def add_degree_features(G: DiGraph, df: pd.DataFrame) -> pd.DataFrame:
source_in_degree = []
source_out_degree = []
source_bi_degree = []
source_nbrs = []
sink_in_degree = []
sink_out_degree = []
sink_bi_degree = []
sink_nbrs = []
common_neighbors = []
total_neighbors = []
transitive_links = []
JC_predecessors = []
JC_successors = []
JC_transient_in = []
JC_transient_out = []
JC_neighbors = []
cos_predecessors = []
cos_successors = []
cos_transient_in = []
cos_transient_out = []
cos_neighbors = []
PA_predecessors = []
PA_successors = []
PA_transient_in = []
PA_transient_out = []
PA_neighbors = []
RA_predecessors = []
RA_successors = []
RA_transient_in = []
RA_transient_out = []
RA_neighbors = []
AA_predecessors = []
AA_successors = []
AA_transient_in = []
AA_transient_out = []
AA_neighbors = []
hub_promoted_index = []
hub_suppressed_index = []
for i, row in tqdm(df.iterrows()):
source, sink = row["edge"]
try:
s_in = set(G.predecessors(source))
s_out = set(G.successors(source))
s_bi = set(s_in.intersection(s_out))
s_nbrs = set(s_in.union(s_out))
except:
s_in = set()
s_out = set()
s_bi = set()
s_nbrs = set()
try:
d_in = set(G.predecessors(sink))
d_out = set(G.successors(sink))
d_bi = set(d_in.intersection(d_out))
d_nbrs = set(d_in.union(d_out))
except:
d_in = set()
d_out = set()
d_bi = set()
d_nbrs = set()
source_in_degree.append(len(s_in))
source_out_degree.append(len(s_out))
source_bi_degree.append(len(s_bi))
source_nbrs.append(len(s_nbrs))
sink_in_degree.append(len(d_in))
sink_out_degree.append(len(d_out))
sink_bi_degree.append(len(d_bi))
sink_nbrs.append(len(d_nbrs))
common = len(s_nbrs.intersection(d_nbrs))
common_neighbors.append(common)
total_neighbors.append(len(s_nbrs.union(d_nbrs)))
transitive_links.append(len(s_out.intersection(d_in)))
JC_predecessors.append(jaccard_coeff(s_in, d_in))
JC_successors.append(jaccard_coeff(s_out, d_out))
JC_transient_in.append(jaccard_coeff(s_out, d_in))
JC_transient_out.append(jaccard_coeff(s_in, d_out))
JC_neighbors.append(jaccard_coeff(s_nbrs, d_nbrs))
cos_predecessors.append(cosine_distance(s_in, d_in))
cos_successors.append(cosine_distance(s_out, d_out))
cos_transient_in.append(cosine_distance(s_out, d_in))
cos_transient_out.append(cosine_distance(s_in, d_out))
cos_neighbors.append(cosine_distance(s_nbrs, d_nbrs))
PA_predecessors.append(preferential_attachment(s_in, d_in))
PA_successors.append(preferential_attachment(s_out, d_out))
PA_transient_in.append(preferential_attachment(s_out, d_in))
PA_transient_out.append(preferential_attachment(s_in, d_out))
PA_neighbors.append(preferential_attachment(s_nbrs, d_nbrs))
RA_predecessors.append(directed_resource_allocation(s_in, d_in, G))
RA_successors.append(directed_resource_allocation(s_out, d_out, G))
RA_transient_in.append(directed_resource_allocation(s_out, d_in, G))
RA_transient_out.append(directed_resource_allocation(s_in, d_out, G))
RA_neighbors.append(directed_resource_allocation(s_nbrs, d_nbrs, G))
AA_predecessors.append(directed_adamic_adar(s_in, d_in, G))
AA_successors.append(directed_adamic_adar(s_out, d_out, G))
AA_transient_in.append(directed_adamic_adar(s_out, d_in, G))
AA_transient_out.append(directed_adamic_adar(s_in, d_out, G))
AA_neighbors.append(directed_adamic_adar(s_nbrs, d_nbrs, G))
try:
hub_promoted_index.append(common / min([len(s_nbrs), len(d_nbrs)]))
except:
hub_promoted_index.append(0.0)
try:
hub_suppressed_index.append(
common / max([len(s_nbrs), len(d_nbrs)]))
except:
hub_suppressed_index.append(0.0)
df = pd.DataFrame({
"edge": df.edge,
"source_in_degree": source_in_degree,
"source_out_degree": source_out_degree,
"source_bi_degree": source_bi_degree,
"source_neighbors": source_nbrs,
"sink_in_degree": sink_in_degree,
"sink_out_degree": sink_out_degree,
