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
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
|
"""Word piece mapping."""
from pathlib import Path
from typing import List, Optional, Union, Set
import torch
from loguru import logger as log
from torch import Tensor
from text_recognizer.data.emnist_mapping import EmnistMapping
from text_recognizer.data.iam_preprocessor import Preprocessor
class WordPieceMapping(EmnistMapping):
def __init__(
self,
data_dir: Optional[Path] = None,
num_features: int = 1000,
tokens: str = "iamdb_1kwp_tokens_1000.txt",
lexicon: str = "iamdb_1kwp_lex_1000.txt",
use_words: bool = False,
prepend_wordsep: bool = False,
special_tokens: Set[str] = {"<s>", "<e>", "<p>"},
extra_symbols: Set[str] = {"\n",},
) -> None:
super().__init__(extra_symbols=extra_symbols)
self.data_dir = (
(
Path(__file__).resolve().parents[2]
/ "data"
/ "downloaded"
/ "iam"
/ "iamdb"
)
if data_dir is None
else Path(data_dir)
)
log.debug(f"Using data dir: {self.data_dir}")
if not self.data_dir.exists():
raise RuntimeError(f"Could not locate iamdb directory at {self.data_dir}")
processed_path = (
Path(__file__).resolve().parents[2] / "data" / "processed" / "iam_lines"
)
tokens_path = processed_path / tokens
lexicon_path = processed_path / lexicon
special_tokens = set(special_tokens)
if self.extra_symbols is not None:
special_tokens = special_tokens | set(extra_symbols)
self.wordpiece_processor = Preprocessor(
data_dir=self.data_dir,
num_features=num_features,
tokens_path=tokens_path,
lexicon_path=lexicon_path,
use_words=use_words,
prepend_wordsep=prepend_wordsep,
special_tokens=special_tokens,
)
def __len__(self) -> int:
return len(self.wordpiece_processor.tokens)
def get_token(self, index: Union[int, Tensor]) -> str:
if (index := int(index)) <= self.wordpiece_processor.num_tokens:
return self.wordpiece_processor.tokens[index]
raise KeyError(f"Index ({index}) not in mapping.")
def get_index(self, token: str) -> Tensor:
if token in self.wordpiece_processor.tokens:
return torch.LongTensor([self.wordpiece_processor.tokens_to_index[token]])
raise KeyError(f"Token ({token}) not found in inverse mapping.")
def get_text(self, indices: Union[List[int], Tensor]) -> str:
if isinstance(indices, Tensor):
indices = indices.tolist()
return self.wordpiece_processor.to_text(indices).replace(" ", "▁")
def get_indices(self, text: str) -> Tensor:
return self.wordpiece_processor.to_index(text)
def emnist_to_wordpiece_indices(self, x: Tensor) -> Tensor:
text = "".join([self.mapping[i] for i in x])
text = text.lower().replace(" ", "▁")
return torch.LongTensor(self.wordpiece_processor.to_index(text))
def __getitem__(self, x: Union[str, int, List[int], Tensor]) -> Union[str, Tensor]:
if isinstance(x, int):
x = [x]
if isinstance(x, str):
return self.get_indices(x)
return self.get_text(x)
|
