Spaces:
Running
on
Zero
Running
on
Zero
jedick
commited on
Commit
·
142bd00
1
Parent(s):
f027363
Don't import tqdm for BM25S tokenizer used in retrieval
Browse files- app.py +2 -2
- mods/bm25s_retriever.py +5 -2
- mods/bm25s_tokenization.py +719 -0
app.py
CHANGED
@@ -58,10 +58,10 @@ def cleanup_graph(request: gr.Request):
|
|
58 |
timestamp = datetime.now().replace(microsecond=0).isoformat()
|
59 |
if request.session_hash in graph_instances["local"]:
|
60 |
del graph_instances["local"][request.session_hash]
|
61 |
-
print(f"{timestamp} -
|
62 |
if request.session_hash in graph_instances["remote"]:
|
63 |
del graph_instances["remote"][request.session_hash]
|
64 |
-
print(f"{timestamp} -
|
65 |
|
66 |
|
67 |
def append_content(chunk_messages, history, thinking_about):
|
|
|
58 |
timestamp = datetime.now().replace(microsecond=0).isoformat()
|
59 |
if request.session_hash in graph_instances["local"]:
|
60 |
del graph_instances["local"][request.session_hash]
|
61 |
+
print(f"{timestamp} - Delete local graph for session {request.session_hash}")
|
62 |
if request.session_hash in graph_instances["remote"]:
|
63 |
del graph_instances["remote"][request.session_hash]
|
64 |
+
print(f"{timestamp} - Delete remote graph for session {request.session_hash}")
|
65 |
|
66 |
|
67 |
def append_content(chunk_messages, history, thinking_about):
|
mods/bm25s_retriever.py
CHANGED
@@ -155,13 +155,16 @@ class BM25SRetriever(BaseRetriever):
|
|
155 |
*,
|
156 |
run_manager: CallbackManagerForRetrieverRun,
|
157 |
) -> List[Document]:
|
158 |
-
from
|
159 |
|
160 |
processed_query = bm25s_tokenize(query, return_ids=False)
|
161 |
if self.activate_numba:
|
162 |
self.vectorizer.activate_numba_scorer()
|
163 |
return_docs = self.vectorizer.retrieve(
|
164 |
-
processed_query,
|
|
|
|
|
|
|
165 |
)
|
166 |
return [self.docs[i] for i in return_docs.documents[0]]
|
167 |
else:
|
|
|
155 |
*,
|
156 |
run_manager: CallbackManagerForRetrieverRun,
|
157 |
) -> List[Document]:
|
158 |
+
from mods.bm25s_tokenization import tokenize as bm25s_tokenize
|
159 |
|
160 |
processed_query = bm25s_tokenize(query, return_ids=False)
|
161 |
if self.activate_numba:
|
162 |
self.vectorizer.activate_numba_scorer()
|
163 |
return_docs = self.vectorizer.retrieve(
|
164 |
+
processed_query,
|
165 |
+
k=self.k,
|
166 |
+
backend_selection="numba",
|
167 |
+
show_progress=False,
|
168 |
)
|
169 |
return [self.docs[i] for i in return_docs.documents[0]]
|
170 |
else:
|
mods/bm25s_tokenization.py
ADDED
@@ -0,0 +1,719 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from ast import Tuple
|
2 |
+
from pathlib import Path
|
3 |
+
import re
|
4 |
+
from typing import Any, Dict, List, Union, Callable, NamedTuple
|
5 |
+
import typing
|
6 |
+
|
7 |
+
from bm25s.utils import json_functions
|
8 |
+
|
9 |
+
try:
|
10 |
+
# To hide progress bars, don't import tqdm
|
11 |
+
# from tqdm.auto import tqdm
|
12 |
+
raise ImportError("Not importing tqdm")
|
13 |
+
except ImportError:
|
14 |
+
|
15 |
+
def tqdm(iterable, *args, **kwargs):
|
16 |
+
return iterable
|
17 |
+
|
18 |
+
|
19 |
+
from bm25s.stopwords import (
|
20 |
+
STOPWORDS_EN,
|
21 |
+
STOPWORDS_EN_PLUS,
|
22 |
+
STOPWORDS_GERMAN,
|
23 |
+
STOPWORDS_DUTCH,
|
24 |
+
STOPWORDS_FRENCH,
|
25 |
+
STOPWORDS_SPANISH,
|
26 |
+
STOPWORDS_PORTUGUESE,
|
27 |
+
STOPWORDS_ITALIAN,
|
28 |
+
STOPWORDS_RUSSIAN,
|
29 |
+
STOPWORDS_SWEDISH,
|
30 |
+
STOPWORDS_NORWEGIAN,
|
31 |
+
STOPWORDS_CHINESE,
|
32 |
+
)
|
33 |
+
|
34 |
+
|
35 |
+
class Tokenized(NamedTuple):
|
36 |
+
"""
|
37 |
+
NamedTuple with two fields: ids and vocab. The ids field is a list of list of token IDs
|
38 |
+
for each document. The vocab field is a dictionary mapping tokens to their index in the
|
39 |
+
vocabulary.
|
40 |
+
"""
|
41 |
+
|
42 |
+
ids: List[List[int]]
|
43 |
+
vocab: Dict[str, int]
|
44 |
+
|
45 |
+
def __repr__(self):
|
46 |
+
"""
|
47 |
+
Returns:
|
48 |
+
a string representation of the class.
