def measure_between_group_connectivity(induced, name = '', plots = False):
overall_conn = approx.node_connectivity(induced)
print('Minimum number of nodes that must be removed to disconnect studets: '+str(overall_conn))
pairwise_conn = approx.all_pairs_node_connectivity(induced)
avg_cluster = approx.average_clustering(induced)
print('Mean of the fraction of triangles that actually exist over all possible triangles in each neighborhood: '+str(avg_cluster))
avg_cluster = approx.average_clustering(induced)
print('Mean of the fraction of triangles that actually exist over all possible triangles in each neighborhood: '+str(avg_cluster))
if plots:
p.pairwise_conn_dist(pairwise_conn, name)
def build_graph():
"""generate graph representation of the training facts
"""
# G = nx.DiGraph()
G = nx.Graph()
with open('data/initial_label_facts.json') as f_in:
label_triples = json.load(f_in)
rel_colors = {
"use":"red",
"different from":'green',
"subclass of":'blue',
"has quality":'purple',
"instance of":'yellow',
"facet of":'brown'}
for rel, vals in label_triples.items():
for pair in vals:
G.add_edge(pair[0], pair[1], color=rel_colors.get(rel, 'black'))
# print(len(list(nx.connected_components(G))))
# print(sorted(d for n, d in G.degree()))
# print(nx.clustering(G))
print(approximation.average_clustering(G))
# write_dot(G, 'test.dot')
# $ dot -Tpng test.dot >test.png
return label_triples
def measure_connectivity(G,grouping = None):
overall_conn = approx.node_connectivity(G)
print('Minimum number of nodes that must be removed to disconnect studets: '+str(overall_conn))
pairwise_conn = approx.all_pairs_node_connectivity(G)
plt.title('Distribution of Min Removed Nodes to Disconnect Pair')
plt.hist(pairwise_conn.values())
avg_cluster = approx.average_clustering(G)
print('Mean of the fraction of triangles that actually exist over all possible triangles in each neighborhood: '+str(avg_cluster))
def _extract_features_for_subgraph(self, graph):
res = {}
deg_list = [i[1] for i in nx.degree(graph)]
weights_list = [
graph[edge[0]][edge[1]]['weight'] for edge in graph.edges
]
res['connected'] = [1 if nx.is_connected(graph) else 0]
res['density'] = ['{:.6f}'.format(nx.density(graph))]
res['Avg_CC'] = [aprox.average_clustering(graph)]
res['Median_deg'] = ['{:.6f}'.format(np.median(deg_list))]
res['Variance_deg'] = ['{:.6f}'.format(np.var(deg_list))]
res['Median_wights'] = [
'{:.6f}'.format(
np.median(weights_list) if len(weights_list) > 0 else -1)
]
res['Variance_wights'] = [
'{:.6f}'.format(
np.var(weights_list) if len(weights_list) > 0 else 0)
]
res['Avg_degree'] = [
'{:.6f}'.format(sum(deg_list) / len(nx.degree(graph)))
]
res['Avg_weight'] = [
'{:.6f}'.format(
sum(weights_list) /
len(weights_list) if len(weights_list) > 0 else -1)
]
res['Avg_weight_abs'] = [
'{:.6f}'.format(
abs(
sum(weights_list) /
len(weights_list) if len(weights_list) > 0 else -1))
]
res['edges'] = [len(graph.edges)]
res['nodes'] = [len(graph.nodes)]
res['self_loops'] = [len(list(nx.nodes_with_selfloops(graph)))]
res['edge_to_node_ratio'] = [
'{:.6f}'.format(
len(graph.nodes) /
len(graph.edges) if len(graph.edges) > 0 else len(graph.nodes))
]
res['negative_edges'] = [
len([
edge for edge in graph.edges
if graph[edge[0]][edge[1]]['weight'] < 0
])
]
res['Num_of_zero_weights'] = [
len([
e for e in graph.edges
if 0.005 > abs(graph[e[0]][e[1]]['weight'] > 0)
])
]
res['min_vc'] = [len(aprox.min_weighted_vertex_cover(graph))]
for key in res.keys():
res[key] = [float(res[key][0])]
return res
def main():
fb = read_graph('facebook_combined.txt.gz')
fb_clustering = average_clustering(fb)
fb_length = estimate_path_length(fb)
n = len(fb)
m = len(fb.edges())
