class PageRank(object):
def __init__(self, jump_prob=0.15, e=math.exp(-8)):
self.db = Datasource()
self.jump_prob = jump_prob
self.e = e
self.doc_ids = sorted([int(v[0]) for v in self.db.get_all_link_docs()])
self.id2idx = {v: idx for idx, v in enumerate(self.doc_ids)}
def __call__(self):
forward_link = self._linkdb_to_dict(self.db.get_forward_link())
doc_size = len(self.doc_ids)
link_matrix = np.ones((doc_size, doc_size),
dtype=np.float32) * (self.jump_prob / doc_size)
scores = np.ones(doc_size, dtype=np.float32) / doc_size
for i in range(doc_size):
doc_id = self.doc_ids[i]
if doc_id in forward_link.keys():
links = [self.id2idx[x] for x in forward_link[doc_id]]
for link in links:
link_matrix[i, link] += (1 - self.jump_prob) / len(links)
else:
link_matrix[i, :] += (1 - self.jump_prob) / doc_size
link_matrix = np.transpose(link_matrix)
scores = scores[:, np.newaxis]
delta = 1
cnt = 0
while delta >= self.e:
prev = scores
scores = np.dot(link_matrix, scores)
delta = np.sum(np.abs(prev - scores))
cnt += 1
print(f"\tPage rank iteration: {cnt}")
return list(zip(self.doc_ids, scores.tolist()))
def _linkdb_to_dict(self, data):
return {
k: [int(tmp) for tmp in v.split(',')] if v is not None else []
for k, v in data
}
