Upload model
Browse files- config.json +15 -0
- configuration_cased.py +26 -0
- modeling_cased.py +252 -0
- pytorch_model.bin +3 -0
- transforms_cased.py +438 -0
config.json
ADDED
@@ -0,0 +1,15 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"alpha": 0.5,
|
3 |
+
"architectures": [
|
4 |
+
"CaSEDModel"
|
5 |
+
],
|
6 |
+
"auto_map": {
|
7 |
+
"AutoConfig": "configuration_cased.CaSEDConfig",
|
8 |
+
"AutoModel": "modeling_cased.CaSEDModel"
|
9 |
+
},
|
10 |
+
"index_name": "cc12m",
|
11 |
+
"model_type": "cased",
|
12 |
+
"retrieval_num_results": 10,
|
13 |
+
"torch_dtype": "float32",
|
14 |
+
"transformers_version": "4.29.2"
|
15 |
+
}
|
configuration_cased.py
ADDED
@@ -0,0 +1,26 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
from transformers.modeling_utils import PretrainedConfig
|
2 |
+
|
3 |
+
|
4 |
+
class CaSEDConfig(PretrainedConfig):
|
5 |
+
"""Configuration class for CaSED.
|
6 |
+
|
7 |
+
Args:
|
8 |
+
index_name (str, optional): Name of the index. Defaults to "cc12m".
|
9 |
+
alpha (float, optional): Weight of the vision loss. Defaults to 0.5.
|
10 |
+
retrieval_num_results (int, optional): Number of results to return. Defaults to 10.
|
11 |
+
"""
|
12 |
+
|
13 |
+
model_type = "cased"
|
14 |
+
is_composition = True
|
15 |
+
|
16 |
+
def __init__(
|
17 |
+
self,
|
18 |
+
index_name: str = "cc12m",
|
19 |
+
alpha: float = 0.5,
|
20 |
+
retrieval_num_results: int = 10,
|
21 |
+
**kwargs,
|
22 |
+
):
|
23 |
+
super().__init__(**kwargs)
|
24 |
+
self.index_name = index_name
|
25 |
+
self.alpha = alpha
|
26 |
+
self.retrieval_num_results = retrieval_num_results
|
modeling_cased.py
ADDED
@@ -0,0 +1,252 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import tarfile
|
3 |
+
from pathlib import Path
|
4 |
+
from typing import Optional
|
5 |
+
|
6 |
+
import faiss
|
7 |
+
import numpy as np
|
8 |
+
import pyarrow as pa
|
9 |
+
import requests
|
10 |
+
import torch
|
11 |
+
from tqdm import tqdm
|
12 |
+
from transformers import CLIPModel, CLIPProcessor
|
13 |
+
from transformers.modeling_utils import PreTrainedModel
|
14 |
+
|
15 |
+
from .configuration_cased import CaSEDConfig
|
16 |
+
from .transforms_cased import default_vocabulary_transforms
|
17 |
+
|
18 |
+
DATABASES = {
|
19 |
+
"cc12m": {
|
20 |
+
"url": "https://storage-cased.alessandroconti.me/cc12m.tar.gz",
|
21 |
+
"cache_subdir": "./cc12m/vit-l-14/",
|
22 |
+
},
|
23 |
+
}
|
24 |
+
|
25 |
+
|
26 |
+
class MetadataProvider:
|
27 |
+
"""Metadata provider.
|
28 |
+
|
29 |
+
It uses arrow files to store metadata and retrieve it efficiently.
|
30 |
+
|
31 |
+
Code reference:
|
32 |
+
- https://github.dev/rom1504/clip-retrieval
|
33 |
+
"""
|
34 |
+
|
35 |
+
def __init__(self, arrow_folder: Path):
|
36 |
+
arrow_files = [str(a) for a in sorted(arrow_folder.glob("**/*")) if a.is_file()]
|
37 |
+
self.table = pa.concat_tables(
|
38 |
+
[
|
39 |
+
pa.ipc.RecordBatchFileReader(pa.memory_map(arrow_file, "r")).read_all()
|
40 |
+
for arrow_file in arrow_files
|
41 |
+
]
|
42 |
+
)
|
43 |
+
|
44 |
+
def get(self, ids: np.ndarray, cols: Optional[list] = None):
|
45 |
+
"""Get arrow metadata from ids.
|
46 |
+
|
47 |
+
Args:
|
48 |
+
ids (np.ndarray): Ids to retrieve.
|
49 |
+
cols (Optional[list], optional): Columns to retrieve. Defaults to None.
