def most_similar_within_model(self, text, topn=10):
words = cut(text)
vec = self.model.infer_vector(words)
s = self.model.docvecs.most_similar([vec], topn=topn)
for index, score in s:
print(index, score, self.docs[int(index)]["artist"])
print(self.docs[int(index)]["desc"], "\n")
def predict(captcha):
imgs = cut(regularized(Image.open(captcha)))
data = [scan(img) for img in imgs]
x = pd.DataFrame(data) / 255.0
x = x.values.reshape(-1, 21, 12, 1)
model = models.load_model('captcha_model', compile=True)
conv = lambda x: chr(x + 48 if 0 <= x <= 9 else x + 87)
return list(map(conv, model.predict_classes(x)))
def test_d2v_with_source(text, model, topn=5):
conn = MyConn()
source_tracks = open(
"../data_related/lyrics_valid_tracks.txt").read().splitlines()
text = replace_noise(text)
text = re.sub(r"( )*[作词|作曲|编曲|制作人|录音|混母带|监制].*\n", "", text)
words = cut(text, join_en=False)
vec = model.infer_vector(words)
s = model.docvecs.most_similar([vec], topn=10)
print(s)
def __iter__(self):
self.artists_documents = []
self.artists_flag = 0
for ar, source in self.artists:
ar = ar.lower().strip()
if source=="KB":
text = str(generate_description_KB(ar, self.KB_df, self.name_2_KB_ents, self.mode))
elif source=="sup":
text = str(generate_description_sup(ar, self.sup_data))
words = cut(text, stops_sup=self.stops_sup, filter_number=True)
self.artists_documents.append((ar, source, text))
self.artists_flag += 1
yield models.doc2vec.TaggedDocument(words, [str(self.artists_flag)])
def count_similarity(self, text1, text2):
words1 = cut(text1)
words2 = cut(text2)
print("simi score: {:.3f}".format(
model.docvecs.similarity_unseen_docs(words1, words2)))
def preprocess(text):
text = re.sub(r"( )*[作词|作曲|编曲|制作人|录音|混母带|监制].*\n", "",
replace_noise(text))
return cut(text, join_en=False)
def get_doc_vector(text, model):
# model = Doc2Vec.load(m_path)
words = cut(replace_noise(text), join_en=False)
vec = model.infer_vector(doc_words=words)
return vec