转换 flair 语言模型张量以在 TensorBoard Projector 中查看

Question

我要转换"vectors,"

vectors = [token.embedding for token in sentence]
print(type(vectors))
<class 'list'>

print(vectors)
[tensor([ 0.0077, -0.0227, -0.0004,  ...,  0.1377, -0.0003,  0.0028]),
...
tensor([ 0.0003, -0.0461,  0.0043,  ..., -0.0126, -0.0004,  0.0142])]

到

0.0077 -0.0227 -0.0004 ... 0.1377 -0.0003 0.0028
...
0.0003 -0.0461 0.0043 ... -0.0126 -0.0004 0.0142

并将其写入 TSV。

旁白：这些嵌入来自 flair (https://github.com/zalandoresearch/flair)：我怎样才能得到完整的输出，而不是 -0.0004 ... 0.1377 缩写的输出？

Answer 1

好的，我四处挖掘...

事实证明这些是 PyTorch 张量（Flair 使用 PyTorch）。对于到 NumPy 数组的简单转换（根据 https://pytorch.org/docs/stable/tensors.html#torch.Tensor.tolist and 处的 PyTorch 文档，使用 tolist()，一种 PyTorch 方法。
```
>>> import torch
>>> a = torch.randn(2, 2)
>>> print(a)
tensor([[-2.1693,  0.7698],
        [ 0.0497,  0.8462]])

>>> a.tolist()
[[-2.1692984104156494, 0.7698001265525818],
 [0.049718063324689865, 0.8462421298027039]]
```

根据我原来的问题，以下是将这些数据转换为纯文本并将它们写入 TSV 文件的方法。

from flair.embeddings import FlairEmbeddings, Sentence
from flair.models import SequenceTagger
from flair.embeddings import StackedEmbeddings

embeddings_f = FlairEmbeddings('pubmed-forward')
embeddings_b = FlairEmbeddings('pubmed-backward')

sentence = Sentence('The RAS-MAPK signalling cascade serves as a central node in transducing signals from membrane receptors to the nucleus.')

tagger = SequenceTagger.load('ner')

tagger.predict(sentence)
embeddings_f.embed(sentence)

stacked_embeddings = StackedEmbeddings([
    embeddings_f,
    embeddings_b,
])

stacked_embeddings.embed(sentence)

# for token in sentence:
#     print(token)
#     print(token.embedding)
#     print(token.embedding.shape)

tokens = [token for token in sentence]
print(tokens)
'''
  [Token: 1 The, Token: 2 RAS-MAPK, Token: 3 signalling, Token: 4 cascade, Token: 5 serves, Token: 6 as, Token: 7 a, Token: 8 central, Token: 9 node, Token: 10 in, Token: 11 transducing, Token: 12 signals, Token: 13 from, Token: 14 membrane, Token: 15 receptors, Token: 16 to, Token: 17 the, Token: 18 nucleus.]
'''

## https://www.geeksforgeeks.org/python-string-split/

tokens = [str(token).split()[2] for token in sentence]
print(tokens)
'''
  ['The', 'RAS-MAPK', 'signalling', 'cascade', 'serves', 'as', 'a', 'central', 'node', 'in', 'transducing', 'signals', 'from', 'membrane', 'receptors', 'to', 'the', 'nucleus.']
'''

tensors = [token.embedding for token in sentence]
print(tensors)
'''
  [tensor([ 0.0077, -0.0227, -0.0004,  ...,  0.1377, -0.0003,  0.0028]),
  tensor([-0.0007, -0.1601, -0.0274,  ...,  0.1982,  0.0013,  0.0042]),
  tensor([ 4.2534e-03, -3.1018e-01, -3.9660e-01,  ...,  5.9336e-02, -9.4445e-05,  1.0025e-02]),
  tensor([ 0.0026, -0.0087, -0.1398,  ..., -0.0037,  0.0012,  0.0274]),
  tensor([-0.0005, -0.0164, -0.0233,  ..., -0.0013,  0.0039,  0.0004]),
  tensor([ 3.8261e-03, -7.6409e-02, -1.8632e-02,  ..., -2.8906e-03, -4.4556e-04,  5.6909e-05]),
  tensor([ 0.0035, -0.0207,  0.1700,  ..., -0.0193,  0.0017,  0.0006]),
  tensor([ 0.0159, -0.4097, -0.0489,  ...,  0.0743,  0.0005,  0.0012]),
  tensor([ 9.7725e-03, -3.3817e-01, -2.2848e-02,  ..., -6.6284e-02, 2.3646e-04,  1.0505e-02]),
  tensor([ 0.0219, -0.0677, -0.0154,  ...,  0.0102,  0.0066,  0.0016]),
  tensor([ 0.0092, -0.0431, -0.0450,  ...,  0.0060,  0.0002,  0.0005]),
  tensor([ 0.0047, -0.2732, -0.0408,  ...,  0.0136,  0.0005,  0.0072]),
  tensor([ 0.0072, -0.0173, -0.0149,  ..., -0.0013, -0.0004,  0.0056]),
  tensor([ 0.0086, -0.1151, -0.0629,  ...,  0.0043,  0.0050,  0.0016]),
  tensor([ 7.6452e-03, -2.3825e-01, -1.5683e-02,  ..., -5.4974e-04, -1.4646e-04,  6.6120e-03]),
  tensor([ 0.0038, -0.0354, -0.1337,  ...,  0.0060, -0.0004,  0.0102]),
  tensor([ 0.0186, -0.0151, -0.0641,  ...,  0.0188,  0.0391,  0.0069]),
  tensor([ 0.0003, -0.0461,  0.0043,  ..., -0.0126, -0.0004,  0.0142])]
'''