|
50 |
+
"""
|
51 |
+
if cols is None:
|
52 |
+
cols = self.table.schema.names
|
53 |
+
else:
|
54 |
+
cols = list(set(self.table.schema.names) & set(cols))
|
55 |
+
t = pa.concat_tables([self.table[i:j] for i, j in zip(ids, ids + 1)])
|
56 |
+
return t.select(cols).to_pandas().to_dict("records")
|
57 |
+
|
58 |
+
|
59 |
+
class CaSEDModel(PreTrainedModel):
|
60 |
+
"""Transformers module for Category Search from External Databases (CaSED).
|
61 |
+
|
62 |
+
Reference:
|
63 |
+
- Conti et al. Vocabulary-free Image Classification. arXiv 2023.
|
64 |
+
|
65 |
+
Args:
|
66 |
+
config (CaSEDConfig): Configuration class for CaSED.
|
67 |
+
"""
|
68 |
+
|
69 |
+
config_class = CaSEDConfig
|
70 |
+
|
71 |
+
def __init__(self, config: CaSEDConfig):
|
72 |
+
super().__init__(config)
|
73 |
+
|
74 |
+
# load CLIP
|
75 |
+
model = CLIPModel.from_pretrained("openai/clip-vit-large-patch14")
|
76 |
+
self.vision_encoder = model.vision_model
|
77 |
+
self.vision_proj = model.visual_projection
|
78 |
+
self.language_encoder = model.text_model
|
79 |
+
self.language_proj = model.text_projection
|
80 |
+
self.logit_scale = model.logit_scale.exp()
|
81 |
+
self.processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")
|
82 |
+
|
83 |
+
# load transforms
|
84 |
+
self.vocabulary_transforms = default_vocabulary_transforms()
|
85 |
+
|
86 |
+
# set hparams
|
87 |
+
self.hparams = {}
|
88 |
+
self.hparams["alpha"] = config.alpha
|
89 |
+
self.hparams["index_name"] = config.index_name
|
90 |
+
self.hparams["retrieval_num_results"] = config.retrieval_num_results
|
91 |
+
|
92 |
+
# set cache dir
|
93 |
+
self.hparams["cache_dir"] = Path(os.path.expanduser("~/.cache/cased"))
|
94 |
+
os.makedirs(self.hparams["cache_dir"], exist_ok=True)
|
95 |
+
|
96 |
+
# download databases
|
97 |
+
self.prepare_data()
|
98 |
+
|
99 |
+
# load faiss indices and metadata providers
|
100 |
+
self.resources = {}
|
101 |
+
for name, items in DATABASES.items():
|
102 |
+
database_path = self.hparams["cache_dir"] / "databases" / items["cache_subdir"]
|
103 |
+
text_index_fp = database_path / "text.index"
|
104 |
+
metadata_fp = database_path / "metadata/"
|
105 |
+
|
106 |
+
text_index = faiss.read_index(
|
107 |
+
str(text_index_fp), faiss.IO_FLAG_MMAP | faiss.IO_FLAG_READ_ONLY
|
108 |
+
)
|
109 |
+
metadata_provider = MetadataProvider(metadata_fp)
|
110 |
+
|
111 |
+
self.resources[name] = {
|
112 |
+
"device": self.device,
|
113 |
+
"model": "ViT-L-14",
|
114 |
+
"text_index": text_index,
|
115 |
+
"metadata_provider": metadata_provider,
|
116 |
+
}
|
117 |
+
|
118 |
+
def prepare_data(self):
|
119 |
+
"""Download data if needed."""
|
120 |
+
databases_path = Path(self.hparams["cache_dir"]) / "databases"
|
121 |
+
|
122 |
+
for name, items in DATABASES.items():
|
123 |
+
url = items["url"]
|
124 |
+
database_path = Path(databases_path, name)
|
125 |
+
if database_path.exists():
|
126 |
+
continue
|
127 |
+
|
128 |
+
# download data
|
129 |
+
target_path = Path(databases_path, name + ".tar.gz")
|
130 |
+
os.makedirs(target_path.parent, exist_ok=True)
|
131 |
+
with requests.get(url, stream=True) as r:
|
132 |
+
r.raise_for_status()
|
133 |
+
total_bytes_size = int(r.headers.get('content-length', 0))
|
134 |
+
chunk_size = 8192
|
135 |
+
p_bar = tqdm(
|
136 |
+
desc="Downloading cc12m index",
|
137 |
+
total=total_bytes_size,
|
138 |
+
unit='iB',
|
139 |
+
unit_scale=True,
|
140 |
+
)
|
141 |
+
with open(target_path, 'wb') as f:
|
142 |
+
for chunk in r.iter_content(chunk_size=chunk_size):
|
143 |
+
f.write(chunk)
|
144 |
+
p_bar.update(len(chunk))
|
145 |
+
p_bar.close()
|
146 |
+
|
147 |
+
# extract data
|
148 |
+
tar = tarfile.open(target_path, "r:gz")
|
149 |
+
tar.extractall(target_path.parent)
|
150 |
+
tar.close()
|
151 |
+
target_path.unlink()
|
152 |
+
|
153 |
+
@torch.no_grad()
|
154 |
+
def query_index(self, sample_z: torch.Tensor) -> torch.Tensor:
|
155 |
+
"""Query the external database index.
