def test_min_edge_dominating_set(self):
graph = nx.path_graph(5)
dom_set = min_edge_dominating_set(graph)
# this is a crappy way to test, but good enough for now.
for edge in graph.edges():
if edge in dom_set:
continue
else:
u, v = edge
found = False
for dom_edge in dom_set:
found |= u == dom_edge[0] or u == dom_edge[1]
ok_(found, "Non adjacent edge found!")
graph = nx.complete_graph(10)
dom_set = min_edge_dominating_set(graph)
# this is a crappy way to test, but good enough for now.
for edge in graph.edges():
if edge in dom_set:
continue
else:
u, v = edge
found = False
for dom_edge in dom_set:
found |= u == dom_edge[0] or u == dom_edge[1]
ok_(found, "Non adjacent edge found!")
def test_min_edge_dominating_set(self):
graph = nx.path_graph(5)
dom_set = apxa.min_edge_dominating_set(graph)
# this is a crappy way to test, but good enough for now.
for edge in graph.edges_iter():
if edge in dom_set:
continue
else:
u, v = edge
found = False
for dom_edge in dom_set:
found |= u == dom_edge[0] or u == dom_edge[1]
ok_(found, "Non adjacent edge found!")
graph = nx.complete_graph(10)
dom_set = apxa.min_edge_dominating_set(graph)
# this is a crappy way to test, but good enough for now.
for edge in graph.edges_iter():
if edge in dom_set:
continue
else:
u, v = edge
found = False
for dom_edge in dom_set:
found |= u == dom_edge[0] or u == dom_edge[1]
ok_(found, "Non adjacent edge found!")
def retrieve_graph_attributes(G):
"""Read in a graph and return an array of graph statistics"""
n_nodes = networkx.number_of_nodes(G)
n_edges = networkx.number_of_edges(G)
min_dom_set = min_weighted_dominating_set(G)
min_edge_dom_set = min_edge_dominating_set(G)
ramsey = networkx.algorithms.approximation.ramsey.ramsey_R2(G)
assortativity = networkx.algorithms.assortativity.degree_assortativity_coefficient(G)
pearson = networkx.algorithms.assortativity.degree_pearson_correlation_coefficient(G)
neighbor = networkx.algorithms.assortativity.neighbor_degree.average_neighbor_degree(G)
average_connectivity = networkx.algorithms.assortativity.average_degree_connectivity(G)
knn = networkx.algorithms.assortativity.connectivity.k_nearest_neighbors(G)
centrality = networkx.algorithms.centrality.degree_alg.degree_centrality(G)
closeness = networkx.algorithms.centrality.closeness.closeness_centrality(G)
betweenness = networkx.algorithms.centrality.betweenness.betweenness_centrality(G)
communicability = networkx.algorithms.centrality.communicability_alg.communicability(G)
estrada = networkx.algorithms.centrality.communicability_alg.estrada_index(G)
shortest = networkx.algorithms.shortest_paths.shortest_path_length(G)
# Of all possible edges in the graph, find the percent of missing edges
missing_edges = 1 - ( n_edges / triangular_number(n_nodes) )
return (n_nodes, n_edges, len(min_dom_set), len(min_edge_dom_set), ramsey, assortativity,
pearson, average_connectivity, knn, centrality, closeness, betweenness, communicability,
estrada, shortest, missing_edges)
def retrieve_graph_attributes(G):
"""Read in a graph and return an array of graph statistics"""
n_nodes = networkx.number_of_nodes(G)
n_edges = networkx.number_of_edges(G)
min_dom_set = min_weighted_dominating_set(G)
min_edge_dom_set = min_edge_dominating_set(G)
ramsey = networkx.algorithms.approximation.ramsey.ramsey_R2(G)
assortativity = networkx.algorithms.assortativity.degree_assortativity_coefficient(
G)
pearson = networkx.algorithms.assortativity.degree_pearson_correlation_coefficient(
G)
neighbor = networkx.algorithms.assortativity.neighbor_degree.average_neighbor_degree(
G)
average_connectivity = networkx.algorithms.assortativity.average_degree_connectivity(
G)
knn = networkx.algorithms.assortativity.connectivity.k_nearest_neighbors(G)
centrality = networkx.algorithms.centrality.degree_alg.degree_centrality(G)
closeness = networkx.algorithms.centrality.closeness.closeness_centrality(
G)
betweenness = networkx.algorithms.centrality.betweenness.betweenness_centrality(
G)
communicability = networkx.algorithms.centrality.communicability_alg.communicability(
G)
estrada = networkx.algorithms.centrality.communicability_alg.estrada_index(
G)
shortest = networkx.algorithms.shortest_paths.shortest_path_length(G)
# Of all possible edges in the graph, find the percent of missing edges
missing_edges = 1 - (n_edges / triangular_number(n_nodes))
return (n_nodes, n_edges, len(min_dom_set), len(min_edge_dom_set), ramsey,
assortativity, pearson, average_connectivity, knn, centrality,
closeness, betweenness, communicability, estrada, shortest,
missing_edges)
def compute_summaries(G):
""" Compute network features, computational times and their nature.
