def main():
interaction_types = ['mentions', 'replies', 'retweets']
for interaction_type in interaction_types:
edge_list = util.get_edge_list(interaction_type)
c_list, m_list, p_list = util.get_lists()
c_df, m_df, p_df = util.get_list_dfs()
node_lists = [c_list, m_list, p_list]
cmp_list = util.append_arrays(c_list, m_list, p_list)
G = gc.create_graph_edge_weights(edge_list)
G = gc.create_graph_subset(G, cmp_list)
# Page Rank
pr = nx.pagerank(G)
with open('{}_pagerank.csv'.format(interaction_type), 'wb') as csv_file:
writer = csv.writer(csv_file)
for key, value in pr.items():
writer.writerow([key, value])
# Betweenness centrality
bc = nx.betweenness_centrality(G)
with open('{}_betweenness.csv'.format(interaction_type), 'wb') as csv_file:
writer = csv.writer(csv_file)
for key, value in bc.items():
writer.writerow([key, value])
# Closeness centrality
cc = nx.closeness_centrality(G)
with open('{}_closeness.csv'.format(interaction_type), 'wb') as csv_file:
writer = csv.writer(csv_file)
for key, value in cc.items():
writer.writerow([key, value])
def attribute_info():
c_list, m_list, p_list = util.get_lists()
cmp_list = util.append_arrays(c_list, m_list, p_list)
interaction_types = ['mentions', 'replies', 'retweets']
for interaction_type in interaction_types:
edge_list = util.get_edge_list(interaction_type)
g = create_graph_edge_weights(edge_list)
cmp_g = create_graph_subset(g, cmp_list)
add_types(cmp_g)
print('{} Assortativity: '.format(interaction_type),
nx.attribute_assortativity_coefficient(cmp_g, 'type'))
print('{} Mixing: '.format(interaction_type),
nx.attribute_mixing_dict(cmp_g, 'type', normalized=True))
def main():
interaction_types = ['mentions', 'replies', 'retweets']
for interaction_type in interaction_types:
edge_list = util.get_edge_list(interaction_type)
c_list, m_list, p_list = util.get_lists()
c_df, m_df, p_df = util.get_list_dfs()
node_lists = [c_list, m_list, p_list]
node_labels = ["celebrities", "media", "politicians", "others"]
cmp_list = util.append_arrays(c_list, m_list, p_list)
g = gc.create_graph_edge_weights(edge_list)
cmp_g = gc.create_graph_subset(g, cmp_list)
add_colors(cmp_g)
filename = "{}.json".format(interaction_type)
json_file = open(filename, 'w')
print "filename: %s" % filename
write_graph_to_json(json_file, cmp_g, interaction_type[:3])
json_file.close()
def test_conductance():
edge_list = util.get_edge_list('replies')
c_list, m_list, p_list = util.get_lists()
c_df, m_df, p_df = util.get_list_dfs()
node_lists = [c_list, m_list, p_list]
node_labels = ["celebrities", "media", "politicians", "others"]
cmp_list = util.append_arrays(c_list, m_list, p_list)
g = gc.create_graph_edge_weights(edge_list)
cmp_g = gc.create_graph_subset(g, cmp_list)
nodes = np.asarray(cmp_g.nodes())
node_lists = [c_list, m_list, p_list]
am = clustering.am_to_vectors(
nodes, nx.adjacency_matrix(
cmp_g, weight='None')) # Vectors of in and out degrees, unweighted
clusters_df = clustering.spectral_clustering(nodes, am, 3)
for i in range(3):
conductance = conductance_score(cmp_g, clusters_df, str(i))
print(conductance)
def main():
interaction_types = ['mentions', 'replies', 'retweets']
f = open("Clustering/Cluster_Info.csv", 'a')
f2 = open("Clustering/Cluster_Stats.csv", 'a')
writer = csv.writer(f2, lineterminator='\n')
writer.writerow([
'Cluster Method', 'Num Clusters', 'Avg Max Percent', 'Avg Min Percent',
'Avg Conductance', 'Homogeneity Score', 'Completeness Score', 'V Score'
])
writer = csv.writer(f, lineterminator='\n')
writer.writerow([
'Cluster Method', 'Cluster Num', 'Conductance',
'Clustering Coefficient', 'Percent Celebrities', 'Percent Media',
'Percent Politicians', 'Number Celebrities', 'Number Media',
'Number Politicians'
])
f.close()
f2.close()
for interaction_type in interaction_types:
edge_list = util.get_edge_list(interaction_type)
c_list, m_list, p_list = util.get_lists()
c_df, m_df, p_df = util.get_list_dfs()
node_lists = [c_list, m_list, p_list]
node_labels = ["celebrities", "media", "politicians", "others"]
cmp_list = util.append_arrays(c_list, m_list, p_list)
g = gc.create_graph_edge_weights(edge_list)
cmp_g = gc.create_graph_subset(g, cmp_list)
# Spectral Clustering
clusters_nums = [2, 3, 4]
nodes = np.asarray(cmp_g.nodes())
am_undir_weight = nx.adjacency_matrix(
cmp_g.to_undirected(), weight='weight') # Undirected, weighted
am_undir_unweight = nx.adjacency_matrix(
cmp_g.to_undirected(), weight='None') # Undirected, unweighted
am_dir_weight = nx.adjacency_matrix(
cmp_g, weight='weight') # Outgoing, weighted
am_dir_unweight = nx.adjacency_matrix(
cmp_g, weight='None') # Outgoing, unweighted
am_sum_weight = am_to_sum(am_dir_weight)
am_sum_unweight = am_to_sum(am_dir_unweight)
am_prod_weight = am_to_prod(am_dir_weight)
am_prod_unweight = am_to_prod(am_dir_unweight)
am_bib_weight = am_bib(am_dir_weight)
am_bib_unweight = am_bib(am_dir_unweight)
am_degdiscount_weight = am_deg_discounted(am_dir_weight)
am_degdiscount_unweight = am_deg_discounted(am_dir_unweight)
names = [
'undir_weight', 'undir_unweight', 'sum_weighted', 'sum_unweighted',
'prod_weight', 'prod_unweighted', 'bib_weight', 'bib_unweight',
'deg_discount_weight', 'deg_discount_unweight'
]
ams = [
am_undir_weight, am_undir_unweight, am_sum_weight, am_sum_unweight,
am_prod_weight, am_prod_unweight, am_bib_weight, am_bib_unweight,
am_degdiscount_weight, am_degdiscount_unweight
]
for clusters_num in clusters_nums:
for i in range(len(names)):
node_lists = [c_list, m_list, p_list]
name = names[i]
am = ams[i]
print(name, '\n')
print(am, '\n')
df = spectral_clustering(nodes, am, clusters_num)
labeled_df = add_types(
df, c_list, m_list, p_list,
"{}_{}clusters_{}".format(interaction_type, clusters_num,
name))
labeled_df = add_labels(
df, c_df, m_df, p_df,
"{}_{}clusters_{}_labels".format(interaction_type,
clusters_num, name))
part_labels = df['Partition']
k = len(part_labels.unique())
clusters_matrix = labeled_df
gc.draw_color_and_shapenodes_df(
cmp_g,
"Clustering\\Graphs\\spectral_{}_{}clusters_{}".format(
interaction_type, k, name),
interaction_type,
node_lists,
node_labels,
k,
clusters_matrix,
weight='weight')