|
156 |
+
|
157 |
+
Args:
|
158 |
+
sample_z (torch.Tensor): Sample to query the index.
|
159 |
+
"""
|
160 |
+
# get the index
|
161 |
+
resources = self.resources[self.hparams["index_name"]]
|
162 |
+
text_index = resources["text_index"]
|
163 |
+
metadata_provider = resources["metadata_provider"]
|
164 |
+
|
165 |
+
# query the index
|
166 |
+
sample_z = sample_z.squeeze(0)
|
167 |
+
sample_z = sample_z / sample_z.norm(dim=-1, keepdim=True)
|
168 |
+
query_input = sample_z.cpu().detach().numpy().tolist()
|
169 |
+
query = np.expand_dims(np.array(query_input).astype("float32"), 0)
|
170 |
+
|
171 |
+
distances, idxs, _ = text_index.search_and_reconstruct(
|
172 |
+
query, self.hparams["retrieval_num_results"]
|
173 |
+
)
|
174 |
+
results = idxs[0]
|
175 |
+
nb_results = np.where(results == -1)[0]
|
176 |
+
nb_results = nb_results[0] if len(nb_results) > 0 else len(results)
|
177 |
+
indices = results[:nb_results]
|
178 |
+
distances = distances[0][:nb_results]
|
179 |
+
|
180 |
+
if len(distances) == 0:
|
181 |
+
return []
|
182 |
+
|
183 |
+
# get the metadata
|
184 |
+
results = []
|
185 |
+
metadata = metadata_provider.get(indices[:20], ["caption"])
|
186 |
+
for key, (d, i) in enumerate(zip(distances, indices)):
|
187 |
+
output = {}
|
188 |
+
meta = None if key + 1 > len(metadata) else metadata[key]
|
189 |
+
if meta is not None:
|
190 |
+
output.update(meta)
|
191 |
+
output["id"] = i.item()
|
192 |
+
output["similarity"] = d.item()
|
193 |
+
results.append(output)
|
194 |
+
|
195 |
+
# get the captions only
|
196 |
+
vocabularies = [result["caption"] for result in results]
|
197 |
+
|
198 |
+
return vocabularies
|
199 |
+
|
200 |
+
@torch.no_grad()
|
201 |
+
def forward(self, images: dict, alpha: Optional[float] = None) -> torch.Tensor():
|
202 |
+
"""Forward pass.
|
203 |
+
|
204 |
+
Args:
|
205 |
+
images (dict): Dictionary with the images. The expected keys are:
|
206 |
+
- pixel_values (torch.Tensor): Pixel values of the images.
|
207 |
+
alpha (Optional[float]): Alpha value for the interpolation.