k = int(round(m / n))
hk = hk_graph_modified(n, 1)
# generate_pmf(fb, hk)
# generate_cdf(fb, hk)
generate_ccdf(fb, hk)
print("Degrees:", len(degrees(fb)), len(degrees(hk)))
print("Clustering:", fb_clustering, average_clustering(hk))
print("Path length:", fb_length, estimate_path_length(hk))
print("Mean degrees:", np.mean(degrees(fb)), np.mean(degrees(hk)))
def analyze_clustering(G):
average_clustering_coefficient = approximation.average_clustering(G)
average_clustering = nx.average_clustering(G)
average_shortest_path_length = nx.average_shortest_path_length(G)
local_efficiency = nx.local_efficiency(G)
global_efficiency = nx.global_efficiency(G)
table = prettytable.PrettyTable(
['Average clustering', 'Average clustering coefficient', 'Average shortest path length'])
table.add_row([average_clustering, average_clustering_coefficient, average_shortest_path_length])
print(table)
table = prettytable.PrettyTable(['Local efficiency', 'Global efficiency'])
table.add_row([local_efficiency, global_efficiency])
print(table)
def extract_graph_features(self, graph):
"""
ref: https://networkx.github.io/documentation/stable/_modules/networkx/algorithms/approximation/vertex_cover.html
ref: https://networkx.github.io/documentation/stable/reference/algorithms/approximation.html#module-networkx.algorithms.approximation
"""
res = {}
deg_list = [i[1] for i in nx.degree(graph)]
weights_list = [
graph[edge[0]][edge[1]]['weight'] for edge in graph.edges
]
if len(weights_list) == 0:
return None
# try:
# weights_list = [graph[edge[0]][edge[1]]['weight'] for edge in graph.edges]
# except:
# return None
res['connected'] = 1 if nx.is_connected(graph) else 0
res['density'] = '{:.6f}'.format(nx.density(graph))
res['Avg_CC'] = aprox.average_clustering(graph)
res['Median_deg'] = '{:.6f}'.format(np.median(deg_list))
res['Variance_deg'] = '{:.6f}'.format(np.var(deg_list))
res['Median_wights'] = '{:.6f}'.format(np.median(weights_list))
res['Variance_wights'] = '{:.6f}'.format(np.var(weights_list))
res['Avg_degree'] = '{:.6f}'.format(
sum(deg_list) / len(nx.degree(graph)))
res['Avg_weight'] = '{:.6f}'.format(
sum(weights_list) / len(weights_list))
res['Avg_weight_abs'] = '{:.6f}'.format(
abs(sum(weights_list) / len(weights_list)))
res['edges'] = len(graph.edges)
res['nodes'] = len(graph.nodes)
res['self_loops'] = len(list(nx.nodes_with_selfloops(graph)))
res['edge_to_node_ratio'] = '{:.6f}'.format(
len(graph.nodes) / len(graph.edges))
res['negative_edges'] = len([
edge for edge in graph.edges
if graph[edge[0]][edge[1]]['weight'] < 0
])
res['Num_of_zero_weights'] = len([
e for e in graph.edges
if 0.005 > abs(graph[e[0]][e[1]]['weight'] > 0)
])
res['min_vc'] = len(aprox.min_weighted_vertex_cover(graph))
return res
def analyze_graph(G):
G.graph['directed'] = nx.is_directed(G)
G_und = G.to_undirected()
G.graph['connected_components'] = nx.number_connected_components(G_und)
G.graph['largest_component'] = len(
max(nx.connected_components(G_und), key=len))
logging.info("Graph ID {}: components analyzed.".format(
G.graph['graph_id']))
G.graph['average_clustering'] = approximation.average_clustering(G_und)
logging.info("Graph ID {}: clustering analyzed.".format(
G.graph['graph_id']))
degrees = [d for n, d in G.degree()]
G.graph['min_degree'] = min(degrees), max(degrees), np.mean(
degrees), np.median(degrees)
G.graph['max_degree'] = max(degrees)
G.graph['avg_degree'] = np.mean(degrees)
G.graph['std_degree'] = np.std(degrees)
G.graph['median_degree'] = np.median(degrees)
logging.info("Graph ID {}: degrees analyzed.".format(
G.graph['graph_id']))
def analyze_graph(G, verbose=False):
"""Analyzes 'G' for relevant characteristics.
