def _init_sampling(self) -> None:
node_degrees = [self._graph.degree(v) for v in self._graph.nodes]
node_degrees = np.array(node_degrees)
self._node_sampling = als.AliasSampling(np.array(node_degrees), np.array(self._graph.nodes))
edge_weights = [[u, v, w] for u, v, w in self._graph.edges.data("weight", default=1)]
edge_weights = np.array(edge_weights)
self._edge_sampling = als.AliasSampling(edge_weights[:, 2], edge_weights[:, :2])
def _precompute_edges(self):
edge_probs = {}
for u, v, w in self._graph.edges.data("weight", default=1):
# === DIRECTION U -> V ===
next_nodes = []
next_weights = []
for t, d in self._graph[v].items():
w = d.get('weight', 1)
next_nodes.append(t)
# BFS - DFS node2vec sampling strategy
if t == u:
prob = 1.0 / self._p
elif self._graph.has_edge(t, u):
prob = 1.0
else:
prob = 1.0 / self._q
# Append to unnormalized probs
next_weights.append(prob * w)
# Generate alias sampling for edge
next_weights = np.array(next_weights)
if len(next_weights > 0):
edge_probs[(u, v)] = als.AliasSampling(
next_weights / sum(next_weights), np.array(next_nodes))
# === DIRECTION V -> U ===
next_nodes = []
next_weights = []
for t, d in self._graph[u].items():
w = d.get('weight', 1)
next_nodes.append(t)
# BFS - DFS node2vec sampling strategy
if t == v:
prob = 1.0 / self._p
elif self._graph.has_edge(t, v):
prob = 1.0
else:
prob = 1.0 / self._q
# Append to unnormalized probs
next_weights.append(prob * w)
# Generate alias sampling for edge
next_weights = np.array(next_weights)
if len(next_weights > 0):
edge_probs[(u, v)] = als.AliasSampling(
next_weights / sum(next_weights), np.array(next_nodes))
# Store the sampled results
self._edge_probs = edge_probs
def _precompute_nodes(self):
node_probs = {}
for u in self._graph.nodes:
nbr_nodes = []
nbr_weights = []
for v, d in self._graph[u].items():
nbr_nodes.append(v)
nbr_weights.append(d.get('weight', 1))
nbr_weights = np.array(nbr_weights)
if len(nbr_weights > 0):
node_probs[u] = als.AliasSampling(nbr_weights / sum(nbr_weights), np.array(nbr_nodes))
self._node_probs = node_probs
np_results = []
als_results = []
for d in CLASSES:
np_class_results = []
als_class_results = []
preproc_times = []
for s in SEEDS:
# Generate probbilities
np.random.seed(s)
probs = np.random.random_sample(d)
probs = probs / np.sum(probs)
# Alias Sampling Preprocessing
preproc_als = als.AliasSampling(probs, probs)
# Numpy Random Choice
list = []
start = time.time()
for i in range(DRAWS):
sample = np.random.choice(probs, 1, p=probs)
list.append(sample)
np_class_results.append((time.time() - start))
print(list)
# Alias Sampling Choice
list = []
start = time.time()
for i in range(DRAWS):
sample = preproc_als.sample(1)