|
208 |
+
"""
|
209 |
+
# forward the images
|
210 |
+
images["pixel_values"] = images["pixel_values"].to(self.device)
|
211 |
+
images_z = self.vision_proj(self.vision_encoder(**images)[1])
|
212 |
+
|
213 |
+
vocabularies, samples_p = [], []
|
214 |
+
for image_z in images_z:
|
215 |
+
# generate a single text embedding from the unfiltered vocabulary
|
216 |
+
vocabulary = self.query_index(image_z)
|
217 |
+
text = self.processor(text=vocabulary, return_tensors="pt", padding=True)
|
218 |
+
text["input_ids"] = text["input_ids"][:, :77].to(self.device)
|
219 |
+
text["attention_mask"] = text["attention_mask"][:, :77].to(self.device)
|
220 |
+
text_z = self.language_encoder(**text)[1]
|
221 |
+
text_z = self.language_proj(text_z)
|
222 |
+
|
223 |
+
# filter the vocabulary, embed it, and get its mean embedding
|
224 |
+
vocabulary = self.vocabulary_transforms(vocabulary) or ["object"]
|
225 |
+
text = self.processor(text=vocabulary, return_tensors="pt", padding=True)
|
226 |
+
text = {k: v.to(self.device) for k, v in text.items()}
|
227 |
+
vocabulary_z = self.language_encoder(**text)[1]
|
228 |
+
vocabulary_z = self.language_proj(vocabulary_z)
|
229 |
+
vocabulary_z = vocabulary_z / vocabulary_z.norm(dim=-1, keepdim=True)
|
230 |
+
|
231 |
+
# get the image and text predictions
|
232 |
+
image_z = image_z / image_z.norm(dim=-1, keepdim=True)
|
233 |
+
text_z = text_z / text_z.norm(dim=-1, keepdim=True)
|
234 |
+
image_p = (torch.matmul(image_z, vocabulary_z.T) * self.logit_scale).softmax(dim=-1)
|
235 |
+
text_p = (torch.matmul(text_z, vocabulary_z.T) * self.logit_scale).softmax(dim=-1)
|
236 |
+
|
237 |
+
# average the image and text predictions
|
238 |
+
alpha = alpha or self.hparams["alpha"]
|
239 |
+
sample_p = alpha * image_p + (1 - alpha) * text_p
|
240 |
+
|
241 |
+
# save the results
|
242 |
+
samples_p.append(sample_p)
|
243 |
+
vocabularies.append(vocabulary)
|
244 |
+
|
245 |
+
# get the scores
|
246 |
+
samples_p = torch.stack(samples_p, dim=0)
|
247 |
+
scores = sample_p.cpu().tolist()
|
248 |
+
|
249 |
+
# define the results
|
250 |
+
results = {"vocabularies": vocabularies, "scores": scores}
|
251 |
+
|
252 |
+
return results
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:91c5a2012ab49580ef33645ef578ab2eab491ace7ed63e856f9ef340f73e0e9e
|
3 |
+
size 1710665929
|
transforms_cased.py
ADDED
@@ -0,0 +1,438 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import re
|
2 |
+
from abc import ABC, abstractmethod
|
3 |
+
from typing import Any, Union
|
4 |
+
|
5 |
+
import inflect
|
6 |
+
import nltk
|
7 |
+
from flair.data import Sentence
|
8 |
+
from flair.models import SequenceTagger
|
9 |
+
|
10 |
+
__all__ = [
|
11 |
+
"DropFileExtensions",
|
12 |
+
"DropNonAlpha",
|
13 |
+
"DropShortWords",
|
14 |
+
"DropSpecialCharacters",
|
15 |
+
"DropTokens",
|
16 |
+
"DropURLs",
|
17 |
+
"DropWords",
|
18 |
+
"FilterPOS",
|
19 |
+
"FrequencyMinWordCount",
|
20 |
+
"FrequencyTopK",
|
21 |
+
"ReplaceSeparators",
|
22 |
+
"ToLowercase",
|
23 |
+
"ToSingular",
|
24 |
+
]
|
25 |
+
|
26 |
+
|
27 |
+
class BaseTextTransform(ABC):
|
28 |
+
"""Base class for string transforms."""
|
29 |
+
|
30 |
+
@abstractmethod
|
31 |
+
def __call__(self, text: str):
|
32 |
+
raise NotImplementedError
|
33 |
+
|
34 |
+
def __repr__(self) -> str:
|
35 |
+
return f"{self.__class__.__name__}()"
|
36 |
+
|
37 |
+
|
38 |
+
class DropFileExtensions(BaseTextTransform):
|
39 |
+
"""Remove file extensions from the input text."""
|
40 |
+
|
41 |
+
def __call__(self, text: str):
|
42 |
+
"""
|
43 |
+
Args:
|
44 |
+
text (str): Text to remove file extensions from.
|
45 |
+
"""
|
46 |
+
text = re.sub(r"\.\w+", "", text)
|
47 |
+
|
48 |
+
return text
|
49 |
+
|
50 |
+
|
51 |
+
class DropNonAlpha(BaseTextTransform):
|
52 |
+
"""Remove non-alpha words from the input text."""
|
53 |
+
|
54 |
+
def __call__(self, text: str):
|
55 |
+
"""
|
56 |
+
Args:
|
57 |
+
text (str): Text to remove non-alpha words from.
|
58 |
+
"""
|
59 |
+
text = re.sub(r"[^a-zA-Z\s]", "", text)
|
60 |
+
|
61 |
+
return text
|
62 |
+
|
63 |
+
|
64 |
+
class DropShortWords(BaseTextTransform):
|
65 |
+
"""Remove short words from the input text.
|
66 |
+
|
67 |
+
Args:
|
68 |
+
min_length (int): Minimum length of words to keep.