G: NetworkX graph
verbose: print characteristics if True
return:
n: number of nodes in G
m: number of edges in G
k: int of average degree (edges per node)
degs: list of degrees of G
"""
n = len(G)
m = len(G.edges())
k = int(round(m/n))
C = average_clustering(G)
L = estimate_path_length(G)
degs = degrees(G)
if verbose:
print('n: %i m: %i k: %i' %(n,m,k))
print('clustering: ',C, 'path length: ',L)
print('average degree: %.2f degree variance: %.2f'
%(np.mean(degs), np.var(degs)))
return n, m, k, degs
def measure_connectivity(G, out, grouping=None):
# overall_conn = approx.node_connectivity(G)
# print('Minimum number of nodes that must be removed to disconnect studets: '+str(overall_conn))
pairwise_conn = approx.all_pairs_node_connectivity(G)
plt.title('Distribution of Min Removed Nodes to Disconnect Pair')
connlist = []
for subdict in pairwise_conn.values():
avgconn = sum(list(subdict.values())) / len(subdict.values())
connlist.append(avgconn)
temp = pd.DataFrame(connlist, columns=['pairwise_conn'])
temp = temp.reset_index()
temp["rank"] = temp['pairwise_conn'].rank(method='average',
ascending=False)
temp = temp.sort_values(by=['index'], ascending=False)
plt.scatter(y=temp['pairwise_conn'], x=temp['rank'], alpha=0.7)
plt.xlabel('Rank')
plt.ylabel('Count')
plt.savefig(out + 'zipf_pairwiseconnrank.png')
plt.close()
avg_cluster = approx.average_clustering(G)
print(
'Mean of the fraction of triangles that actually exist over all possible triangles in each neighborhood: '
+ str(avg_cluster))
def test_complete():
G = nx.complete_graph(5)
assert average_clustering(G, trials=int(len(G) / 2)) == 1
G = nx.complete_graph(7)
assert average_clustering(G, trials=int(len(G) / 2)) == 1
def test_empty():
G = nx.empty_graph(5)
assert average_clustering(G, trials=int(len(G) / 2)) == 0
def test_dodecahedral():
# Actual coefficient is 0
G = nx.dodecahedral_graph()
assert (average_clustering(G, trials=int(len(G) / 2)) ==
nx.average_clustering(G))
return lengths
def estimate_path_length(G, nodes=None, trials=1000):
""" estimates the average shortest path length of a given Graph """
return np.mean(sample_path_lengths(G, nodes, trials))
fb = read_graph('facebook_combined.txt.gz')
n = len(fb)
m = len(fb.edges())
k_fb = int(round(2 * m / n)) # I don't get why we double count the edges, but
print(f"fb Graph nodes = {n}, edges = {m}")
C_fb = average_clustering(fb)
L_fb = estimate_path_length(fb)
""" now construct a WS graph with the same n, k; Downey figured out by trial
and error that when p = 0.05, the C and Ls are comparable. """
ws = nx.watts_strogatz_graph(n, k_fb, 0.05, seed=15)
print(f"Constructing Watts-Strogatz Graph, WS({n}, {k_fb}, 0.05)")
C_ws = average_clustering(ws)
L_ws = estimate_path_length(ws)
print("graph \t n \t C \t L \t mu_k \t sigma_k")
print(f"fb \t {n} \t {C_fb} \t {L_fb} \t "
f"{np.mean(degrees(fb)):.1f} \t {np.std(degrees(fb)):.1f}")
print(f"WS \t {n} \t {C_ws} \t {L_ws} \t "
f"{np.mean(degrees(ws)):.1f} \t {np.std(degrees(ws)):.1f}")
""" now use probability mass function objects to check the probability that a
node has a particular degree """
def test_dodecahedral():
# Actual coefficient is 0
G = nx.dodecahedral_graph()
assert_equal(average_clustering(G, trials=int(len(G) / 2)),
nx.average_clustering(G))
def average_clustering(graph: nx.Graph, approximate=True):
# only for undirected graphs
if approximate:
return approximation.average_clustering(graph)
return cluster.average_clustering(graph)
def ver_medidas(G):
print(function.info(G))
"""
Numero minimo de nodos que deben ser removidos para desconectar G
"""
print("Numero minimo de nodos que deben ser removidos para desconectar G :"+str(approximation.node_connectivity(G)))
"""
average clustering coefficient of G.
