-
-
Notifications
You must be signed in to change notification settings - Fork 7
/
main.py
38 lines (26 loc) · 1015 Bytes
/
main.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
import spacy
import numpy as np
from sentence_transformers import SentenceTransformer
def vectorise(sent):
return model.encode([sent.text])[0]
def overwrite_vectors(doc):
doc.user_hooks['vector'] = vectorise
doc.user_span_hooks['vector'] = vectorise
doc.user_token_hooks['vector'] = vectorise
return doc
nlp = spacy.blank('en')
nlp.add_pipe(overwrite_vectors)
# https://github.com/UKPLab/sentence-transformers
model = SentenceTransformer('bert-base-nli-mean-tokens') # 768
model = SentenceTransformer('roberta-large-nli-stsb-mean-tokens') # 1024
sentences = ['This framework generates embeddings for each input sentence',
'Sentences are passed as a list of string.',
'The quick brown fox jumps over the lazy dog.',
'Sentences are given as a list of strings']
docs = [nlp(s) for s in sentences]
print(docs[0].vector.shape)
m = np.zeros((len(docs), len(docs)))
for i, d_i in enumerate(docs):
for j, d_j in enumerate(docs):
m[i,j] = d_i.similarity(d_j)
print(m)