def __init__(self, nx_graph=nx.Graph()):
self.bipartite_graph = nx.Graph()
self.node = self.bipartite_graph.node
self.next_edge_index = 0
self.next_hedge_index = 0
self.nodes_count = 0
self.edges_count = 0
self.hedges_count = 0
# ready sets
self.reset_nodes_with_more_labels()
self.reset_self_loops()
self.reset_parallel_hedges_groups()
# add nodes
for node in nx_graph.nodes_iter():
self.add_node(node, attr_dict=copy.deepcopy(nx_graph.node[node]))
# add edges of order 2
if nx_graph.is_directed():
adj_nodes = nxext.get_all_adjacent_nodes(nx_graph)
for pair in adj_nodes:
edges = nxext.get_edge_labels_and_dirs(nx_graph, pair[0],
pair[1])
u = Hypergraph.format_node_id(pair[0])
v = Hypergraph.format_node_id(pair[1])
for edge in edges:
dir_code = edge[1]
if dir_code == 0:
direction = None
elif dir_code > 0:
direction = set([(u, v)])
else:
direction = set([(v, u)])
self.add_edge(set([u, v]),
direction=direction,
label=copy.deepcopy(edge[0]))
else:
for edge_endpoints in nx_graph.edges_iter():
u = u"n_{0}".format(edge_endpoints[0])
v = u"n_{0}".format(edge_endpoints[1])
if nx_graph.is_multigraph():
edges = nx_graph[edge_endpoints[0]][edge_endpoints[1]]
for i in range(len(edges)):
self.add_edge(set([u, v]), label=edges[i]["label"])
else:
edge_label = nx_graph.edge[edge_endpoints[0]][
edge_endpoints[1]]["label"]
self.add_edge(set([u, v]), label=copy.deepcopy(edge_label))
# Initialize ready sets
self.init_parallel_edges_groups()
self.init_nodes_with_n_neighbors()
def __init__(self, nx_graph = nx.Graph()):
self.bipartite_graph = nx.Graph()
self.node = self.bipartite_graph.node
self.next_edge_index = 0
self.next_hedge_index = 0
self.nodes_count = 0
self.edges_count = 0
self.hedges_count = 0
# ready sets
self.reset_nodes_with_more_labels()
self.reset_self_loops()
self.reset_parallel_hedges_groups()
# add nodes
for node in nx_graph.nodes_iter():
self.add_node(node, attr_dict=copy.deepcopy(nx_graph.node[node]))
# add edges of order 2
if nx_graph.is_directed():
adj_nodes = nxext.get_all_adjacent_nodes(nx_graph)
for pair in adj_nodes:
edges = nxext.get_edge_labels_and_dirs(nx_graph, pair[0], pair[1])
u = Hypergraph.format_node_id(pair[0])
v = Hypergraph.format_node_id(pair[1])
for edge in edges:
dir_code = edge[1]
if dir_code == 0:
direction = None
elif dir_code > 0:
direction = set([(u, v)])
else:
direction = set([(v, u)])
self.add_edge(set([u, v]), direction=direction, label=copy.deepcopy(edge[0]))
else:
for edge_endpoints in nx_graph.edges_iter():
u = u"n_{0}".format(edge_endpoints[0])
v = u"n_{0}".format(edge_endpoints[1])
if nx_graph.is_multigraph():
edges = nx_graph[edge_endpoints[0]][edge_endpoints[1]]
for i in range(len(edges)):
self.add_edge(set([u, v]), label=edges[i]["label"])
else:
edge_label = nx_graph.edge[edge_endpoints[0]][edge_endpoints[1]]["label"]
self.add_edge(set([u, v]), label=copy.deepcopy(edge_label))
# Initialize ready sets
self.init_parallel_edges_groups()
self.init_nodes_with_n_neighbors()
def get_all_parallel_edge_free_groupings():
'''Finds all groups of parallel edges and returns all possible graphs
where only one edge per group of parallel edges remains.
'''
def process_parallel_edgess(all_adj_nodes, i, new_feature):
'''Process parallel edges between nodes u and v recursively
through all adjacent pairs of nodes.
'''
u, v = all_adj_nodes[i]
edges = nxext.get_edge_labels_and_dirs(feature, u, v)
edges_count = len(edges)
for j, edge in enumerate(edges):
if j < edges_count - 1:
_new_feature = new_feature.copy()
else:
# no need to copy when the graph will not be used for other iterations
_new_feature = new_feature
if edge[1] <= 0:
_new_feature.add_edge(v, u, label=edge[0])
if edge[1] >= 0:
_new_feature.add_edge(u, v, label=edge[0])
if i < len(all_adj_nodes) - 1:
for processed_feature in process_parallel_edgess(all_adj_nodes, i + 1, _new_feature):
yield processed_feature
else:
yield _new_feature
all_adj_nodes = list(nxext.get_all_adjacent_nodes(feature))
if len(all_adj_nodes) == 0:
return [feature]
else:
new_feature = nx.MultiDiGraph()
new_feature.add_nodes_from(feature.nodes_iter(data=True))
return process_parallel_edgess(all_adj_nodes, 0, new_feature)
def process_raw_feature(raw_feature, hypergraph, max_nodes=6):
'''Turns a raw feature to a usable feature or a collection of features, depending on
the type of the rule according to which the feature was reduced (fixed, pattern or dynamic).