|
49 |
+
for example, for a small corpus, it would be something like:
|
50 |
+
----
|
51 |
+
Tokenized(
|
52 |
+
"ids": [
|
53 |
+
0: [0, 1, 2, 3]
|
54 |
+
],
|
55 |
+
"vocab": [
|
56 |
+
'': 4
|
57 |
+
'cat': 0
|
58 |
+
'feline': 1
|
59 |
+
'likes': 2
|
60 |
+
'purr': 3
|
61 |
+
],
|
62 |
+
)
|
63 |
+
----
|
64 |
+
|
65 |
+
and, for example, for a large corpus, it would be something like:
|
66 |
+
----
|
67 |
+
Tokenized(
|
68 |
+
"ids": [
|
69 |
+
0: [0, 1, 2, 3]
|
70 |
+
1: [4, 5, 6, 7, 8, 9]
|
71 |
+
2: [10, 11, 12, 13, 14]
|
72 |
+
3: [15, 16, 17, 18, 19]
|
73 |
+
4: [0, 1, 2, 3, 0, 20, 21, 22, 23, 24, ...]
|
74 |
+
5: [0, 1, 2, 3]
|
75 |
+
6: [4, 5, 6, 7, 8, 9]
|
76 |
+
7: [10, 11, 12, 13, 14]
|
77 |
+
8: [15, 16, 17, 18, 19]
|
78 |
+
9: [0, 1, 2, 3, 0, 20, 21, 22, 23, 24, ...]
|
79 |
+
... (total 500000 docs)
|
80 |
+
],
|
81 |
+
"vocab": [
|
82 |
+
'': 29
|
83 |
+
'animal': 12
|
84 |
+
'beautiful': 11
|
85 |
+
'best': 6
|
86 |
+
'bird': 10
|
87 |
+
'can': 13
|
88 |
+
'carefully': 27
|
89 |
+
'casually': 28
|
90 |
+
'cat': 0
|
91 |
+
'creature': 16
|
92 |
+
... (total 30 tokens)
|
93 |
+
],
|
94 |
+
)
|
95 |
+
----
|
96 |
+
"""
|
97 |
+
lines_print_max_num = 10
|
98 |
+
single_doc_print_max_len = 10
|
99 |
+
lines = ["Tokenized(", ' "ids": [']
|
100 |
+
for doc_idx, document in enumerate(self.ids[:lines_print_max_num]):
|
101 |
+
preview = document[:single_doc_print_max_len]
|
102 |
+
if len(document) > single_doc_print_max_len:
|
103 |
+
preview += ["..."]
|
104 |
+
lines.append(f" {doc_idx}: [{', '.join([str(x) for x in preview])}]")
|
105 |
+
if len(self.ids) > lines_print_max_num:
|
106 |
+
lines.append(f" ... (total {len(self.ids)} docs)")
|
107 |
+
lines.append(f' ],\n "vocab": [')
|
108 |
+
vocab_keys = sorted(list(self.vocab.keys()))
|
109 |
+
for vocab_idx, key_ in enumerate(vocab_keys[:lines_print_max_num]):
|
110 |
+
val_ = self.vocab[key_]
|
111 |
+
lines.append(f" {key_!r}: {val_}")
|
112 |
+
if len(list(vocab_keys)) > 10:
|
113 |
+
lines.append(f" ... (total {len(vocab_keys)} tokens)")
|
114 |
+
lines.append(" ],\n)")
|
115 |
+
return "\n".join(lines)
|
116 |
+
|
117 |
+
|
118 |
+
class Tokenizer:
|
119 |
+
"""
|
120 |
+
Tokenizer class for tokenizing a list of strings and converting them to token IDs.
|
121 |
+
|
122 |
+
Parameters
|
123 |
+
----------
|
124 |
+
lower : bool, optional
|
125 |
+
Whether to convert the text to lowercase before tokenization
|
126 |
+
|
127 |
+
splitter : Union[str, Callable], optional
|
128 |
+
If a string is provided, the tokenizer will interpret it as a regex pattern,
|
129 |
+
and use the `re.compile` function to compile the pattern and use the `findall` method
|
130 |
+
to split the text. If a callable is provided, the tokenizer will use the callable to
|
131 |
+
split the text. The callable should take a string as input and return a list of strings.
|
132 |
+
|
133 |
+
stopwords : Union[str, List[str]], optional
|
134 |
+
The list of stopwords to remove from the text. If "english" or "en" is provided,
|
135 |
+
the function will use the default English stopwords. If None or False is provided,
|
136 |
+
no stopwords will be removed. If a list of strings is provided, the tokenizer will
|
137 |
+
use the list of strings as stopwords.
|
138 |
+
|
139 |
+
stemmer : Callable, optional
|
140 |
+
The stemmer to use for stemming the tokens. It is recommended
|
141 |
+
to use the PyStemmer library for stemming, but you can also any callable that
|
142 |
+
takes a list of strings and returns a list of strings.
