def compute_user_pagerank(network, user, directed_networks, weighted):
if len(network.es) == 0:
return 1
options = igraph.ARPACKOptions()
options.mxiter = 100000
# Igraph 0.7 has a bad bug! It does not give pagerank of a specific user. It
# only returns pagerank of first user in the graph, no matter what the
# vertices argument is. So we get a list of all pageranks, then query in the
# list with the index of node with name user
user_index = network.vs.find(user).index
if weighted:
# if weights are requested, we should use the original weights rather than their inverse
# (in contrast to closeness). In case of pagerank, large edge weights are good for the
# incoming node. They give it more credit.
weights = network.es['weight']
else:
weights = None
user_pagerank = network.pagerank(directed=directed_networks,
damping=0.85,
weights=weights,
implementation="prpack",
arpack_options=options)[user_index]
if user_pagerank > 1:
sys.stderr.write('Error: Pagerank of user %s is greater than one: %f' %
(user, user_pagerank))
# Igraph returns tiny non-zero values for page ranks that should be really zero.
# manually zero them out.
if user_pagerank < pow(10, -15):
user_pagerank = 0
return utils.round_to_sigfigs(user_pagerank, NUM_SIGFIGS)
def compute_user_pagerank(network, user, directed_networks, weighted):
if len(network.es) == 0:
return 1
options = igraph.ARPACKOptions()
options.mxiter = 100000
# Igraph 0.7 has a bad bug! It does not give pagerank of a specific user. It
# only returns pagerank of first user in the graph, no matter what the
# vertices argument is. So we get a list of all pageranks, then query in the
# list with the index of node with name user
user_index = network.vs.find(user).index
if weighted:
# if weights are requested, we should use the original weights rather than their inverse
# (in contrast to closeness). In case of pagerank, large edge weights are good for the
# incoming node. They give it more credit.
weights = network.es['weight']
else:
weights = None
user_pagerank = network.pagerank(directed=directed_networks,
damping=0.85, weights=weights,
implementation = "prpack",
arpack_options = options)[user_index]
if user_pagerank > 1:
sys.stderr.write('Error: Pagerank of user %s is greater than one: %f' %
(user, user_pagerank))
# Igraph returns tiny non-zero values for page ranks that should be really zero.
# manually zero them out.
if user_pagerank < pow(10, -15):
user_pagerank = 0
return utils.round_to_sigfigs(user_pagerank, NUM_SIGFIGS)
def compute_network_weighted_average_path_length(network, directed_networks):
if len(network.es) == 0:
return 0
# weights are requested, we should use the inverse of weights, since higher weights
# means more interactions between two users
weights = [1.0 / w for w in network.es['weight']]
if directed_networks:
shortest_paths = network.shortest_paths(source=network.vs,
target=network.vs,
weights=weights,
mode=igraph.OUT)
else:
shortest_paths = network.shortest_paths(source=network.vs,
target=network.vs,
weights=weights,
mode=igraph.ALL)
# iterate over all shortest paths and add them up
# assign the largest path length to unconnected nodes
max_path_length = -1
for i in shortest_paths:
for path_length in i:
if path_length != float('Inf') and path_length > max_path_length:
max_path_length = path_length
num_pairs = 0
sum_path_lengths = 0
for i in shortest_paths:
for path_length in i:
sum_path_lengths += min(path_length, max_path_length)
num_pairs += 1
average_path_length = sum_path_lengths / num_pairs
return utils.round_to_sigfigs(average_path_length, NUM_SIGFIGS)
def compute_network_weighted_average_path_length(network, directed_networks):
if len(network.es) == 0:
return 0
# weights are requested, we should use the inverse of weights, since higher weights
# means more interactions between two users
weights = [1.0/w for w in network.es['weight']]
if directed_networks:
shortest_paths = network.shortest_paths(source = network.vs, target = network.vs,
weights = weights, mode = igraph.OUT)
else:
shortest_paths = network.shortest_paths(source = network.vs, target = network.vs,
weights = weights, mode = igraph.ALL)
# iterate over all shortest paths and add them up
# assign the largest path length to unconnected nodes
max_path_length = -1
for i in shortest_paths:
for path_length in i:
if path_length != float('Inf') and path_length > max_path_length:
max_path_length = path_length
num_pairs = 0
sum_path_lengths = 0
for i in shortest_paths:
for path_length in i:
sum_path_lengths += min(path_length, max_path_length)
num_pairs += 1
average_path_length = sum_path_lengths / num_pairs
return utils.round_to_sigfigs(average_path_length, NUM_SIGFIGS)
def compute_user_closeness_centrality(network, user, mode, weighted):
if weighted:
# if weights are requested, we should use the inverse of weights, since higher weights
# means more interactions between two users
weights = [1.0 / w for w in network.es['weight']]