"""
print("average clustering coefficient of G: "+str(approximation.average_clustering(G)))
"""
Densidad de un Grafo
"""
print("Densidad de G: "+str(function.density(G)))
"""
Assortativity measures the similarity of connections in
the graph with respect to the node degree.
Valores positivos de r indican que existe una correlacion entre nodos
con grado similar, mientras que un valor negativo indica
correlaciones entre nodos de diferente grado
"""
print("degree assortativity:"+str(assortativity.degree_assortativity_coefficient(G)))
"""
Assortativity measures the similarity of connections
in the graph with respect to the given attribute.
"""
print("assortativity for node attributes: "+str(assortativity.attribute_assortativity_coefficient(G,"crime")))
"""
Grado promedio vecindad
"""
plt.plot(assortativity.average_neighbor_degree(G).values())
plt.title("Grado promedio vecindad")
plt.xlabel("Nodo")
plt.ylabel("Grado")
plt.show();
"""
Grado de Centralidad de cada nodo
"""
plt.plot(centrality.degree_centrality(G).values())
plt.title("Grado de centralidad")
plt.xlabel("Nodo")
plt.ylabel("Centralidad")
plt.show();
"""
Calcular el coeficiente de agrupamiento para nodos
"""
plt.plot(cluster.clustering(G).values())
plt.title("coeficiente de agrupamiento")
plt.xlabel("Nodo")
plt.show();
"""
Media coeficiente de Agrupamiento
"""
print("Coeficiente de agrupamiento de G:"+str(cluster.average_clustering(G)))
"""
Centro del grafo
El centro de un grafo G es el subgrafo inducido por el
conjunto de vertices de excentricidad minima.
La excentricidad de v in V se define como la
distancia maxima desde v a cualquier otro vertice del
grafo G siguiendo caminos de longitud minima.
"""
print("Centro de G:"+ str(distance_measures.center(G)))
"""
Diametro de un grafo
The diameter is the maximum eccentricity.
"""
print("Diametro de G:"+str(distance_measures.diameter(G)))
"""
Excentricidad de cada Nodo
The eccentricity of a node v is the maximum distance
from v to all other nodes in G.
"""
plt.plot(distance_measures.eccentricity(G).values())
plt.title("Excentricidad de cada Nodo")
plt.xlabel("Nodo")
plt.show();
"""
Periferia
The periphery is the set of nodes with eccentricity equal to the diameter.
"""
print("Periferia de G:")
print(distance_measures.periphery(G))
"""
Radio
The radius is the minimum eccentricity.
"""
print("Radio de G:"+str(distance_measures.radius(G)))
"""
PageRank calcula una clasificacion de los nodos
en el grafico G en funcion de la estructura de
los enlaces entrantes. Originalmente fue disenado
como un algoritmo para clasificar paginas web.
"""
plt.plot(link_analysis.pagerank_alg.pagerank(G).values())
plt.title("Puntaje de cada Nodo")
plt.xlabel("Nodo")
plt.show();
"""
Coeficiente de Small World.
A graph is commonly classified as small-world if sigma>1.
"""
print("Coeficiente de Small World: " + str(smallworld.sigma(G)))
"""
The small-world coefficient (omega) ranges between -1 and 1.
Values close to 0 means the G features small-world characteristics.
Values close to -1 means G has a lattice shape whereas values close
to 1 means G is a random graph.