Evaluate 54 summary statistics of a network G, plus 4 noise variables,
store the computational time to evaluate each summary statistic, and keep
track of their nature (discrete or not).
Args:
G (networkx.classes.graph.Graph):
an undirected networkx graph.
Returns:
resDicts (tuple):
a tuple containing the elements:
- dictSums (dict): a dictionary with the name of the summaries
as keys and the summary statistic values as values;
- dictTimes (dict): a dictionary with the name of the summaries
as keys and the time to compute each one as values;
- dictIsDist (dict): a dictionary indicating if the summary is
discrete (True) or not (False).
"""
dictSums = dict() # Will store the summary statistic values
dictTimes = dict() # Will store the evaluation times
dictIsDisc = dict() # Will store the summary statistic nature
# Extract the largest connected component
Gcc = sorted(nx.connected_components(G), key=len, reverse=True)
G_lcc = G.subgraph(Gcc[0])
# Number of edges
start = time.time()
dictSums["num_edges"] = G.number_of_edges()
dictTimes["num_edges"] = time.time() - start
dictIsDisc["num_edges"] = True
# Number of connected components
start = time.time()
dictSums["num_of_CC"] = nx.number_connected_components(G)
dictTimes["num_of_CC"] = time.time() - start
dictIsDisc["num_of_CC"] = True
# Number of nodes in the largest connected component
start = time.time()
dictSums["num_nodes_LCC"] = nx.number_of_nodes(G_lcc)
dictTimes["num_nodes_LCC"] = time.time() - start
dictIsDisc["num_nodes_LCC"] = True
# Number of edges in the largest connected component
start = time.time()
dictSums["num_edges_LCC"] = G_lcc.number_of_edges()
dictTimes["num_edges_LCC"] = time.time() - start
dictIsDisc["num_edges_LCC"] = True
# Diameter of the largest connected component
start = time.time()
dictSums["diameter_LCC"] = nx.diameter(G_lcc)
dictTimes["diameter_LCC"] = time.time() - start
dictIsDisc["diameter_LCC"] = True
# Average geodesic distance (shortest path length in the LCC)
start = time.time()
dictSums["avg_geodesic_dist_LCC"] = nx.average_shortest_path_length(G_lcc)
dictTimes["avg_geodesic_dist_LCC"] = time.time() - start
dictIsDisc["avg_geodesic_dist_LCC"] = False
# Average degree of the neighborhood of each node
start = time.time()
dictSums["avg_deg_connectivity"] = np.mean(
list(nx.average_degree_connectivity(G).values()))
dictTimes["avg_deg_connectivity"] = time.time() - start
dictIsDisc["avg_deg_connectivity"] = False
# Average degree of the neighbors of each node in the LCC
start = time.time()
dictSums["avg_deg_connectivity_LCC"] = np.mean(
list(nx.average_degree_connectivity(G_lcc).values()))
dictTimes["avg_deg_connectivity_LCC"] = time.time() - start
dictIsDisc["avg_deg_connectivity_LCC"] = False
# Recover the degree distribution
start_degree_extract = time.time()
degree_vals = list(dict(G.degree()).values())
degree_extract_time = time.time() - start_degree_extract
# Entropy of the degree distribution
start = time.time()
dictSums["degree_entropy"] = ss.entropy(degree_vals)
dictTimes["degree_entropy"] = time.time() - start + degree_extract_time
dictIsDisc["degree_entropy"] = False
# Maximum degree
start = time.time()
dictSums["degree_max"] = max(degree_vals)
dictTimes["degree_max"] = time.time() - start + degree_extract_time
dictIsDisc["degree_max"] = True
# Average degree
start = time.time()
dictSums["degree_mean"] = np.mean(degree_vals)
dictTimes["degree_mean"] = time.time() - start + degree_extract_time
dictIsDisc["degree_mean"] = False
# Median degree
start = time.time()
dictSums["degree_median"] = np.median(degree_vals)