:param raw_feature: A ReducibleFeature extracted from Arnborg & Proskurowski
:param hypergraph: A Hypergraph which contains the nodes of the raw feature.
:param max_nodes: (default value 6) A number of nodes that a pattern or a dynamic feature can
have before being disassembled in subfeatures of size max_nodes.
:return A collection containing one or more features depending on the type of the raw feature.
'''
assert type(raw_feature) is ReducibleFeature
assert type(hypergraph) is Hypergraph
assert max_nodes > 3
def get_feature_type(raw_feature):
'''Get the type of the raw feature according to the rule it was reduced by.
:param raw_feature: A ReducibleFeature.
:return Type of the raw feature: 0 for "fixed", 1 for "pattern", 2 for "dynamic".
'''
rule = raw_feature.get_full_rule()
if rule in ["2.1.0.0", "2.2.0.0", "4.2.0.0", "4.3.0.0", "5.2.3.1"]:
return 1
elif rule in ["5.2.2.0", "5.2.3.2", "5.2.4.0"]:
return 2
else:
return 0
def sliding_window(seq, window_size):
# Returns a sliding window (of width window_size) over data from the iterable seq
# s -> (s0,s1,...s[n-1]), (s1,s2,...,sn), ...
it = iter(seq)
result = tuple(islice(it, window_size))
if len(result) == window_size:
yield result
for elem in it:
result = result[1:] + (elem,)
yield result
feature_type = get_feature_type(raw_feature)
if feature_type == 1:
# pattern
nodes_count = raw_feature.number_of_nodes()
if nodes_count > max_nodes:
rule = raw_feature.get_full_rule()
if rule == "2.1.0.0":
# chain: extract all subpaths of length max_nodes using a sliding window
s = raw_feature.peripheral_nodes[0]
t = raw_feature.peripheral_nodes[1]
path = [s] + raw_feature.reducible_nodes + [t]
for subpath in sliding_window(path, max_nodes):
yield hypergraph.subgraph_with_labels(set(subpath))
raise StopIteration
elif rule == "2.2.0.0":
# ring: extract all subpaths of length max_nodes using a sliding window
cycle = raw_feature.peripheral_nodes + raw_feature.reducible_nodes
for subpath in sliding_window(cycle + cycle[:max_nodes - 1], max_nodes):
yield hypergraph.subgraph_with_labels(set(subpath))
raise StopIteration
elif rule == "4.2.0.0":
# buddy: If there are more than 3 buddies create a buddy
# configuration with 3 buddies for each possible combination
assert len(raw_feature.peripheral_nodes) == 3
buddy_nodes = raw_feature.reducible_nodes
for buddy_nodes_subgroup in itertools.combinations(buddy_nodes, max_nodes - 3):
yield hypergraph.subgraph_with_labels(set(buddy_nodes_subgroup) | set(raw_feature.peripheral_nodes))
raise StopIteration
elif rule == "4.3.0.0":
# cube: similar approach as for wheel (5.2.3.1)
if nodes_count > max_nodes + 1:
reducible_neigh = [set(hypergraph.neighbors(node)) for node in raw_feature.reducible_nodes]
hub_node = reduce(lambda a, b: a.intersection(b), reducible_neigh)
ring_subgraph = hypergraph.subgraph(raw_feature.reducible_nodes | (raw_feature.peripheral_nodes - hub_node))
ring = nx.cycle_basis(ring_subgraph)
if len(ring) > 0:
ring = ring[0]
for subpath in sliding_window(ring + ring[:max_nodes - 1], max_nodes):
yield hypergraph.subgraph_with_labels(set(subpath) | hub_node)
raise StopIteration
else:
# if there is no ring in the feature, treat it as a fixed feature
# TODO: This can lead to a large number of shingles. Better solution?
pass
elif rule == "5.2.3.1":
# wheel: extract all cake-slice subpaths of length max_node using a sliding window
if nodes_count > max_nodes + 1:
ring_subgraph = hypergraph.subgraph(raw_feature.reducible_nodes)
ring = nx.cycle_basis(ring_subgraph)
if len(ring) > 0:
ring = ring[0]
for subpath in sliding_window(ring + ring[:max_nodes - 1], max_nodes):
yield hypergraph.subgraph_with_labels(set(subpath) | set(raw_feature.peripheral_nodes))
raise StopIteration
else:
# if there is no ring in the feature, treat it as a fixed feature
# TODO: This can lead to a large number of shingles. Better solution?
pass
elif feature_type == 2:
# dynamic (the reducible nodes are always of degree 3):
# for each pair of adjacent nodes u, v let a new feature be
# the subgraph that contains u, v and all neighbors of u and v.
nodes_count = raw_feature.number_of_nodes()
if nodes_count > max_nodes:
nodes = set(raw_feature.reducible_nodes) | set(raw_feature.peripheral_nodes)
feature_graph = hypergraph.subgraph(nodes)
adj_nodes = nxext.get_all_adjacent_nodes(feature_graph)
neighbors = {node: nxext.get_all_neighbors(feature_graph, node) for node in nodes}
for u, v in adj_nodes:
node_subgroup = set([u, v] + neighbors[u] + neighbors[v])
yield hypergraph.subgraph_with_labels(set(node_subgroup))
raise StopIteration
# fixed or pattern/dynamic with <= max_nodes number of nodes
yield raw_feature.as_subgraph(hypergraph)