|
143 |
+
"""
|
144 |
+
|
145 |
+
def __init__(
|
146 |
+
self,
|
147 |
+
lower: bool = True,
|
148 |
+
splitter: Union[str, Callable] = r"(?u)\b\w\w+\b",
|
149 |
+
stopwords: Union[str, List[str]] = "english",
|
150 |
+
stemmer: Callable = None, # type: ignore
|
151 |
+
):
|
152 |
+
self.lower = lower
|
153 |
+
if isinstance(splitter, str):
|
154 |
+
splitter = re.compile(splitter).findall
|
155 |
+
if not callable(splitter):
|
156 |
+
raise ValueError("splitter must be a callable or a regex pattern.")
|
157 |
+
|
158 |
+
# Exception handling for stemmer when we are using PyStemmer, which has a stemWords method
|
159 |
+
if hasattr(stemmer, "stemWord"):
|
160 |
+
stemmer = stemmer.stemWord
|
161 |
+
if not callable(stemmer) and stemmer is not None:
|
162 |
+
raise ValueError("stemmer must be callable or have a `stemWord` method.")
|
163 |
+
|
164 |
+
self.stopwords = _infer_stopwords(stopwords)
|
165 |
+
self.splitter = splitter
|
166 |
+
self.stemmer = stemmer
|
167 |
+
|
168 |
+
self.reset_vocab()
|
169 |
+
|
170 |
+
def reset_vocab(self):
|
171 |
+
"""
|
172 |
+
Reset the vocabulary dictionaries to empty dictionaries, allowing you to
|
173 |
+
tokenize a new set of texts without reusing the previous vocabulary.
|
174 |
+
"""
|
175 |
+
self.word_to_stem = {} # word -> stemmed word, e.g. "apple" -> "appl"
|
176 |
+
self.stem_to_sid = {} # stem -> stemmed id, e.g. "appl" -> 0
|
177 |
+
# word -> {stemmed, unstemmed} id, e.g. "apple" -> 0 (appl) or "apple" -> 2 (apple)
|
178 |
+
self.word_to_id = {}
|
179 |
+
|
180 |
+
def save_vocab(self, save_dir: str, vocab_name: str = "vocab.tokenizer.json"):
|
181 |
+
"""
|
182 |
+
Save the vocabulary dictionaries to a file. The file is saved in JSON format.
|
183 |
+
|
184 |
+
Parameters
|
185 |
+
----------
|
186 |
+
save_dir : str
|
187 |
+
The directory where the vocabulary file is saved.
|
188 |
+
|
189 |
+
vocab_name : str, optional
|
190 |
+
The name of the vocabulary file. Default is "vocab.tokenizer.json". Make
|
191 |
+
sure to not use the same name as the vocab.index.json file saved by the BM25
|
192 |
+
model, as it will overwrite the vocab.index.json file and cause errors.
|
193 |
+
"""
|
194 |
+
save_dir: Path = Path(save_dir)
|
195 |
+
path = save_dir / vocab_name
|
196 |
+
|
197 |
+
save_dir.mkdir(parents=True, exist_ok=True)
|
198 |
+
with open(path, "w", encoding="utf-8") as f:
|
199 |
+
d = {
|
200 |
+
"word_to_stem": self.word_to_stem,
|
201 |
+
"stem_to_sid": self.stem_to_sid,
|
202 |
+
"word_to_id": self.word_to_id,
|
203 |
+
}
|
204 |
+
f.write(json_functions.dumps(d, ensure_ascii=False))
|
205 |
+
|
206 |
+
def load_vocab(self, save_dir: str, vocab_name: str = "vocab.tokenizer.json"):
|
207 |
+
"""
|
208 |
+
Load the vocabulary dictionaries from a file. The file should be saved in JSON format.
|
209 |
+
|
210 |
+
Parameters
|
211 |
+
----------
|
212 |
+
save_dir : str
|
213 |
+
The directory where the vocabulary file is saved.
|
214 |
+
|
215 |
+
vocab_name : str, optional
|
216 |
+
The name of the vocabulary file.
|
217 |
+
|
218 |
+
Note
|
219 |
+
----
|
220 |
+
The vocabulary file should be saved in JSON format, with the following keys:
|
221 |
+
- word_to_stem: a dictionary mapping words to their stemmed words
|
222 |
+
- stem_to_sid: a dictionary mapping stemmed words to their stemmed IDs
|
223 |
+
- word_to_id: a dictionary mapping words to their word
|
224 |
+
"""
|
225 |
+
path = Path(save_dir) / vocab_name
|
226 |
+
|
227 |
+
with open(path, "r", encoding="utf-8") as f:
|
228 |
+
d = json_functions.loads(f.read())
|
229 |
+
self.word_to_stem = d["word_to_stem"]
|
230 |
+
self.stem_to_sid = d["stem_to_sid"]
|
231 |
+
self.word_to_id = d["word_to_id"]
|
232 |
+
|
233 |
+
def save_stopwords(
|
234 |
+
self, save_dir: str, stopwords_name: str = "stopwords.tokenizer.json"
|
235 |
+
):
|
236 |
+
"""
|
237 |
+
Save the stopwords to a file. The file is saved in JSON format.
|
238 |
+
|
239 |
+
Parameters
|
240 |
+
----------
|
241 |
+
save_dir : str
|
242 |
+
The directory where the stopwords file is saved.
|
243 |
+
|
244 |
+
stopwords_name : str, optional
|
245 |
+
The name of the stopwords file. Default is "stopwords.tokenizer.json".
