def test_child_non_critical(self):
# tests D consisting of non-critical path.
# Source cover is expected to be empty
D: nx.DiGraph = nx.DiGraph()
D.add_edge(0, 1)
cover = source_cover(D, set())
assert_set_equal(cover, set())
def test_tree_leafs_critical(self):
# tests D to be a tree, only leafs are critical
# Source cover is expected to be {0} - the root
D: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
critical: Set = {node for node in D.nodes if D.out_degree(node) == 0}
cover: Set = source_cover(D, critical)
assert_set_equal(cover, {0})
def test_no_critical(self):
# tests D consisting of single directed edge
# Source cover expected to not cover it
D: nx.DiGraph = nx.DiGraph()
D.add_node(0)
cover = source_cover(D, set())
assert_set_equal(cover, set())
def test_child_critical(self):
# tests D consisting of non-critical 0 and path to critical 4.
# Source cover is expected to be {0}
D: nx.DiGraph = nx.path_graph(5, nx.DiGraph())
D.add_edge(0, 1)
cover = source_cover(D, {4})
assert_set_equal(cover, {0})
def test_single_critical(self):
# tests D consisting of single critical vertex
# Source cover expected to cover it
D: nx.DiGraph = nx.DiGraph()
D.add_node(0)
cover = source_cover(D, {0})
assert_set_equal(cover, {0})
def test_tree_all_critical(self):
# tests D to be a tree, all vertices are critical
# Source cover is expected to be the root
# tests correct functionality of deletion of vertices
# reachable from a critical vertex
D: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
cover: Set = source_cover(D, set(D.nodes))
assert_set_equal(cover, {0})
def test_tree_leafs_critical_and_isolated(self):
# tests D to be a tree, only leafs are critical
# and set of isolated
# Source cover is expected to be root(D1) | isolated
D1: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
D2: nx.DiGraph = nx.DiGraph()
D2.add_nodes_from({i for i in range(100)})
D = nx.compose(D1, D2)
critical: Set = {node for node in D.nodes if D.in_degree(node) == 0}
cover: Set = source_cover(D, critical)
assert_set_equal(cover, critical)
def test_two_disjoint_trees_leafs_critical(self):
# tests two trees, only leafs are critical
# Source cover is expected to be {0} - the root
D1: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
D2: nx.DiGraph = nx.balanced_tree(2, 6, nx.DiGraph())
D = nx.compose(D1, D2)
critical: Set = {node for node in D.nodes if D.out_degree(node) == 0}
cover: Set = source_cover(D, critical)
assert_set_equal(cover,
{node
for node in D.nodes
if D.in_degree(node) == 0}) # Only roots form cover
def test_tree_multiple_sources_leafs_critical(self):
# tests D to be a tree, only leafs are critical
# There are multiple sources in the graphs, including two
# covering all the vertices. However, the optimal solution is
# still of size 1. tests if it correctly does not consider these
# additional source.
D: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
D.add_edges_from({(65, 1), (65, 2), (66, 15), (65, 29), (66, 38),
(67, 58)})
critical = {node for node in D.nodes if D.out_degree(node) == 0}
cover = source_cover(D, critical)
assert_true(cover == {0} or cover == {65}) # Only two correct options
def test_tree_all_critical_decoy(self):
# tests D to be a tree, all vertices are critical
# Source cover is expected to be the root
# tests correct functionality of deletion of vertices
# reachable from a critical vertex
# There is also one "decoy" - a non-critical vertex 100
# that covers two leafs. However, after the deletion,
# the vertex 100 does not cover any critical vertex
# any more. Hence, it should not be in the cover.
D: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
D.add_edges_from({(100, 50), (100, 51)})
cover: Set = source_cover(D, set(D.nodes) - {100})
assert_set_equal(cover, {0})
def test_not_covering_non_critical(self):