"""
print("Omega coeficiente: "+str(smallworld.omega(G)))
edgecnt = G.number_of_edges()
print('{} has {} nodes in its network.'.format(that_year, nodecnt))
print('{} has {} edges in its network.'.format(that_year, edgecnt))
###GRAPH DENSITY
den = nx.density(G)
print('{} has a density of {}.'.format(that_year, round(den, 5)))
###DEGREE ASSORTATIVITY
deg_assort = nx.degree_assortativity_coefficient(G)
print('{} has a degree assortativity of {}.'.format(
that_year, round(deg_assort, 5)))
###APPROXIMATE AVERAGE CLUSTERING COEFFICIENT###
avg_clust = approx.average_clustering(G, trials=10000)
print('{} has an average clustering of {}.'.format(that_year, avg_clust))
###DEGREE OF FRAGMENTATION--BORGATTI'S KEY PLAYER PROBLEM###
denominator = nodecnt * (nodecnt - 1)
sum_of_distance = 0.0
distance_matrix = nx.all_pairs_shortest_path_length(G)
for source, destinations in distance_matrix:
for destination, length in destinations.items():
if not source == destination:
sum_of_distance += 1 / length
degree_of_fragmentation = (2 * sum_of_distance) / denominator
print('{} has a degree of fragmentation of {}.'.format(
that_year, degree_of_fragmentation))
###NUMBER OF CLIQUES###
return lengths
def estimate_path_length(G, nodes=None, trials=1000):
return np.mean(sample_path_lengths(G, nodes, trials))
def read_graph(filename):
G = nx.Graph()
array = np.loadtxt(filename, dtype=int)
G.add_edges_from(array)
return G
# https://snap.stanford.edu/data/facebook_combined.txt.gz
fb = read_graph('../../data/facebook_combined.txtz')
n, m = len(fb), len(fb.edges()) # (4039, 88234)
C = average_clustering(fb)
L = estimate_path_length(fb)
k = int(round(2 * m / n)) # 44
lattice = nx.watts_strogatz_graph(n, k, p=0)
len(lattice), len(lattice.edges())
# (4039, 88858)
C, average_clustering(lattice)
# (0.615, 0.747)
L, estimate_path_length(lattice)
# (3.717, 47.088)
def test_complete():
G = nx.complete_graph(5)
assert_equal(average_clustering(G, trials=int(len(G) / 2)), 1)
G = nx.complete_graph(7)
assert_equal(average_clustering(G, trials=int(len(G) / 2)), 1)
def test_petersen():
# Actual coefficient is 0
G = nx.petersen_graph()
assert_equal(average_clustering(G, trials=int(len(G) / 2)),
nx.average_clustering(G))
def test_complete():
G = nx.complete_graph(5)
assert_equal(average_clustering(G, trials=int(len(G) / 2)), 1)
G = nx.complete_graph(7)
assert_equal(average_clustering(G, trials=int(len(G) / 2)), 1)
def test_empty():
G = nx.empty_graph(5)
assert_equal(average_clustering(G, trials=int(len(G) / 2)), 0)
def test_petersen():
# Actual coefficient is 0
G = nx.petersen_graph()
assert (average_clustering(G, trials=int(len(G) / 2)) ==
nx.average_clustering(G))
annot = sep + "Metrics added {} from {}.py".format(now, this_file) + \
"\n\n" + sep
with open("network_metrics.txt", 'a') as metrics_file:
metrics_file.write(annot)
metrics_file.write(nx.info(latinx_g) + "\n\n")
metrics_file.write(nx.info(todes_g) + "\n\n")
# ---------------------------------------------------------------------------- #
# AVG CLUSTER COEFFICIENT
# ---------------------------------------------------------------------------- #
logging.info("calculating cluster coeff for each network")
todes_gu = nx.to_undirected(todes_g)
latinx_gu = nx.to_undirected(latinx_g)
t_cluster_coeff = appx.average_clustering(todes_gu, trials=10000, seed=115)
l_cluster_coeff = appx.average_clustering(latinx_gu, trials=10000, seed=115)
# write each cluster coefficient
with open("network_metrics.txt", 'a') as metrics_file:
metrics_file.write("Latinx network average cluster coeff: {} \n\n".format(