|
246 |
+
"""
|
247 |
+
save_dir: Path = Path(save_dir)
|
248 |
+
path = save_dir / stopwords_name
|
249 |
+
|
250 |
+
save_dir.mkdir(parents=True, exist_ok=True)
|
251 |
+
with open(path, "w") as f:
|
252 |
+
f.write(json_functions.dumps(self.stopwords))
|
253 |
+
|
254 |
+
def load_stopwords(
|
255 |
+
self, save_dir: str, stopwords_name: str = "stopwords.tokenizer.json"
|
256 |
+
):
|
257 |
+
"""
|
258 |
+
Load the stopwords from a file. The file should be saved in JSON format.
|
259 |
+
|
260 |
+
Parameters
|
261 |
+
----------
|
262 |
+
save_dir : str
|
263 |
+
The directory where the stopwords file is saved.
|
264 |
+
|
265 |
+
stopwords_name : str, optional
|
266 |
+
The name of the stopwords file.
|
267 |
+
"""
|
268 |
+
path = Path(save_dir) / stopwords_name
|
269 |
+
|
270 |
+
with open(path, "r") as f:
|
271 |
+
self.stopwords = json_functions.loads(f.read())
|
272 |
+
|
273 |
+
def streaming_tokenize(
|
274 |
+
self,
|
275 |
+
texts: List[str],
|
276 |
+
update_vocab: Union[bool, str] = True,
|
277 |
+
allow_empty: bool = True,
|
278 |
+
):
|
279 |
+
"""
|
280 |
+
Tokenize a list of strings and return a generator of token IDs.
|
281 |
+
|
282 |
+
Parameters
|
283 |
+
----------
|
284 |
+
texts : List[str]
|
285 |
+
A list of strings to tokenize.
|
286 |
+
|
287 |
+
update_vocab : bool, optional
|
288 |
+
Whether to update the vocabulary dictionary with the new tokens. If true,
|
289 |
+
the different dictionaries making up the vocabulary will be updated with the
|
290 |
+
new tokens. If False, the function will not update the vocabulary. Unless you have
|
291 |
+
a stemmer and the stemmed word is in the stem_to_sid dictionary. If "never",
|
292 |
+
the function will never update the vocabulary, even if the stemmed word is in
|
293 |
+
the stem_to_sid dictionary. Note that update_vocab="if_empty" is not supported
|
294 |
+
in this method, only in the `tokenize` method.
|
295 |
+
|
296 |
+
allow_empty : bool, optional
|
297 |
+
Whether to allow the splitter to return an empty string. If False, the splitter
|
298 |
+
will return an empty list, which may cause issues if the tokenizer is not expecting
|
299 |
+
an empty list. If True, the splitter will return a list with a single empty string.
|
300 |
+
"""
|
301 |
+
stopwords_set = set(self.stopwords) if self.stopwords is not None else None
|
302 |
+
using_stopwords = stopwords_set is not None
|
303 |
+
using_stemmer = self.stemmer is not None
|
304 |
+
|
305 |
+
if allow_empty is True and update_vocab is True and "" not in self.word_to_id:
|
306 |
+
idx = max(self.word_to_id.values(), default=-1) + 1
|
307 |
+
self.word_to_id[""] = idx
|
308 |
+
|
309 |
+
if using_stemmer:
|
310 |
+
if "" not in self.word_to_stem:
|
311 |
+
self.word_to_stem[""] = ""
|
312 |
+
if "" not in self.stem_to_sid:
|
313 |
+
self.stem_to_sid[""] = idx
|
314 |
+
|
315 |
+
for text in texts:
|
316 |
+
if self.lower:
|
317 |
+
text = text.lower()
|
318 |
+
|
319 |
+
splitted_words = list(self.splitter(text))
|
320 |
+
|
321 |
+
if allow_empty is True and len(splitted_words) == 0:
|
322 |
+
splitted_words = [""]
|
323 |
+
|
324 |
+
doc_ids = []
|
325 |
+
for word in splitted_words:
|
326 |
+
if word in self.word_to_id:
|
327 |
+
wid = self.word_to_id[word]
|
328 |
+
doc_ids.append(wid)
|
329 |
+
continue
|
330 |
+
|
331 |
+
if using_stopwords and word in stopwords_set:
|
332 |
+
continue
|
333 |
+
|
334 |
+
# We are always updating the word_to_stem mapping since even new
|
335 |
+
# words that we have never seen before can be stemmed, with the
|
336 |
+
# possibility that the stemmed ID is already in the stem_to_sid
|
337 |
+
if using_stemmer:
|
338 |
+
if word in self.word_to_stem:
|
339 |
+
stem = self.word_to_stem[word]
|
340 |
+
else:
|
341 |
+
stem = self.stemmer(word)
|
342 |
+
self.word_to_stem[word] = stem
|
343 |
+
|
344 |
+
# if the stem is already in the stem_to_sid, we can just use the ID
|
345 |
+
# and update the word_to_id dictionary, unless update_vocab is "never"
|
346 |
+
# in which case we skip this word
|
347 |
+
if update_vocab != "never" and stem in self.stem_to_sid:
|
348 |
+
sid = self.stem_to_sid[stem]
|
349 |
+
self.word_to_id[word] = sid
|
350 |
+
doc_ids.append(sid)
|
351 |
+
|
352 |
+
elif update_vocab is True:
|
353 |
+
sid = len(self.stem_to_sid)
|
354 |
+
self.stem_to_sid[stem] = sid
|
355 |
+
self.word_to_id[word] = sid
|