# Test for unbounded approximation factor as discussed
# in section 2.1 the thesis. The optimal solution is 1. We use the
# minimal solution. Expected source cover is of length 1 if it works correctly.
D: nx.DiGraph = nx.DiGraph()
D.add_nodes_from([
"s_1,2", "s_3,4", "s_5,6", "t_1", "t_2,3", "t_4,5", "t_6,7",
"t_8,9"
])
D.add_edges_from({("s_1,2", "t_1"), ("s_3,4", "t_1"), ("s_5,6", "t_1"),
("s_3,4", "t_2,3"), ("s_3,4", "t_4,5"),
("s_5,6", "t_6,7"), ("s_5,6", "t_8,9")})
cover = source_cover(D, {"t_1"})
assert_true(len(cover) == 1)
def test_tree_multiple_sources_tree_critical(self):
# tests D to be a tree, whole tree is critical
# There are multiple sources in the graphs, including two
# covering all the vertices. However, the optimal solution is
# still of size 1. In addition to test_tree_multiple_sources_leafs_critical(),
# this test also tests correctness of the deletion procedure.
# If it proceeds correctly, it does not consider the newly added sources
# as critical sinks, so the solution is still of size 1.
D: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
critical = set(D.nodes)
D.add_edges_from({(65, 1), (65, 2), (66, 15), (65, 29), (66, 38),
(67, 58)})
cover = source_cover(D, critical)
assert_true(
cover == {0}
) # All others are reachable from a critical vertex, so this is only viable option.
def test_tree_leafs_non_critical(self):
# tests D to be a tree, no vertices are critcal
# Source cover is expected to be empty
D: nx.DiGraph = nx.balanced_tree(2, 5, nx.DiGraph())
cover: Set = source_cover(D, set())
assert_set_equal(cover, set())
def bipartite_matching_augmentation(G: nx.Graph, A: Set, M: Dict = None):
"""Returns a set of edges A such that G(V, E + A) is strongly connected.
Parameters
----------
G : NetworkX Graph
A bipartite graph G = (A + B, E), where G can be augmented, that is |A + B| >= 4.
G must also admit a perfect matching.
A : Set
A bipartition of G, where |A| = |A + B| / 2
M: Dict = None
A perfect bipartite matching of G, for each edge {a, b} in M holds M[a] = b, M[b] = a.
If M is not given, it will be computed using eppstein_matching(G, A)
Returns
-------
L : Set
Set of edges from E(G) - M such that G admits a perfect matching even after a single arbitrary
edge is removed. Edges are in form of (a, b), where a is from A and b from B
Raises
------
NetworkX.NotImplemented:
If G is directed or a multigraph.
Notes
-----
Implementation is based on BINDEWALD, Viktor; HOMMELSHEIM, Felix; MÜHLENTHALER, Moritz; SCHAUDT, Oliver.
How to Secure Matchings Against Edge Failures. CoRR. 2018, vol. abs/1805.01299. Available from arXiv:
1805.01299
References
----------
[1] BINDEWALD, Viktor; HOMMELSHEIM, Felix; MÜHLENTHALER, Moritz; SCHAUDT, Oliver.
How to Secure Matchings Against Edge Failures. CoRR. 2018, vol. abs/1805.01299. Available from arXiv:
1805.01299
"""
if len(
A
) <= 1: # Graph consisting of only one vertex at each bipartition cannot be augmented.
raise bipartite_ghraph_not_augmentable_exception(
"G cannot be augmented.")
if M is None: # User can specify her own matching for speed-up
M: Dict = nx.algorithms.bipartite.eppstein_matching(G, A)
D: nx.DiGraph = nx.DiGraph()
for u in M: # Construction of D, iterate over all keys in M
if u in A: # Add all edges from A to D
w = M[u]
D.add_node(u)
outgoing: int = 0
for uPrime in G.neighbors(w): # Construct edges of D
if uPrime != u:
D.add_edge(u, uPrime)
D_condensation: nx.DiGraph = nx.algorithms.components.condensation(
D) # Condensation - acyclic digraph
X: Set = set(
) # A set of vertices of D_condensation corresponding to trivial strong components of D