# ----------------------------------------
## Write those data to TSV files.

## 

import csv

metadata_f = 'metadata.tsv'
tensors_f = 'tensors.tsv'

with open(metadata_f, 'w', encoding='utf8', newline='') as tsv_file:
    tsv_writer = csv.writer(tsv_file, delimiter='\t', lineterminator='\n')
    for token in tokens:
        ## Assign to a dummy variable ( _ ) to suppress character counts;
        ## if I use (token), rather than ([token]), I get spaces between all characters:
        _ = tsv_writer.writerow([token])

## metadata.tsv :
'''
  The
  RAS-MAPK
  signalling
  cascade
  serves
  as
  a
  central
  node
  in
  transducing
  signals
  from
  membrane
  receptors
  to
  the
  nucleus.
'''

with open(metadata_f, 'w', encoding='utf8', newline='') as tsv_file:
    tsv_writer = csv.writer(tsv_file, delimiter='\t', lineterminator='\n')
    _ = tsv_writer.writerow(tokens)

## metadata.tsv :
'''
  The   RAS-MAPK    signalling  cascade serves  as  a   central node    in  transducing signals from    membrane    receptors   to  the nucleus.
'''

tensors = [token.embedding for token in sentence]
print(tensors)
'''
  [tensor([ 0.0077, -0.0227, -0.0004,  ...,  0.1377, -0.0003,  0.0028]),
  tensor([-0.0007, -0.1601, -0.0274,  ...,  0.1982,  0.0013,  0.0042]),
  tensor([ 4.2534e-03, -3.1018e-01, -3.9660e-01,  ...,  5.9336e-02, -9.4445e-05,  1.0025e-02]),
  tensor([ 0.0026, -0.0087, -0.1398,  ..., -0.0037,  0.0012,  0.0274]),
  tensor([-0.0005, -0.0164, -0.0233,  ..., -0.0013,  0.0039,  0.0004]),
  tensor([ 3.8261e-03, -7.6409e-02, -1.8632e-02,  ..., -2.8906e-03, -4.4556e-04,  5.6909e-05]),
  tensor([ 0.0035, -0.0207,  0.1700,  ..., -0.0193,  0.0017,  0.0006]),
  tensor([ 0.0159, -0.4097, -0.0489,  ...,  0.0743,  0.0005,  0.0012]),
  tensor([ 9.7725e-03, -3.3817e-01, -2.2848e-02,  ..., -6.6284e-02, 2.3646e-04,  1.0505e-02]),
  tensor([ 0.0219, -0.0677, -0.0154,  ...,  0.0102,  0.0066,  0.0016]),
  tensor([ 0.0092, -0.0431, -0.0450,  ...,  0.0060,  0.0002,  0.0005]),
  tensor([ 0.0047, -0.2732, -0.0408,  ...,  0.0136,  0.0005,  0.0072]),
  tensor([ 0.0072, -0.0173, -0.0149,  ..., -0.0013, -0.0004,  0.0056]),
  tensor([ 0.0086, -0.1151, -0.0629,  ...,  0.0043,  0.0050,  0.0016]),
  tensor([ 7.6452e-03, -2.3825e-01, -1.5683e-02,  ..., -5.4974e-04, -1.4646e-04,  6.6120e-03]),
  tensor([ 0.0038, -0.0354, -0.1337,  ...,  0.0060, -0.0004,  0.0102]),
  tensor([ 0.0186, -0.0151, -0.0641,  ...,  0.0188,  0.0391,  0.0069]),
  tensor([ 0.0003, -0.0461,  0.0043,  ..., -0.0126, -0.0004,  0.0142])]
'''

with open(tensors_f, 'w', encoding='utf8', newline='') as tsv_file:
    tsv_writer = csv.writer(tsv_file, delimiter='\t', lineterminator='\n')
    for token in sentence:
        embedding = token.embedding
        _ = tsv_writer.writerow(embedding.tolist())

## tensors.tsv (18 lines: one embedding per token in metadata.tsv):
## note: enormous output, even for this simple sentence.
'''
  0.007691788021475077  -0.02268664352595806    -0.0004340760060586035  ...
'''

最后，我的目的是将上下文语言嵌入（Flair 等）加载到 TensorFlow 的 Embedding Projector 中。事实证明，我需要做的就是将（此处为 Flair 数据）转换为 NumPy 数组，并将它们加载到 TensorFlow TensorBoard 实例中（不需要 TSV 文件！）。

我在我的博客 post 中对此进行了详细描述，此处：Visualizing Language Model Tensors (Embeddings) in TensorFlow's TensorBoard [TensorBoard Projector: PCA; t-SNE; ...]。

Answer 2

要获取令牌，您可以使用 token.text 和 token.embedding.tolist() 来获取嵌入：

def flair_embeddings(sentences, output_file=None):
    if output_file:
        f = open(output_file, 'w')
    # init embedding
    flair_embedding_forward = FlairEmbeddings('news-forward')
    
    for i, sent in enumerate(sentences):
        print("Encoding the {}th input sentence!".format(i))
        # create a sentence
        sentence = Sentence(sent)

        # embed words in sentence
        flair_embedding_forward.embed(sentence)

        for token in sentence:
            if output_file:

                f.write(token.text + "\t" + "\t".join([str(num) for num in token.embedding.tolist()]) + '\n')
            else:
                print(token.text + "\t" + "\t".join([str(num) for num in token.embedding.tolist()]) + '\n')

转换 flair 语言模型张量以在 TensorBoard Projector 中查看

Transform flair language model tensors for viewing in TensorBoard Projector

python

numpy

pytorch

tensorboard

flair