356 |
+
doc_ids.append(sid)
|
357 |
+
else:
|
358 |
+
# if we are not using a stemmer, we can just update the word_to_id
|
359 |
+
# directly rather than going through the stem_to_sid dictionary
|
360 |
+
if update_vocab is True and word not in self.word_to_id:
|
361 |
+
wid = len(self.word_to_id)
|
362 |
+
self.word_to_id[word] = wid
|
363 |
+
doc_ids.append(wid)
|
364 |
+
|
365 |
+
if len(doc_ids) == 0 and allow_empty is True and "" in self.word_to_id:
|
366 |
+
doc_ids = [self.word_to_id[""]]
|
367 |
+
|
368 |
+
yield doc_ids
|
369 |
+
|
370 |
+
def tokenize(
|
371 |
+
self,
|
372 |
+
texts: List[str],
|
373 |
+
update_vocab: Union[bool, str] = "if_empty",
|
374 |
+
leave_progress: bool = False,
|
375 |
+
show_progress: bool = True,
|
376 |
+
length: Union[int, None] = None,
|
377 |
+
return_as: str = "ids",
|
378 |
+
allow_empty: bool = True,
|
379 |
+
) -> Union[List[List[int]], List[List[str]], typing.Generator, Tokenized]:
|
380 |
+
"""
|
381 |
+
Tokenize a list of strings and return the token IDs.
|
382 |
+
|
383 |
+
Parameters
|
384 |
+
----------
|
385 |
+
texts : List[str]
|
386 |
+
A list of strings to tokenize.
|
387 |
+
|
388 |
+
update_vocab : bool, optional
|
389 |
+
Whether to update the vocabulary dictionary with the new tokens. If true,
|
390 |
+
the different dictionaries making up the vocabulary will be updated with the
|
391 |
+
new tokens. If False, the vocabulary will not be updated unless you have a stemmer
|
392 |
+
and the stemmed word is in the stem_to_sid dictionary. If update_vocab="if_empty",
|
393 |
+
the function will only update the vocabulary if it is empty, i.e. when the
|
394 |
+
function is called for the first time, or if the vocabulary has been reset with
|
395 |
+
the `reset_vocab` method. If update_vocab="never", the "word_to_id" will never
|
396 |
+
be updated, even if the stemmed word is in the stem_to_sid dictionary. Only use
|
397 |
+
this if you are sure that the stemmed words are already in the stem_to_sid dictionary.
|
398 |
+
|
399 |
+
leave_progress : bool, optional
|
400 |
+
Whether to leave the progress bar after completion. If False, the progress bar
|
401 |
+
will disappear after completion. If True, the progress bar will stay on the screen.
|
402 |
+
|
403 |
+
show_progress : bool, optional
|
404 |
+
Whether to show the progress bar for tokenization. If False, the function will
|
405 |
+
not show the progress bar. If True, it will use tqdm.auto to show the progress bar.
|
406 |
+
|
407 |
+
length : int, optional
|
408 |
+
The length of the texts. If None, the function will call `len(texts)` to get the length.
|
409 |
+
This is mainly used when `texts` is a generator or a stream instead of a list, in which case
|
410 |
+
`len(texts)` will raise a TypeError, and you need to provide the length manually.
|
411 |
+
|
412 |
+
return_as : str, optional
|
413 |
+
The type of object to return by this function.
|
414 |
+
If "tuple", this returns a Tokenized namedtuple, which contains the token IDs
|
415 |
+
and the vocab dictionary.
|
416 |
+
If "string", this return a list of lists of strings, each string being a token.
|
417 |
+
If "ids", this return a list of lists of integers corresponding to the token IDs,
|
418 |
+
or stemmed IDs if a stemmer is used.
|
419 |
+
|
420 |
+
allow_empty : bool, optional
|
421 |
+
Whether to allow the splitter to return an empty string. If False, the splitter
|
422 |
+
will return an empty list, which may cause issues if the tokenizer is not expecting
|
423 |
+
an empty list. If True, the splitter will return a list with a single empty string.
|
424 |
+
|
425 |
+
Returns
|
426 |
+
-------
|
427 |
+
List[List[int]] or Generator[List[int]] or List[List[str]] or Tokenized object
|
428 |
+
If `return_as="stream"`, a Generator[List[int]] is returned, each integer being a token ID.
|
429 |
+
If `return_as="ids"`, a List[List[int]] is returned, each integer being a token ID.
|
430 |
+
If `return_as="string"`, a List[List[str]] is returned, each string being a token.
|
431 |
+
If `return_as="tuple"`, a Tokenized namedtuple is returned, with names `ids` and `vocab`.
|
432 |
+
"""
|
433 |
+
incorrect_return_error = (
|
434 |
+
"return_as must be either 'tuple', 'string', 'ids', or 'stream'."
|
435 |
+
)
|
436 |
+
incorrect_update_vocab_error = (
|
437 |
+
"update_vocab must be either True, False, 'if_empty', or 'never'."