dictTimes["degree_median"] = time.time() - start + degree_extract_time
dictIsDisc["degree_median"] = False
# Standard deviation of the degree distribution
start = time.time()
dictSums["degree_std"] = np.std(degree_vals)
dictTimes["degree_std"] = time.time() - start + degree_extract_time
dictIsDisc["degree_std"] = False
# Quantile 25%
start = time.time()
dictSums["degree_q025"] = np.quantile(degree_vals, 0.25)
dictTimes["degree_q025"] = time.time() - start + degree_extract_time
dictIsDisc["degree_q025"] = False
# Quantile 75%
start = time.time()
dictSums["degree_q075"] = np.quantile(degree_vals, 0.75)
dictTimes["degree_q075"] = time.time() - start + degree_extract_time
dictIsDisc["degree_q075"] = False
# Average geodesic distance
start = time.time()
dictSums["avg_shortest_path_length_LCC"] = nx.average_shortest_path_length(
G_lcc)
dictTimes["avg_shortest_path_length_LCC"] = time.time() - start
dictIsDisc["avg_shortest_path_length_LCC"] = False
# Average global efficiency:
# The efficiency of a pair of nodes in a graph is the multiplicative
# inverse of the shortest path distance between the nodes.
# The average global efficiency of a graph is the average efficiency of
# all pairs of nodes.
start = time.time()
dictSums["avg_global_efficiency"] = nx.global_efficiency(G)
dictTimes["avg_global_efficiency"] = time.time() - start
dictIsDisc["avg_global_efficiency"] = False
# Harmonic mean which is 1/avg_global_efficiency
start = time.time()
dictSums["harmonic_mean"] = nx.global_efficiency(G)
dictTimes["harmonic_mean"] = time.time() - start
dictIsDisc["harmonic_mean"] = False
# Average local efficiency
# The local efficiency of a node in the graph is the average global
# efficiency of the subgraph induced by the neighbors of the node.
# The average local efficiency is the average of the
# local efficiencies of each node.
start = time.time()
dictSums["avg_local_efficiency_LCC"] = nx.local_efficiency(G_lcc)
dictTimes["avg_local_efficiency_LCC"] = time.time() - start
dictIsDisc["avg_local_efficiency_LCC"] = False
# Node connectivity
# The node connectivity is equal to the minimum number of nodes that
# must be removed to disconnect G or render it trivial.
# Only on the largest connected component here.
start = time.time()
dictSums["node_connectivity_LCC"] = nx.node_connectivity(G_lcc)
dictTimes["node_connectivity_LCC"] = time.time() - start
dictIsDisc["node_connectivity_LCC"] = True
# Edge connectivity
# The edge connectivity is equal to the minimum number of edges that
# must be removed to disconnect G or render it trivial.
# Only on the largest connected component here.
start = time.time()
dictSums["edge_connectivity_LCC"] = nx.edge_connectivity(G_lcc)
dictTimes["edge_connectivity_LCC"] = time.time() - start
dictIsDisc["edge_connectivity_LCC"] = True
# Graph transitivity
# 3*times the number of triangles divided by the number of triades
start = time.time()
dictSums["transitivity"] = nx.transitivity(G)
dictTimes["transitivity"] = time.time() - start
dictIsDisc["transitivity"] = False
# Number of triangles
start = time.time()
dictSums["num_triangles"] = np.sum(list(nx.triangles(G).values())) / 3
dictTimes["num_triangles"] = time.time() - start
dictIsDisc["num_triangles"] = True
# Estimate of the average clustering coefficient of G:
# Average local clustering coefficient, with local clustering coefficient
# defined as C_i = (nbr of pairs of neighbors of i that are connected)/(nbr of pairs of neighbors of i)
start = time.time()
dictSums["avg_clustering_coef"] = nx.average_clustering(G)
