. L., and Stein, C., 2009,
Introduction to Algorithms, third edition, The MIT Press,
Cambridge, London.
https://en.wikipedia.org/wiki/Johnson's_algorithm
"""
def __init__(self, graph):
"""The algorithm initialization.
Parameters
----------
graph : directed weighted graph
"""
if not graph.is_directed():
raise ValueError("the graph is not directed")
self.graph = graph
self.positive_weights = all(edge.weight >= 0
for edge in self.graph.iteredges()) # O(E) time
self.distance = None
def run(self):
"""Finding all shortest paths."""
if self.positive_weights:
self._new_graph = self.graph
else:
# graph copy
self._new_graph = self.graph.__class__(self.graph.v()+1, directed=True)
for node in self.graph.iternodes(): # O(V) time
self._new_graph.add_node(node)
for edge in self.graph.iteredges(): # O(E) time
self._new_graph.add_edge(edge)
self._new_node = self.graph.v()
self._new_graph.add_node(self._new_node)
for node in self.graph.iternodes(): # O(V) time
self._new_graph.add_edge(Edge(self._new_node, node, 0))
self._bf = BellmanFord(self._new_graph)
# If this step detects a negative cycle,
# the algorithm is terminated.
self._bf.run(self._new_node) # O(V*E) time
# Edges are reweighted.
for edge in list(self._new_graph.iteredges()): # O(E) time
edge.weight = (edge.weight
+ self._bf.distance[edge.source]
- self._bf.distance[edge.target])
self._new_graph.del_edge(edge)
self._new_graph.add_edge(edge)
# Remove _new_node with edges.
self._new_graph.del_node(self._new_node)
self.distance = dict()
for source in self.graph.iternodes():
self.distance[source] = dict()
algorithm = Dijkstra(self._new_graph) # O(V*E*log(V)) total time
algorithm.run(source)
for target in self.graph.iternodes(): # O(V**2) total time
if self.positive_weights:
self.distance[source][target] = algorithm.distance[target]
else:
self.distance[source][target] = (
algorithm.distance[target]
- self._bf.distance[source]
+ self._bf.distance[target])