|
438 |
+
)
|
439 |
+
if return_as not in ["tuple", "string", "ids", "stream"]:
|
440 |
+
raise ValueError(incorrect_return_error)
|
441 |
+
|
442 |
+
if update_vocab not in [True, False, "if_empty", "never"]:
|
443 |
+
raise ValueError(incorrect_update_vocab_error)
|
444 |
+
|
445 |
+
if update_vocab == "if_empty":
|
446 |
+
update_vocab = len(self.word_to_id) == 0
|
447 |
+
|
448 |
+
stream_fn = self.streaming_tokenize(
|
449 |
+
texts=texts, update_vocab=update_vocab, allow_empty=allow_empty
|
450 |
+
)
|
451 |
+
|
452 |
+
if return_as == "stream":
|
453 |
+
return stream_fn
|
454 |
+
|
455 |
+
if length is None:
|
456 |
+
length = len(texts)
|
457 |
+
|
458 |
+
tqdm_kwargs = dict(
|
459 |
+
desc="Tokenize texts",
|
460 |
+
leave=leave_progress,
|
461 |
+
disable=not show_progress,
|
462 |
+
total=length,
|
463 |
+
)
|
464 |
+
|
465 |
+
token_ids = []
|
466 |
+
for doc_ids in tqdm(stream_fn, **tqdm_kwargs):
|
467 |
+
token_ids.append(doc_ids)
|
468 |
+
|
469 |
+
if return_as == "ids":
|
470 |
+
return token_ids
|
471 |
+
elif return_as == "string":
|
472 |
+
return self.decode(token_ids)
|
473 |
+
elif return_as == "tuple":
|
474 |
+
return self.to_tokenized_tuple(token_ids)
|
475 |
+
else:
|
476 |
+
raise ValueError(incorrect_return_error)
|
477 |
+
|
478 |
+
def get_vocab_dict(self) -> Dict[str, Any]:
|
479 |
+
if self.stemmer is None:
|
480 |
+
# if we are not using a stemmer, we return the word_to_id dictionary
|
481 |
+
# which maps the words to the word IDs
|
482 |
+
return self.word_to_id
|
483 |
+
else:
|
484 |
+
# if we are using a stemmer, we return the stem_to_sid dictionary,
|
485 |
+
# which we will use to map the stemmed words to the stemmed IDs
|
486 |
+
return self.stem_to_sid
|
487 |
+
|
488 |
+
def to_tokenized_tuple(self, docs: List[List[int]]) -> Tokenized:
|
489 |
+
"""
|
490 |
+
Convert the token IDs to a Tokenized namedtuple, which contains the word IDs, or the stemmed IDs
|
491 |
+
if a stemmer is used. The Tokenized namedtuple contains two fields: ids and vocab. The latter
|
492 |
+
is a dictionary mapping the token IDs to the tokens, or a dictionary mapping the stemmed IDs to
|
493 |
+
the stemmed tokens (if a stemmer is used).
|
494 |
+
"""
|
495 |
+
return Tokenized(ids=docs, vocab=self.get_vocab_dict())
|
496 |
+
|
497 |
+
def decode(self, docs: List[List[int]]) -> List[List[str]]:
|
498 |
+
"""
|
499 |
+
Convert word IDs (or stemmed IDs if a stemmer is used) back to strings using the vocab dictionary,
|
500 |
+
which is a dictionary mapping the word IDs to the words or a dictionary mapping the stemmed IDs
|
501 |
+
to the stemmed words (if a stemmer is used).
|
502 |
+
|
503 |
+
Parameters
|
504 |
+
----------
|
505 |
+
docs : List[List[int]]
|
506 |
+
A list of lists of word IDs or stemmed IDs.
|
507 |
+
|
508 |
+
Returns
|
509 |
+
-------
|
510 |
+
List[List[str]]
|
511 |
+
A list of lists of strings, each string being a word or a stemmed word if a stemmer is used.
|
512 |
+
"""
|
513 |
+
vocab = self.get_vocab_dict()
|
514 |
+
reverse_vocab = {v: k for k, v in vocab.items()}
|
515 |
+
return [[reverse_vocab[token_id] for token_id in doc] for doc in docs]
|
516 |
+
|
517 |
+
|
518 |
+
def convert_tokenized_to_string_list(tokenized: Tokenized) -> List[List[str]]:
|
519 |
+
"""
|
520 |
+
Convert the token IDs back to strings using the vocab dictionary.