dictTimes["avg_clustering_coef"] = time.time() - start
dictIsDisc["avg_clustering_coef"] = False
# Square clustering (averaged over nodes):
# the fraction of possible squares that exist at the node.
# We average it over nodes
start = time.time()
dictSums["square_clustering_mean"] = np.mean(
list(nx.square_clustering(G).values()))
dictTimes["square_clustering_mean"] = time.time() - start
dictIsDisc["square_clustering_mean"] = False
# We compute the median
start = time.time()
dictSums["square_clustering_median"] = np.median(
list(nx.square_clustering(G).values()))
dictTimes["square_clustering_median"] = time.time() - start
dictIsDisc["square_clustering_median"] = False
# We compute the standard deviation
start = time.time()
dictSums["square_clustering_std"] = np.std(
list(nx.square_clustering(G).values()))
dictTimes["square_clustering_std"] = time.time() - start
dictIsDisc["square_clustering_std"] = False
# Number of 2-cores
start = time.time()
dictSums["num_2cores"] = len(nx.k_core(G, k=2))
dictTimes["num_2cores"] = time.time() - start
dictIsDisc["num_2cores"] = True
# Number of 3-cores
start = time.time()
dictSums["num_3cores"] = len(nx.k_core(G, k=3))
dictTimes["num_3cores"] = time.time() - start
dictIsDisc["num_3cores"] = True
# Number of 4-cores
start = time.time()
dictSums["num_4cores"] = len(nx.k_core(G, k=4))
dictTimes["num_4cores"] = time.time() - start
dictIsDisc["num_4cores"] = True
# Number of 5-cores
start = time.time()
dictSums["num_5cores"] = len(nx.k_core(G, k=5))
dictTimes["num_5cores"] = time.time() - start
dictIsDisc["num_5cores"] = True
# Number of 6-cores
start = time.time()
dictSums["num_6cores"] = len(nx.k_core(G, k=6))
dictTimes["num_6cores"] = time.time() - start
dictIsDisc["num_6cores"] = True
# Number of k-shells
# The k-shell is the subgraph induced by nodes with core number k.
# That is, nodes in the k-core that are not in the k+1-core
# Number of 2-shells
start = time.time()
dictSums["num_2shells"] = len(nx.k_shell(G, 2))
dictTimes["num_2shells"] = time.time() - start
dictIsDisc["num_2shells"] = True
# Number of 3-shells
start = time.time()
dictSums["num_3shells"] = len(nx.k_shell(G, 3))
dictTimes["num_3shells"] = time.time() - start
dictIsDisc["num_3shells"] = True
# Number of 4-shells
start = time.time()
dictSums["num_4shells"] = len(nx.k_shell(G, 4))
dictTimes["num_4shells"] = time.time() - start
dictIsDisc["num_4shells"] = True
# Number of 5-shells
start = time.time()
dictSums["num_5shells"] = len(nx.k_shell(G, 5))
dictTimes["num_5shells"] = time.time() - start
dictIsDisc["num_5shells"] = True
# Number of 6-shells
start = time.time()
dictSums["num_6shells"] = len(nx.k_shell(G, 6))
dictTimes["num_6shells"] = time.time() - start
dictIsDisc["num_6shells"] = True
start = time.time()
listOfCliques = list(nx.enumerate_all_cliques(G))