else:
weights = None
user_closeness_centrality = network.closeness(vertices=user,
mode=mode,
weights=weights)
return utils.round_to_sigfigs(user_closeness_centrality, NUM_SIGFIGS)
def compute_user_closeness_centrality(network, user, mode, weighted):
if weighted:
# if weights are requested, we should use the inverse of weights, since higher weights
# means more interactions between two users
weights = [1.0/w for w in network.es['weight']]
else:
weights = None
user_closeness_centrality = network.closeness(vertices = user,
mode = mode,
weights = weights)
return utils.round_to_sigfigs(user_closeness_centrality, NUM_SIGFIGS)
def compute_laplacian_centrality(network, vertex):
# find mapping from vertex to sum of edge weights
vertex_weights = collections.defaultdict(float)
for edge in network.es:
source = edge["source"]
target = edge["target"]
weight = edge["weight"]
vertex_weights[source] += weight
vertex_weights[target] += weight
neighbors = network.neighbors(vertex, mode="all")
result = (vertex_weights[vertex]**2 + vertex_weights[vertex] +
2 * sum(vertex_weights[i] for i in neighbors))
return utils.round_to_sigfigs(result, NUM_SIGFIGS)
def compute_laplacian_centrality(network, vertex):
# find mapping from vertex to sum of edge weights
vertex_weights = collections.defaultdict(float)
for edge in network.es:
source = edge["source"]
target = edge["target"]
weight = edge["weight"]
vertex_weights[source] += weight
vertex_weights[target] += weight
neighbors = network.neighbors(vertex, mode="all")
result = (vertex_weights[vertex]**2 + vertex_weights[vertex] +
2 * sum(vertex_weights[i] for i in neighbors) )
return utils.round_to_sigfigs(result, NUM_SIGFIGS)
def compute_network_weighted_density(network, directed_networks):
if len(network.es) == 0:
return 0
num_vertices = len(network.vs)
num_max_edges = num_vertices * (num_vertices - 1)
sum_edge_weights = 0
for edge_weight in network.es['weight']:
sum_edge_weights += edge_weight
if directed_networks:
density = (sum_edge_weights / num_max_edges)
else:
density = (2.0 * sum_edge_weights / num_max_edges)
return utils.round_to_sigfigs(density, NUM_SIGFIGS)
def compute_network_weighted_density(network, directed_networks):
if len(network.es) == 0:
return 0
num_vertices = len(network.vs)
num_max_edges = num_vertices * (num_vertices - 1)
sum_edge_weights = 0
for edge_weight in network.es['weight']:
sum_edge_weights += edge_weight
if directed_networks:
density = (sum_edge_weights / num_max_edges)
else:
density = (2.0 * sum_edge_weights / num_max_edges)
return utils.round_to_sigfigs(density, NUM_SIGFIGS)
def compute_user_betweenness_centrality(network, user, directed_networks, weighted):
if len(network.es) == 0:
return 1
if weighted:
# if weights are requested, we should use the inverse of weights, since higher weights
# means more interactions between two users so the path length should be smaller
weights = [1.0/w for w in network.es['weight']]
else:
weights = None
user_betweenness = network.betweenness(vertices = user,
directed = directed_networks,
weights = weights,
nobigint = False)
return utils.round_to_sigfigs(user_betweenness, NUM_SIGFIGS)
def compute_user_satoshi_pagerank(network, user, weighted):
if len(network.es) == 0:
return 1
if 'satoshi' not in network.vs["name"]:
return 0
# increase the pagerank computation max number of iterations. sometimes arpack does not
# converge.
options = igraph.ARPACKOptions()
options.mxiter = 100000
# Igraph 0.7 has a bad bug! It does not give pagerank of a specific user. It
# only returns pagerank of first user in the graph, no matter what the
# vertices argument is. So we get a list of all pageranks, then query in the
# list with the index of node with name user
user_index = network.vs.find(user).index
if weighted:
# if weights are requested, we should use the original weights rather than their inverse
# (in contrast to closeness). In case of pagerank, large edge weights are good for the
# incoming node. They give it more credit.
weights = network.es['weight']
else:
weights = None
# Vertices should be the active user whose page rank we want to compute. directed
# should be true so that directed paths are considered. damping is the probability
# that we reset the random walk on Satoshi at each step. reset_vertices should only
# contain Satoshi, since we are interested in flow from Satoshi. weights should be
# the 'weight' attribute so that weights are used in page rank computation.
user_satoshi_pagerank = network.personalized_pagerank(
directed=True,
damping=0.85,
reset_vertices='satoshi',
weights=weights,
implementation="prpack",
arpack_options=options)[user_index]
if user_satoshi_pagerank > 1:
sys.stderr.write(
'Error: Satoshi pagerank of user %s is greater than one: %f' %
(user, user_satoshi_pagerank))