isolated: Set = set() # Set of isolated vertices
sources: Set = set(
) # Set of vertices that are not isolated and have no ingoing arc
sinks: Set = set(
) # Set of vertices that are not isolated and have no outgoing arc
# We do not use function sources, sinks, isolated for performance reasons
# Because we would either have to loop twice or check one more condition in the loop
# if we were to modify the function.
for vertex in D_condensation.nodes:
inDegree: int = D_condensation.in_degree(vertex)
outDegree: int = D_condensation.out_degree(vertex)
if len(D_condensation.nodes[vertex]['members']) == 1:
# Each trivial strong component is incident to some critical edge
X.add(vertex)
if inDegree == 0 and outDegree == 0:
# Isolated: neither ingoing nor outgoing arc
isolated.add(vertex)
elif inDegree == 0:
# Source: no ingoing arc and not isolated
sources.add(vertex)
# Sink: no outgoing arc and not isolated
elif outDegree == 0:
sinks.add(vertex)
A_0 = D_condensation
A_1 = D_condensation.reverse(copy=False)
if len(
X
) == 0: # If there is no trivial strong component, G admits a perfect matching after edge removal
return set()
# Use source_cover to choose ln(n) approximation of choice of sources that cover all sinks in C_0, resp. C_1
C_0 = source_cover(A_0, X, (sources, sinks, isolated))
C_1 = source_cover(A_1, X, (sinks, sources, isolated))
# We now determine vertices that lie either on C_1X paths or XC_2 paths
# Vertices on C_1X paths are those visited when traveling from C_1 to X on
# D_condensation and from X to C_1 on D_condensation_reverse
CX_vertices = set()
XC_vertices = set()
D_condensation_reverse = D_condensation.reverse(copy=False)
# Now, specify arguments for the fast DFS traversal
def action_on_vertex_CX(current_vertex):
CX_vertices.add(
current_vertex) # We have reached it from C, so add it in
return True
def action_on_neighbor_CX(neighbor, current_vertex):
return neighbor not in CX_vertices
def action_on_vertex_XC(current_vertex):
XC_vertices.add(current_vertex)
return True
def action_on_neighbor_XC(neighbor, current_vertex):
return neighbor not in XC_vertices
for source in C_0:
# Reachable from C_0 (search for X)
fast_traversal(D_condensation, source, action_on_vertex_CX,
action_on_neighbor_CX)
for critical in X:
# Reachable from X (search for C_2)
fast_traversal(D_condensation, critical, action_on_vertex_CX,
action_on_neighbor_CX)
# Reachable from X (search for C_1)
fast_traversal(D_condensation_reverse, critical, action_on_vertex_XC,
action_on_neighbor_XC)
for source in C_1:
# Reachable from C_2 (search for X)
fast_traversal(D_condensation_reverse, source, action_on_vertex_XC,
action_on_neighbor_XC)
D_hat_vertices = CX_vertices & XC_vertices # Intersection
# Marginal case when single vertex cannot be connected to form non-trivial strongly connected component.
# We need to add another arbitrary vertex, which always exists as |V(D)| is guaranteed to be > 2 and contains
# at least one trivial strong component.
if len(D_hat_vertices) == 1:
vert = next(iter(D_hat_vertices))
D_hat_vertices.add(next(iter(set(D_condensation.nodes) - {vert})))
# Update sources, sinks, isolated as intersection with D_hat_vertices
sources &= D_hat_vertices
sinks &= D_hat_vertices
isolated &= D_hat_vertices
D_hat = nx.classes.function.induced_subgraph(D_condensation,
D_hat_vertices)
L_star: Set = eswaran_tarjan(D_hat,
is_condensation=True,
sourcesSinksIsolated=(sources, sinks,
isolated))
# Map vertices from L to vertices of L*
return set(
map(
lambda e:
(next(iter(D_condensation.nodes[e[1]]['members'])), M[next(
iter(D_condensation.nodes[e[0]]['members']))]), L_star))