"sink_bi_degree": sink_out_degree,
"sink_neighbors": sink_nbrs,
"common_neighbors": common_neighbors,
"total_neighbors": total_neighbors,
"transitive_links": transitive_links,
"JC_predecessors": JC_predecessors,
"JC_successors": JC_successors,
"JC_transient_in": JC_transient_in,
"JC_transient_out": JC_transient_out,
"JC_neighbors": JC_neighbors,
"cos_predecessors": cos_predecessors,
"cos_successors": cos_successors,
"cos_transient_in": cos_transient_in,
"cos_transient_out": cos_transient_out,
"cos_neighbors": cos_neighbors,
"PA_predecessors": PA_predecessors,
"PA_successors": PA_successors,
"PA_transient_in": PA_transient_in,
"PA_transient_out": PA_transient_out,
"PA_neighbors": PA_neighbors,
"RA_predecessors": RA_predecessors,
"RA_successors": RA_successors,
"RA_transient_in": RA_transient_in,
"RA_transient_out": RA_transient_out,
"RA_neighbors": RA_neighbors,
"AA_predecessors": AA_predecessors,
"AA_successors": AA_successors,
"AA_transient_in": AA_transient_in,
"AA_transient_out": AA_transient_out,
"AA_neighbors": AA_neighbors,
"hub_promoted_index": hub_promoted_index,
"hub_suppressed_index": hub_suppressed_index,
})
# Other indices
df["sorensen_index"] = 2 * (
df["common_neighbors"] /
(df["source_neighbors"] + df["sink_neighbors"]))
df["LHN_index"] = df["common_neighbors"] / (df["source_neighbors"] *
df["sink_neighbors"])
# Calculate degree densities
df["source_in_density"] = df["source_in_degree"] / df["source_neighbors"]
df["source_out_density"] = df["source_out_degree"] / df["source_neighbors"]
df["source_bi_density"] = df["source_bi_degree"] / df["source_neighbors"]
df["sink_in_density"] = df["sink_in_degree"] / df["sink_neighbors"]
df["sink_out_density"] = df["sink_out_degree"] / df["sink_neighbors"]
df["sink_bi_density"] = df["sink_bi_degree"] / df["sink_neighbors"]
return df
def tree0(weight_value, startwindow, term):
print 'start window:', startwindow
# windowGraph = {}
cliqueGraph = DiGraph()
dic_term = {}
dic_last_time = {}
dic_temp = {}
dic_term_num = {}
dic_intersect_level = {}
# term = 183
root = 0
cliqueGraph.add_node(root, annotation='root', windowsize='root', weight_value='root')
w = data.shape[1]
i = 0
q = 0
for window in range(startwindow, w):
dic_intersect_level.clear()
#print window ## mine
if window == startwindow:
for clique in find_cliques(windowGraph[window]):
if len(clique) >size_clique:
cliqueGraph.add_node(term, annotation=list(clique), windowsize=[window],
weight=weight_value) # generate a term
cliqueGraph.add_edge(root, term)
dic_term[frozenset(clique)] = [window] # dic_term 记录 window和clique or Dic_term records window and clique
dic_term_num[frozenset(clique)] = term # dic_term_num 记录 term 序号和clique or Dic_term_num record term number and clique
dic_last_time[frozenset(clique)] = [window] # dic_last_time 记录上一时刻生成的交集 用于下一时刻的比较 or Dic_last_time records the intersection generated at the last moment for comparison at the next moment
term = term + 1
print 'for start window '
else:
continue
# print len(dic_last_time), len(dic_term), cliqueGraph.number_of_nodes()
else:
for clique in find_cliques(windowGraph[window]):
if len(clique) > size_clique:
#print window, 'clique:', clique ## mine
for key, value in dic_last_time.items(): # key 是clique ,value是 [window] or Key is clique, value is [window]
intersect = sorted(set(key).intersection(set(clique)))
q = 0
# if len(intersect) >= size_clique:
if len(intersect) >= size_clique:
#print 'intersect', intersect
# 同一层判断交集之间是否有重复的父子关系。 每生成一个交集， 判断当前层的其他term和交集的关系。or The same layer determines whether there are
#duplicate parent-child relationships between intersections. Each generation of an intersection determines the relationship
#between other terms and intersections of the current layer.
