def visualize_col():
"""
"""
adj_matrix = col_matrix(col_k, nnc_mps, agreement_matrix, nnc_arbit, mps)
classes = get_equiv_classes(nnc_mps, agreement_matrix, nnc_arbit, mps)
classes_set = set(classes)
classes_set.remove(-1)
num_classes = len(classes_set)
classes_ord = list(classes_set)
classes_ord.sort()
class_indices = {classes_ord[i]: i for i in range(num_classes)}
class_sizes = [classes.count(cls) for cls in classes_ord]
groups = [None for i in range(len(classes))]
for i in range(len(classes_ord)):
class_index = i
rel_p = [mp[1][0] for mp in mps if classes[mps.index(mp)] == classes_ord[class_index]]
counts = [rel_p.count(rel_p[j]) for j in range(len(rel_p))]
dominant = rel_p[counts.index(max(counts))]
groups[i] = dominant
names = [str(i) for i in range(len(adj_matrix))]
#groups = [str(i) for i in range(len(adj_matrix))]
write_json_graph_general(adj_matrix,
names,
groups,
filename = "viz/col/" + agr_method + '_' + str(start_session) +
'_' + str(end_session) + 'minsess' +
str(min_sessions) + '_' + 'k=' + str(col_k) +
'_' + str(nnc_arbit) + ".json")
def top_k(k, nnc_mps, internal, agreement_matrix, arbit, mps, common_session_matrix=None):
"""
Compute the matrix for the NNC graph.
Parameters
==========
k - number of top neighbors to be included
nnc_mps - the set of MPs to be considered
internal - if True, we take the top k neighbors among nnc_mps;
if False, we take the top k neighbors among all MPs
(but visualize only the connections to neighbors among
the nnc_mps)
agreement_matrix - the agreement matrix used
arbit - whether to take all neighbors corresponding to the top k values
of similarity, or take the first in each group
mps - all the nodes in the network
common_session_matrix - the matrix holding common session counts over the
min_sessions threshold, indexed in the same way as the agreement matrix
Returns
=======
An adjacency matrix over the MPs so that there is a edge
from i to j if j is among the top k neighbors of i.
Notes
=====
I am stupid and have implemented this with an adjacency matrix.
However, the graph is very likely to be sufficiently sparse, so
an adjacency list will be much better.
"""
# MPs to be considered
nnc_mps_ind = [mps.index(mp) for mp in nnc_mps]
adj_matrix = [[0 for i in mps] for j in mps]
for i in nnc_mps_ind:
row = [agreement_matrix[i][j] for j in nnc_mps_ind if j != i]
for j in range(len(row)):
if row[j] == none_value:
row[j] = 0
row_values = list(set(row))
row_values.sort(key = lambda x: -x)
top_ki = row_values[:k]
top_ki = [value for value in top_ki if value != 0]
if arbit == False:
for j in nnc_mps_ind:
if j != i and agreement_matrix[i][j] in top_ki:
adj_matrix[i][j] = 1
if arbit == True:
marked = [False]*len(top_ki)
for j in nnc_mps_ind:
if j != i and agreement_matrix[i][j] in top_ki:
ind = top_ki.index(agreement_matrix[i][j])
if marked[ind] == False:
marked[ind] = True
adj_matrix[i][j] = 1
return adj_matrix
def get_equiv_classes(nnc_mps, agreement_matrix, nnc_arbit, mps):
adj_matrix = top_k(1, nnc_mps, nnc_internal, agreement_matrix, nnc_arbit, mps)
rel_indices = [mps.index(mp) for mp in nnc_mps]
size = len(adj_matrix)
classes = [-1]*size
marked = [False]*size
for i in rel_indices:
if not marked[i]:
marked[i] = True
classes[i] = i
cloud = set([i])
new = True
while new:
new_cloud = set()
for j in cloud:
temp = [l for l in rel_indices if not marked[l] and (adj_matrix[j][l] == 1 or adj_matrix[l][j] == 1)]
for l in temp:
marked[l] = True
new_cloud.add(l)
classes[l] = i
if len(new_cloud) == 0:
new = False
cloud = new_cloud
return classes
def top_k_old(k, nnc_mps, internal, agreement_matrix, arbit, mps, common_session_matrix=None):
"""
Compute the matrix for the NNC graph.
Parameters
==========
k - number of top neighbors to be included
nnc_mps - the set of MPs to be considered
internal - if True, we take the top k neighbors among nnc_mps;
if False, we take the top k neighbors among all MPs
(but visualize only the connections to neighbors among
the nnc_mps)
agreement_matrix - the agreement matrix used
arbit - whether to take all neighbors corresponding to the top k values
of similarity, or take the first in each group
mps - all the nodes in the network
common_session_matrix - the matrix holding common session counts over the
min_sessions threshold, indexed in the same way as the agreement matrix
Returns
=======
An adjacency matrix over the MPs so that there is a edge
from i to j if j is among the top k neighbors of i.
Notes
=====
I am stupid and have implemented this with an adjacency matrix.
However, the graph is very likely to be sufficiently sparse, so
an adjacency list will be much better.
"""
# MPs to be considered
graph_mps = nnc_mps
nnc_mps_ind = [mps.index(mp) for mp in nnc_mps]
adj_matrix = [[None for i in mps] for j in mps]
# compute the matrix
for i in nnc_mps_ind:
row = []
if internal == True:
for j in range(len(mps)):
if agreement_matrix[i][j] == None or j not in nnc_mps_ind:
row.append(-1)
else:
row.append(agreement_matrix[i][j])
if internal == False:
for j in range(len(mps)):
if agreement_matrix[i][j] == None:
row.append(-1)
else:
row.append(agreement_matrix[i][j])
ind = range(len(mps))
ind.sort(key = lambda index: -row[index])
ind = [index for index in ind if index != i and row[index] != -1]
if arbit == False:
trunc = ind[:min(len(ind), k)]
rel_neigh = [t for t in nnc_mps_ind if agreement_matrix[i][t] in [agreement_matrix[i][l] for l in trunc]]
for t in rel_neigh:
adj_matrix[i][t] = 1
else:
for t in range(min(len(ind), k)):
adj_matrix[i][ind[t]] = 1
return adj_matrix