enum_all_cliques_time = time.time() - start
# Number of 4-cliques
start = time.time()
n4Clique = 0
for li in listOfCliques:
if len(li) == 4:
n4Clique += 1
dictSums["num_4cliques"] = n4Clique
dictTimes["num_4cliques"] = time.time() - start + enum_all_cliques_time
dictIsDisc["num_4cliques"] = True
# Number of 5-cliques
start = time.time()
n5Clique = 0
for li in listOfCliques:
if len(li) == 5:
n5Clique += 1
dictSums["num_5cliques"] = n5Clique
dictTimes["num_5cliques"] = time.time() - start + enum_all_cliques_time
dictIsDisc["num_5cliques"] = True
# Maximal size of a clique in the graph
start = time.time()
dictSums["max_clique_size"] = len(approximation.clique.max_clique(G))
dictTimes["max_clique_size"] = time.time() - start
dictIsDisc["max_clique_size"] = True
# Approximated size of a large clique in the graph
start = time.time()
dictSums["large_clique_size"] = approximation.large_clique_size(G)
dictTimes["large_clique_size"] = time.time() - start
dictIsDisc["large_clique_size"] = True
# Number of shortest path of size k
start = time.time()
listOfPLength = list(nx.shortest_path_length(G))
path_length_time = time.time() - start
# when k = 3
start = time.time()
n3Paths = 0
for node in G.nodes():
tmp = list(listOfPLength[node][1].values())
n3Paths += tmp.count(3)
dictSums["num_shortest_3paths"] = n3Paths / 2
dictTimes["num_shortest_3paths"] = time.time() - start + path_length_time
dictIsDisc["num_shortest_3paths"] = True
# when k = 4
start = time.time()
n4Paths = 0
for node in G.nodes():
tmp = list(listOfPLength[node][1].values())
n4Paths += tmp.count(4)
dictSums["num_shortest_4paths"] = n4Paths / 2
dictTimes["num_shortest_4paths"] = time.time() - start + path_length_time
dictIsDisc["num_shortest_4paths"] = True
# when k = 5
start = time.time()
n5Paths = 0
for node in G.nodes():
tmp = list(listOfPLength[node][1].values())
n5Paths += tmp.count(5)
dictSums["num_shortest_5paths"] = n5Paths / 2
dictTimes["num_shortest_5paths"] = time.time() - start + path_length_time
dictIsDisc["num_shortest_5paths"] = True
# when k = 6
start = time.time()
n6Paths = 0
for node in G.nodes():
tmp = list(listOfPLength[node][1].values())
n6Paths += tmp.count(6)
dictSums["num_shortest_6paths"] = n6Paths / 2
dictTimes["num_shortest_6paths"] = time.time() - start + path_length_time
dictIsDisc["num_shortest_6paths"] = True
# Size of the minimum (weight) node dominating set:
# A subset of nodes where each node not in the subset has for direct
# neighbor a node of the dominating set.
start = time.time()
T = approximation.min_weighted_dominating_set(G)
dictSums["size_min_node_dom_set"] = len(T)
dictTimes["size_min_node_dom_set"] = time.time() - start
dictIsDisc["size_min_node_dom_set"] = True
# Idem but with the edge dominating set
start = time.time()
T = approximation.min_edge_dominating_set(G)
