def detect_hubs(self):
self.hubs = []
# [(node1, node2...), (degree1, degree2, ...)]
node_degrees = list(zip(*self.G.degree()))
degrees = node_degrees[1]
nodes = node_degrees[0]
avg_degree = np.mean(degrees)
std_degree = np.std(degrees)
for idx, node in enumerate(nodes):
deg = degrees[idx]
if(deg > (avg_degree + std_degree)):
self.hubs.append(node)
debug('Hubs: ', self.hubs, ", avg degree: ", avg_degree, ", std degree: ", std_degree)
def run(opts: Options,
method: Node2Vec,
title=None,
cluster2color=['green', 'red', 'blue', 'orange', 'purple', 'cyan'],
num_clusters=6,
zero_indexed=False,
tsne_perplexity=12,
draw=["graph", "tsne"],
save=False,
save_path=None):
G, clf_results = main(opts, method)
model = method.model
if (draw or save):
debug("Generating colors...")
colors = cluster.cluster_colors(graph=G,
model=model,
cluster2color=cluster2color,
num_clusters=num_clusters)
colors_idx = utils.color_by_index(colors, list(model.wv.vocab),
zero_indexed)
if (draw):
debug("Drawing...")
if ("graph" in draw):
debug("Drawing graph...")
layout = nx.drawing.layout.spring_layout(G)
drawing.graph(graph=G,
model=model,
colors=colors,
layout=layout,
title=title)
if ("tsne" in draw):
debug("Drawing TSNE...")
tsne_model = tsne.model(model, perplexity=tsne_perplexity)
drawing.tsne(tsne_model, colors_idx, title=title)
if (save):
assert title, "title must be set if save=True"
assert save_path, "save_path must be set if save=True"
utils.save_gefx(G, colors, title, save_path, zero_indexed)
return clf_results
def preprocess_transition_probs(self):
'''
Preprocessing of transition probabilities for guiding the random walks.
'''
G = self.G
debug("Preprocessing nodes...")
alias_nodes = {}
for node in tqdm(G.nodes(), disable=(not DEBUG)):
# get all weights of node's neighbors
unnormalized_probs = [
G[node][nbr]['weight'] for nbr in sorted(G.neighbors(node))
]
# normalize weights with sum
norm_const = sum(unnormalized_probs)
normalized_probs = [
float(u_prob) / norm_const for u_prob in unnormalized_probs
]
alias_nodes[node] = self.alias_setup(normalized_probs)
debug("Preprocessing edges...")
alias_edges = {}
# Split edges up in chunks
workers = os.cpu_count()
edge_divisor = workers**2
chunk_size = int(G.size() / edge_divisor) + 1
chunks = list(self.split_chunks(G.edges(), chunk_size))
# Creating worker pool & running
#maxtasks = int(len(chunks) / 10)
#with mp.Pool(processes=workers, maxtasksperchild=maxtasks) as pool:
with mp.Pool(processes=workers) as pool:
debug(
f"Workers {workers}, chunks {len(chunks)} of size {chunk_size}"
)
result = list(
tqdm(pool.imap(self.preprocess_edge_chunk, chunks),
total=len(chunks),
disable=(not DEBUG)))
# Combining results into one
for d in result:
alias_edges.update(d)
del result
pool.close()
pool.join()
self.alias_nodes = alias_nodes
self.alias_edges = alias_edges
return
def run(self):
debug("Preprocessing transition probs...")
self.walker.preprocess_transition_probs()
debug("Simulating walks...")
walks = self.walker.simulate_walks(self.opts.num_walks,
self.opts.walk_length)
debug("Learning model...")
# Converts every walk value (= node) from int to str for word2vec (=> words)
walks_str = tqdm([list(map(str, walk)) for walk in walks],
disable=(not DEBUG))
self.model = Word2Vec(walks_str,
workers=self.opts.workers,
size=self.opts.dim,
min_count=self.opts.min_count,
window=self.opts.window,
sg=1)
def __init__(self, opts: MethodOpts, h=1):
debug("Node2Vec opts: ", opts.__dict__)
self.opts = opts
self.h = h
def main(opts: Options, model: node2vec.Node2Vec):
debug("Options: ", opts.__dict__)
random.seed(32)
np.random.seed(32)
# Reading graph
debug("Reading graph...")
G = None
if opts.graph_format == 'adjlist':
G = graph.read_adjlist(filename=opts.input)
elif opts.graph_format == 'edgelist':
G = graph.read_edgelist(filename=opts.input, weighted=opts.weighted)
debug(f"Graph: {len(G.nodes())} nodes, {G.size()} edges")
# Loading/learning embeddings
if (opts.output and os.path.isfile(opts.output)):
debug(f"Model {opts.output} exists, loading from file...")
model.load_embeddings(opts.output)
else:
debug("Training model...")
model.init_walker(G)
model.run()
if opts.output:
debug("Saving embeddings...")
model.save_embeddings(opts.output)
# Classification
clf_results = None
if opts.label_file:
X, Y = classify.read_node_label(opts.label_file)
debug("Training classifier using {:.2f}% nodes...".format(
opts.training_ratio * 100))
clf = classify.Classifier(vectors=model.get_vectors(),
clf=LogisticRegression(solver='liblinear'))
clf_results = clf.split_train_evaluate(X,
Y,
opts.training_ratio,
iter=opts.clf_iterations)
return G, clf_results
def __init__(self, opts: MethodOpts, p=1.0, q=1.0):
debug("Node2Vec opts: ", opts.__dict__)
self.opts = opts
self.p = p
self.q = q
def __init__(self, opts: MethodOpts, jump_prob=0):
debug("Node2Vec opts: ", opts.__dict__)
self.opts = opts
self.jump_prob = jump_prob