for ik, iv in dic_intersect_level.items():
if set(intersect) == (set(ik)): # 生成一模一样的交集 or Generate exactly the same intersection
# 判断两个的编号是否一样？or Is the two numbers the same?
if dic_term_num[frozenset(key)] != dic_term_num[frozenset(ik)]:
cliqueGraph.add_edge(dic_term_num[frozenset(key)], dic_term_num[frozenset(ik)])
q = 1
break
elif set(intersect).issuperset(set(ik)): # 生成了超集 or Superset generated
cliqueGraph.remove_node(dic_term_num[frozenset(ik)])
dic_term.pop(frozenset(ik)) # 从四个字典中都删除该节点的信息 or Delete the node's information from all four dictionaries
dic_term_num.pop(frozenset(ik))
dic_intersect_level.pop(frozenset(ik))
dic_temp.pop(frozenset(ik))
elif set(intersect).issubset(set(ik)): # 生成了子集 or Generated subset
q = 1
break
if q == 1:
continue
dic_intersect_level[frozenset(intersect)] = 1
if dic_term.has_key(frozenset(intersect)):
# 交集已经出现过 or Intersection has appeared
parent = cliqueGraph.predecessors(dic_term_num[frozenset(intersect)])
children = cliqueGraph.successors(dic_term_num[frozenset(intersect)])
#print 'parent',len(parent)
if len(parent) > 0:
# 是交集生成的term，则重定向 or Is the intersection of generated term, then redirect
cliqueGraph.add_node(term, annotation=list(intersect),
windowsize=value + [window],
weight=weight_value)
for p in parent:
cliqueGraph.add_edge(p, term) # 连边 // Edge
for c in children:
cliqueGraph.add_edge(term, c) # 连边 // edge
cliqueGraph.remove_node(dic_term_num[frozenset(intersect)]) # 从图中删除冗余结点 or Remove redundant nodes from the figure
# print 'deleted intersect nodes:',dic_term_num[frozenset(intersect)]
i = i + 1
dic_term.pop(frozenset(intersect)) # 字典中删除 // Delete in dictionary
dic_term_num.pop(frozenset(intersect))
dic_term[frozenset(intersect)] = value + [window] # 新节点插入字典 // New node insert dictionary
dic_term_num[frozenset(intersect)] = term
dic_temp[frozenset(intersect)] = value + [window] # 记录到dic_temp里 // Record to dic_temp
term = term + 1
continue
else:
# 是window生成的term // Is the term generated by the window
continue
else:
# 交集没有出现过， 则生成新的term // No intersection occurs, then a new term is generated
# print 'new term intersect never appear:', term
cliqueGraph.add_node(term, annotation=list(intersect), windowsize=value + [window],
weight=weight_value) # generate a term
cliqueGraph.add_edge(dic_term_num[frozenset(key)], term) # 连边，变化：只连接交集作为父亲。// Edge, change: Only connect intersections as fathers.
dic_term[frozenset(intersect)] = value + [window] # 新节点插入字典 // New node insert dictionary
dic_term_num[frozenset(intersect)] = term
dic_temp[frozenset(intersect)] = value + [window] # 记录到dic_temp里 // Record to dic_temp
term = term + 1
else:
continue
else:
continue
dic_last_time.clear()
for key, value in dic_temp.items():
dic_last_time[key] = value
dic_temp.clear()
print 'window', startwindow, 'size is', cliqueGraph.number_of_nodes(), cliqueGraph.number_of_edges()## mine
# print 'deleted nodes:', i
# fw = open('0904edges_remove.txt', 'w')
# fw2 = open('0904terms_remove.txt', 'w')
# fw.write('parent' + '\t' + 'child' + '\n')
# for edge in cliqueGraph.edges():
# fw.write(str(edge[0]) + '\t' + str(edge[1]) + '\n')
# fw.close()
# fw2.write('term_id' + '\t' + 'anno_genes' + '\t' + 'window' + '\t' + 'gene_size' + '\t' + 'window_size' + '\n')
# for key, value in dic_term.items():
# fw2.write(str(dic_term_num[key]) + '\t' + str(key) + '\t' + str(value) + '\t' + str(len(key)) + '\t' + str(len(value)) + '\n')
# fw2.close()
# for nodes in cliqueGraph.nodes():
# if cliqueGraph.degree(nodes) == 0:
# print nodes
return cliqueGraph, dic_term, dic_term_num, term