dictSums["size_min_edge_dom_set"] = 2 * len(
T) # times 2 to have a number of nodes
dictTimes["size_min_edge_dom_set"] = time.time() - start
dictIsDisc["size_min_edge_dom_set"] = True
# The Wiener index of a graph is the sum of the shortest-path distances
# between each pair of reachable nodes. For pairs of nodes in undirected graphs,
# only one orientation of the pair is counted.
# (On LCC otherwise inf)
start = time.time()
dictSums["wiener_index_LCC"] = nx.wiener_index(G_lcc)
dictTimes["wiener_index_LCC"] = time.time() - start
dictIsDisc["wiener_index_LCC"] = True
# Betweenness node centrality (averaged over nodes):
# at node u it is defined as B_u = sum_i,j sigma(i,u,j)/sigma(i,j)
# where sigma is the number of shortest path between i and j going through u or not
start = time.time()
betweenness = list(nx.betweenness_centrality(G).values())
time_betweenness = time.time() - start
# Averaged across nodes
start = time.time()
dictSums["betweenness_centrality_mean"] = np.mean(betweenness)
dictTimes["betweenness_centrality_mean"] = time.time(
) - start + time_betweenness
dictIsDisc["betweenness_centrality_mean"] = False
# Maximum across nodes
start = time.time()
dictSums["betweenness_centrality_max"] = max(betweenness)
dictTimes["betweenness_centrality_max"] = time.time(
) - start + time_betweenness
dictIsDisc["betweenness_centrality_max"] = False
# Central point dominance
# CPD = sum_u(B_max - B_u)/(N-1)
start = time.time()
dictSums["central_point_dominance"] = sum(
max(betweenness) - np.array(betweenness)) / (len(betweenness) - 1)
dictTimes["central_point_dominance"] = time.time(
) - start + time_betweenness
dictIsDisc["central_point_dominance"] = False
# Estrata index : sum_i^n exp(lambda_i)
# with n the number of nodes, lamda_i the i-th eigen value of the adjacency matrix of G
start = time.time()
dictSums["Estrata_index"] = nx.estrada_index(G)
dictTimes["Estrata_index"] = time.time() - start
dictIsDisc["Estrata_index"] = False
# Eigenvector centrality
# For each node, it is the average eigenvalue centrality of its neighbors,
# where centrality of node i is taken as the i-th coordinate of x
# such that Ax = lambda*x (for the maximal eigen value)
# Averaged
start = time.time()
dictSums["avg_eigenvec_centrality"] = np.mean(
list(nx.eigenvector_centrality_numpy(G).values()))
dictTimes["avg_eigenvec_centrality"] = time.time() - start
dictIsDisc["avg_eigenvec_centrality"] = False
# Maximum
start = time.time()
dictSums["max_eigenvec_centrality"] = max(
list(nx.eigenvector_centrality_numpy(G).values()))
dictTimes["max_eigenvec_centrality"] = time.time() - start
dictIsDisc["max_eigenvec_centrality"] = False
### Noise generation ###
# Noise simulated from a Normal(0,1) distribution
start = time.time()
dictSums["noise_Gauss"] = ss.norm.rvs(0, 1)
dictTimes["noise_Gauss"] = time.time() - start
dictIsDisc["noise_Gauss"] = False
# Noise simulated from a Uniform distribution [0-50]
start = time.time()
dictSums["noise_Unif"] = ss.uniform.rvs(0, 50)
dictTimes["noise_Unif"] = time.time() - start
dictIsDisc["noise_Unif"] = False
# Noise simulated from a Bernoulli B(0.5) distribution
start = time.time()
dictSums["noise_Bern"] = ss.bernoulli.rvs(0.5)
dictTimes["noise_Bern"] = time.time() - start
dictIsDisc["noise_Bern"] = True
# Noise simulated from a discrete uniform distribution [0,50[
start = time.time()
dictSums["noise_disc_Unif"] = ss.randint.rvs(0, 50)
dictTimes["noise_disc_Unif"] = time.time() - start
dictIsDisc["noise_disc_Unif"] = True
resDicts = (dictSums, dictTimes, dictIsDisc)
return resDicts
def create_graph(edgelist):
G = nx.DiGraph()
G.add_weighted_edges_from(edgelist)
# Graph layout
pos = nx.spring_layout(G)
# Remove nodes with out-degree = 1------------------------------------------
outdeg = nx.degree(G)
while 1 in outdeg.values():
to_remove = [n for n in outdeg if outdeg[n] == 1]
G.remove_nodes_from(to_remove)
# Remove edges also checking the two values of the tuple
edgelist = [x for x in edgelist if x[0] not in to_remove]
edgelist = [x for x in edgelist if x[1] not in to_remove]
outdeg = nx.degree(G)
# ---------------------------------------------------------------------------
# Node size ================================================================
d = nx.degree(G)
nx.draw_networkx_nodes(G, pos, nodelist=d.keys(), node_size=[v * 100 for v in d.values()])
# ===========================================================================
nx.draw_networkx_nodes(G, pos, cmap=plt.get_cmap("jet"), node_size=1)
edge_labels = dict([((u, v), d["weight"]) for u, v, d in G.edges(data=True)])
nx.draw_networkx_edges(G, pos, width=0.5, alpha=0.2)
nx.draw_networkx_edge_labels(G, pos, edge_labels=edge_labels)
labels = {}
for node in G.nodes():
labels[node] = node
nx.draw_networkx_labels(G, pos, labels, font_size=8)
if not len(G) == 1:
from networkx.algorithms import approximation as apxm
print apxm.min_edge_dominating_set(G)
print apxm.max_clique(G)
print apxm.maximum_independent_set(G)
print apxm.min_maximal_matching(G)
print nx.degree_centrality(G)
plt.savefig("directed.png") # save as png
# clean old graph
plt.clf()