|
521 |
+
"""
|
522 |
+
reverse_vocab = {v: k for k, v in tokenized.vocab.items()}
|
523 |
+
|
524 |
+
return [
|
525 |
+
[reverse_vocab[token_id] for token_id in doc_ids] for doc_ids in tokenized.ids
|
526 |
+
]
|
527 |
+
|
528 |
+
|
529 |
+
def _infer_stopwords(stopwords: Union[str, List[str]]) -> Union[List[str], tuple]:
|
530 |
+
# Source of stopwords: https://github.com/nltk/nltk/blob/96ee715997e1c8d9148b6d8e1b32f412f31c7ff7/nltk/corpus/__init__.py#L315
|
531 |
+
if stopwords in ["english", "en", True]: # True is added to support the default
|
532 |
+
return STOPWORDS_EN
|
533 |
+
elif stopwords in ["english_plus", "en_plus"]:
|
534 |
+
return STOPWORDS_EN_PLUS
|
535 |
+
elif stopwords in ["german", "de"]:
|
536 |
+
return STOPWORDS_GERMAN
|
537 |
+
elif stopwords in ["dutch", "nl"]:
|
538 |
+
return STOPWORDS_DUTCH
|
539 |
+
elif stopwords in ["french", "fr"]:
|
540 |
+
return STOPWORDS_FRENCH
|
541 |
+
elif stopwords in ["spanish", "es"]:
|
542 |
+
return STOPWORDS_SPANISH
|
543 |
+
elif stopwords in ["portuguese", "pt"]:
|
544 |
+
return STOPWORDS_PORTUGUESE
|
545 |
+
elif stopwords in ["italian", "it"]:
|
546 |
+
return STOPWORDS_ITALIAN
|
547 |
+
elif stopwords in ["russian", "ru"]:
|
548 |
+
return STOPWORDS_RUSSIAN
|
549 |
+
elif stopwords in ["swedish", "sv"]:
|
550 |
+
return STOPWORDS_SWEDISH
|
551 |
+
elif stopwords in ["norwegian", "no"]:
|
552 |
+
return STOPWORDS_NORWEGIAN
|
553 |
+
elif stopwords in ["chinese", "zh"]:
|
554 |
+
return STOPWORDS_CHINESE
|
555 |
+
elif stopwords in [None, False]:
|
556 |
+
return []
|
557 |
+
elif isinstance(stopwords, str):
|
558 |
+
raise ValueError(
|
559 |
+
f"{stopwords} not recognized. Only English stopwords as default, German, Dutch, French, Spanish, Portuguese, Italian, Russian, Swedish, Norwegian, and Chinese are currently supported. "
|
560 |
+
"Please input a list of stopwords"
|
561 |
+
)
|
562 |
+
else:
|
563 |
+
return stopwords
|
564 |
+
|
565 |
+
|
566 |
+
def tokenize(
|
567 |
+
texts: Union[str, List[str]],
|
568 |
+
lower: bool = True,
|
569 |
+
token_pattern: str = r"(?u)\b\w\w+\b",
|
570 |
+
stopwords: Union[str, List[str]] = "english",
|
571 |
+
stemmer: Callable = None, # type: ignore
|
572 |
+
return_ids: bool = True,
|
573 |
+
show_progress: bool = True,
|
574 |
+
leave: bool = False,
|
575 |
+
allow_empty: bool = True,
|
576 |
+
) -> Union[List[List[str]], Tokenized]:
|
577 |
+
"""
|
578 |
+
Tokenize a list using the same method as the scikit-learn CountVectorizer,
|
579 |
+
and optionally apply a stemmer to the tokens or stopwords removal.
|
580 |
+
|
581 |
+
If you provide stemmer, it must have a `stemWords` method, or be callable
|
582 |
+
that takes a list of strings and returns a list of strings. If your stemmer
|
583 |
+
can only be called on a single word, you can use a lambda function to wrap it,
|
584 |
+
e.g. `lambda lst: list(map(stemmer.stem, lst))`.
|
585 |
+
|
586 |
+
If return_ids is True, the function will return a namedtuple with: (1) the tokenized
|
587 |
+
IDs and (2) the token_to_index dictionary. You can access the tokenized IDs using
|
588 |
+
the `ids` attribute and the token_to_index dictionary using the `vocab` attribute,
|
589 |
+
You can also destructure the namedtuple to get the ids and vocab_dict variables,
|
590 |
+
e.g. `token_ids, vocab = tokenize(...)`.
|
591 |
+
|
592 |
+
Parameters
|
593 |
+
----------
|
594 |
+
texts : Union[str, List[str]]
|
595 |
+
A list of strings to tokenize. If a single string is provided, it will be
|
596 |
+
converted to a list with a single element.
|
597 |
+
|
598 |
+
lower : bool, optional
|
599 |
+
Whether to convert the text to lowercase before tokenization
|
600 |
+
|
601 |
+
token_pattern : str, optional
|
602 |
+
The regex pattern to use for tokenization, by default, r"(?u)\\b\\w\\w+\\b"
|
603 |
+
|
604 |
+
stopwords : Union[str, List[str]], optional
|
605 |
+
The list of stopwords to remove from the text. If "english" or "en" is provided,
|
606 |
+
the function will use the default English stopwords
|
607 |
+
|
608 |
+
stemmer : Callable, optional
|
609 |
+
The stemmer to use for stemming the tokens. It is recommended
|
610 |
+
to use the PyStemmer library for stemming, but you can also any callable that
|
611 |
+
takes a list of strings and returns a list of strings.
|
612 |
+
|
613 |
+
return_ids : bool, optional
|
614 |
+
Whether to return the tokenized IDs and the vocab dictionary. If False, the
|
615 |
+
function will return the tokenized strings. If True, the function will return
|
616 |
+
a namedtuple with the tokenized IDs and the vocab dictionary.
|
617 |
+
|
618 |
+
show_progress : bool, optional
|
619 |
+
Whether to show the progress bar for tokenization. If False, the function will
|
620 |
+
not show the progress bar. If True, it will use tqdm.auto to show the progress bar.