# Igraph returns tiny non-zero values for page ranks that should be really zero.
# manually zero them out.
if user_satoshi_pagerank < pow(10, -15):
user_satoshi_pagerank = 0
return utils.round_to_sigfigs(user_satoshi_pagerank, NUM_SIGFIGS)
def compute_user_betweenness_centrality(network, user, directed_networks,
weighted):
if len(network.es) == 0:
return 1
if weighted:
# if weights are requested, we should use the inverse of weights, since higher weights
# means more interactions between two users so the path length should be smaller
weights = [1.0 / w for w in network.es['weight']]
else:
weights = None
user_betweenness = network.betweenness(vertices=user,
directed=directed_networks,
weights=weights,
nobigint=False)
return utils.round_to_sigfigs(user_betweenness, NUM_SIGFIGS)
def compute_user_satoshi_pagerank(network, user, weighted):
if len(network.es) == 0:
return 1
if 'satoshi' not in network.vs["name"]:
return 0
# increase the pagerank computation max number of iterations. sometimes arpack does not
# converge.
options = igraph.ARPACKOptions()
options.mxiter = 100000
# Igraph 0.7 has a bad bug! It does not give pagerank of a specific user. It
# only returns pagerank of first user in the graph, no matter what the
# vertices argument is. So we get a list of all pageranks, then query in the
# list with the index of node with name user
user_index = network.vs.find(user).index
if weighted:
# if weights are requested, we should use the original weights rather than their inverse
# (in contrast to closeness). In case of pagerank, large edge weights are good for the
# incoming node. They give it more credit.
weights = network.es['weight']
else:
weights = None
# Vertices should be the active user whose page rank we want to compute. directed
# should be true so that directed paths are considered. damping is the probability
# that we reset the random walk on Satoshi at each step. reset_vertices should only
# contain Satoshi, since we are interested in flow from Satoshi. weights should be
# the 'weight' attribute so that weights are used in page rank computation.
user_satoshi_pagerank = network.personalized_pagerank(
directed=True, damping=0.85, reset_vertices='satoshi', weights=weights,
implementation = "prpack", arpack_options = options)[user_index]
if user_satoshi_pagerank > 1:
sys.stderr.write('Error: Satoshi pagerank of user %s is greater than one: %f' %
(user, user_satoshi_pagerank))
# Igraph returns tiny non-zero values for page ranks that should be really zero.
# manually zero them out.
if user_satoshi_pagerank < pow(10, -15):
user_satoshi_pagerank = 0
return utils.round_to_sigfigs(user_satoshi_pagerank, NUM_SIGFIGS)
def compute_user_clustering_coefficient(network, user):
user_clustering_coefficient = network.transitivity_local_undirected(
vertices=user, mode="zero")
return utils.round_to_sigfigs(user_clustering_coefficient, NUM_SIGFIGS)
def compute_network_average_degree(network):
average_degree = network.degree_distribution().mean
return utils.round_to_sigfigs(average_degree, NUM_SIGFIGS)
def compute_network_clustering_coefficient(network):
clustering_coefficient = network.transitivity_undirected(mode="zero")
return utils.round_to_sigfigs(clustering_coefficient, NUM_SIGFIGS)
def compute_network_average_path_length(network, directed_networks):
average_path_length = network.average_path_length(directed = directed_networks,
unconn = True)
return utils.round_to_sigfigs(average_path_length, NUM_SIGFIGS)
def compute_network_unweighted_average_path_length(network, directed_networks):
average_path_length = network.average_path_length(
directed=directed_networks, unconn=False)
return utils.round_to_sigfigs(average_path_length, NUM_SIGFIGS)
def compute_network_unweighted_density(network):
return utils.round_to_sigfigs(network.density(loops=True), NUM_SIGFIGS)
def compute_network_unweighted_density(network): return utils.round_to_sigfigs(network.density(loops=True), NUM_SIGFIGS)
def compute_network_clustering_coefficient(network): clustering_coefficient = network.transitivity_undirected(mode="zero") return utils.round_to_sigfigs(clustering_coefficient, NUM_SIGFIGS)
def compute_network_average_degree(network): average_degree = network.degree_distribution().mean return utils.round_to_sigfigs(average_degree, NUM_SIGFIGS)
def compute_user_clustering_coefficient(network, user):
user_clustering_coefficient = network.transitivity_local_undirected(
vertices = user, mode = "zero")
return utils.round_to_sigfigs(user_clustering_coefficient, NUM_SIGFIGS)