l_cluster_coeff))
metrics_file.write(
"Todes network average cluster coeff: {} \n\n".format(t_cluster_coeff))
# ---------------------------------------------------------------------------- #
# NETWORK DENSITY
# ---------------------------------------------------------------------------- #
logging.info("calculating network density")
with open("network_metrics.txt", 'a') as metrics_file:
metrics_file.write("Latinx Density: {}\n\n".format(nx.density(latinx_g)))
def test_tetrahedral():
# Actual coefficient is 1
G = nx.tetrahedral_graph()
assert average_clustering(G, trials=int(len(G) /
2)) == nx.average_clustering(G)
print("Instances topics\n")
for key, value in instance2topics.items():
print(key,value)
""" Assortativity """
r=nx.degree_pearson_correlation_coefficient(mastodon_digraph)
print(r)
""" Network Analysis Undirected """
mastodon_undirected = mastodon_digraph.to_undirected()
average_node_degree = nx.average_degree_connectivity(mastodon_undirected)
print ("average node degree", average_node_degree.keys())
average_clustering = approx.average_clustering(mastodon_undirected)
print("Average Clustering: ", average_clustering)
node_connectivity = approx.node_connectivity(mastodon_undirected)
print("Node connectivity: ", node_connectivity)
""" Max degree node + Ego Network """
#find node with largest degree
node_and_degree = mastodon_digraph.degree()
(largest_hub, degree) = sorted(node_and_degree, key=itemgetter(1))[-1]
# Create ego graph of main hub
hub_ego = nx.ego_graph(mastodon_digraph, largest_hub)
# Draw graph
pos = nx.spring_layout(hub_ego)
# exact, slow
# print('node connectivity:', node_connectivity(G.to_undirected()))
# print('edge connectivity:', edge_connectivity(G.to_undirected()))
#print('------------DAG-------------', file = f)
print('Is DAG (should be true):\t',
is_directed_acyclic_graph(G),
file=f)
print('Longest path length:\t', dag_longest_path_length(G), file=f)
#print('--------Clustering--------')
# exact, slow
# print('Average clustering coefficient (undirected):', average_clustering(G.to_undirected()))
# approximate, fast
print('Average clustering coefficient (undirected, approx.):\t',
approx.average_clustering(G.to_undirected()),
file=f)
# reprint in the file \t \t \t \t \t
print("\n\n\n\n", file=f)
#---------Nodes & Edges--------'
print(G.number_of_nodes(), end="\t", file=f)
print(len([
k for k in list(nx.weakly_connected_components(G)) if (len(k) == 1)
]),
end="\t",
file=f)
print(percentage, end="\t", file=f)
print(G.number_of_edges(), end="\t", file=f)
#--------Connectivity--------
net = nx.Graph(nx.read_pajek(path))
d = dict()
d['Graph'] = g
d['n_nodes'] = net.number_of_nodes()
d['n_edges'] = net.number_of_edges()
node_list = list(net.nodes())
degrees = [x[1] for x in net.degree(node_list)]
average_dg = np.average(degrees)
max_dg = np.max(degrees)
min_dg = np.min(degrees)
d['min_degree'] = round(min_dg,4)
d['max_degree'] = round(max_dg,4)
d['av_degree'] = round(average_dg,4)
#print('Maximum degree: ', max_dg, ' Minimum degree: ', min_dg, ' Average degree: ', average_dg)
d['clustering_cof'] = round(approximation.average_clustering(net),4)
d['assortativity'] = round(degree_assortativity_coefficient(net),4)
d['av_path_length'] = round(average_shortest_path_length(net),4)
d['diameter'] = round(diameter(net),4)
df = df.append(d, ignore_index=True)
df.to_csv('descriptors.csv', index=False)
def test_petersen_seed():
# Actual coefficient is 0
G = nx.petersen_graph()
assert_equal(average_clustering(G, trials=int(len(G) / 2), seed=1),
nx.average_clustering(G))