|
621 |
+
|
622 |
+
leave : bool, optional
|
623 |
+
Whether to leave the progress bar after completion. If False, the progress bar
|
624 |
+
will disappear after completion. If True, the progress bar will stay on the screen.
|
625 |
+
|
626 |
+
allow_empty : bool, optional
|
627 |
+
Whether to allow the splitter to return an empty string. If False, the splitter
|
628 |
+
will return an empty list, which may cause issues if the tokenizer is not expecting
|
629 |
+
an empty list. If True, the splitter will return a list with a single empty string.
|
630 |
+
Note
|
631 |
+
-----
|
632 |
+
You may pass a single string or a list of strings. If you pass a single string,
|
633 |
+
this function will convert it to a list of strings with a single element.
|
634 |
+
"""
|
635 |
+
if isinstance(texts, str):
|
636 |
+
texts = [texts]
|
637 |
+
|
638 |
+
split_fn = re.compile(token_pattern).findall
|
639 |
+
stopwords = _infer_stopwords(stopwords)
|
640 |
+
|
641 |
+
# Step 1: Split the strings using the regex pattern
|
642 |
+
corpus_ids = []
|
643 |
+
token_to_index = {}
|
644 |
+
|
645 |
+
for text in tqdm(
|
646 |
+
texts, desc="Split strings", leave=leave, disable=not show_progress
|
647 |
+
):
|
648 |
+
stopwords_set = set(stopwords)
|
649 |
+
if lower:
|
650 |
+
text = text.lower()
|
651 |
+
|
652 |
+
splitted = split_fn(text)
|
653 |
+
|
654 |
+
if allow_empty is False and len(splitted) == 0:
|
655 |
+
splitted = [""]
|
656 |
+
|
657 |
+
doc_ids = []
|
658 |
+
|
659 |
+
for token in splitted:
|
660 |
+
if token in stopwords_set:
|
661 |
+
continue
|
662 |
+
|
663 |
+
if token not in token_to_index:
|
664 |
+
token_to_index[token] = len(token_to_index)
|
665 |
+
|
666 |
+
token_id = token_to_index[token]
|
667 |
+
doc_ids.append(token_id)
|
668 |
+
|
669 |
+
corpus_ids.append(doc_ids)
|
670 |
+
|
671 |
+
# Create a list of unique tokens that we will use to create the vocabulary
|
672 |
+
unique_tokens = list(token_to_index.keys())
|
673 |
+
|
674 |
+
# Step 2: Stem the tokens if a stemmer is provided
|
675 |
+
if stemmer is not None:
|
676 |
+
if hasattr(stemmer, "stemWords"):
|
677 |
+
stemmer_fn = stemmer.stemWords
|
678 |
+
elif callable(stemmer):
|
679 |
+
stemmer_fn = stemmer
|
680 |
+
else:
|
681 |
+
error_msg = "Stemmer must have a `stemWord` method, or be callable. For example, you can use the PyStemmer library."
|
682 |
+
raise ValueError(error_msg)
|
683 |
+
|
684 |
+
# Now, we use the stemmer on the token_to_index dictionary to get the stemmed tokens
|
685 |
+
tokens_stemmed = stemmer_fn(unique_tokens)
|
686 |
+
vocab = set(tokens_stemmed)
|
687 |
+
vocab_dict = {token: i for i, token in enumerate(vocab)}
|
688 |
+
stem_id_to_stem = {v: k for k, v in vocab_dict.items()}
|
689 |
+
# We create a dictionary mapping the stemmed tokens to their index
|
690 |
+
doc_id_to_stem_id = {
|
691 |
+
token_to_index[token]: vocab_dict[stem]
|
692 |
+
for token, stem in zip(unique_tokens, tokens_stemmed)
|
693 |
+
}
|
694 |
+
|
695 |
+
# Now, we simply need to replace the tokens in the corpus with the stemmed tokens
|
696 |
+
for i, doc_ids in enumerate(
|
697 |
+
tqdm(corpus_ids, desc="Stem Tokens", leave=leave, disable=not show_progress)
|
698 |
+
):
|
699 |
+
corpus_ids[i] = [doc_id_to_stem_id[doc_id] for doc_id in doc_ids]
|
700 |
+
else:
|
701 |
+
vocab_dict = token_to_index
|
702 |
+
|
703 |
+
# Step 3: Return the tokenized IDs and the vocab dictionary or the tokenized strings
|
704 |
+
if return_ids:
|
705 |
+
return Tokenized(ids=corpus_ids, vocab=vocab_dict)
|
706 |
+
else:
|
707 |
+
# We need a reverse dictionary to convert the token IDs back to tokens
|
708 |
+
reverse_dict = stem_id_to_stem if stemmer is not None else unique_tokens
|
709 |
+
# We convert the token IDs back to tokens in-place
|
710 |
+
for i, token_ids in enumerate(
|
711 |
+
tqdm(
|
712 |
+
corpus_ids,
|
713 |
+
desc="Reconstructing token strings",
|
714 |
+
leave=leave,
|
715 |
+
disable=not show_progress,
|
716 |
+
)
|
717 |
+
):
|
718 |
+
corpus_ids[i] = [reverse_dict[token_id] for token_id in token_ids]
|
719 |
+
return corpus_ids
|