|
69 |
+
"""
|
70 |
+
|
71 |
+
def __init__(self, min_length) -> None:
|
72 |
+
super().__init__()
|
73 |
+
self.min_length = min_length
|
74 |
+
|
75 |
+
def __call__(self, text: str):
|
76 |
+
"""
|
77 |
+
Args:
|
78 |
+
text (str): Text to remove short words from.
|
79 |
+
"""
|
80 |
+
text = " ".join([word for word in text.split() if len(word) >= self.min_length])
|
81 |
+
|
82 |
+
return text
|
83 |
+
|
84 |
+
def __repr__(self) -> str:
|
85 |
+
return f"{self.__class__.__name__}(min_length={self.min_length})"
|
86 |
+
|
87 |
+
|
88 |
+
class DropSpecialCharacters(BaseTextTransform):
|
89 |
+
"""Remove special characters from the input text.
|
90 |
+
|
91 |
+
Special characters are defined as any character that is not a word character, whitespace,
|
92 |
+
hyphen, period, apostrophe, or ampersand.
|
93 |
+
"""
|
94 |
+
|
95 |
+
def __call__(self, text: str):
|
96 |
+
"""
|
97 |
+
Args:
|
98 |
+
text (str): Text to remove special characters from.
|
99 |
+
"""
|
100 |
+
text = re.sub(r"[^\w\s\-\.\'\&]", "", text)
|
101 |
+
|
102 |
+
return text
|
103 |
+
|
104 |
+
|
105 |
+
class DropTokens(BaseTextTransform):
|
106 |
+
"""Remove tokens from the input text.
|
107 |
+
|
108 |
+
Tokens are defined as strings enclosed in angle brackets, e.g. <token>.
|
109 |
+
"""
|
110 |
+
|
111 |
+
def __call__(self, text: str):
|
112 |
+
"""
|
113 |
+
Args:
|
114 |
+
text (str): Text to remove tokens from.
|
115 |
+
"""
|
116 |
+
text = re.sub(r"<[^>]+>", "", text)
|
117 |
+
|
118 |
+
return text
|
119 |
+
|
120 |
+
|
121 |
+
class DropURLs(BaseTextTransform):
|
122 |
+
"""Remove URLs from the input text."""
|
123 |
+
|
124 |
+
def __call__(self, text: str):
|
125 |
+
"""
|
126 |
+
Args:
|
127 |
+
text (str): Text to remove URLs from.
|
128 |
+
"""
|
129 |
+
text = re.sub(r"http\S+", "", text)
|
130 |
+
|
131 |
+
return text
|
132 |
+
|
133 |
+
|
134 |
+
class DropWords(BaseTextTransform):
|
135 |
+
"""Remove words from the input text.
|
136 |
+
|
137 |
+
It is case-insensitive and supports singular and plural forms of the words.
|
138 |
+
"""
|
139 |
+
|
140 |
+
def __init__(self, words: list[str]) -> None:
|
141 |
+
super().__init__()
|
142 |
+
self.words = words
|
143 |
+
self.pattern = r"\b(?:{})\b".format("|".join(words))
|
144 |
+
|
145 |
+
def __call__(self, text: str):
|
146 |
+
"""
|
147 |
+
Args:
|
148 |
+
text (str): Text to remove words from.
|
149 |
+
"""
|
150 |
+
text = re.sub(self.pattern, "", text, flags=re.IGNORECASE)
|
151 |
+
|
152 |
+
return text
|
153 |
+
|
154 |
+
def __repr__(self) -> str:
|
155 |
+
return f"{self.__class__.__name__}(pattern={self.pattern})"
|
156 |
+
|
157 |
+
|
158 |
+
class FilterPOS(BaseTextTransform):
|
159 |
+
"""Filter words by POS tags.
|
160 |
+
|
161 |
+
Args:
|
162 |
+
tags (list): List of POS tags to remove.
|
163 |
+
engine (str): POS tagger to use. Must be one of "nltk" or "flair". Defaults to "nltk".
|
164 |
+
keep_compound_nouns (bool): Whether to keep composed words. Defaults to True.
|
165 |
+
"""
|
166 |
+
|
167 |
+
def __init__(self, tags: list, engine: str = "nltk", keep_compound_nouns: bool = True) -> None:
|
168 |
+
super().__init__()
|
169 |
+
self.tags = tags
|
170 |
+
self.engine = engine
|
171 |
+
self.keep_compound_nouns = keep_compound_nouns
|
172 |
+
|
173 |
+
if engine == "nltk":
|
174 |
+
nltk.download("averaged_perceptron_tagger", quiet=True)
|
175 |
+
nltk.download("punkt", quiet=True)
|
176 |
+
self.tagger = lambda x: nltk.pos_tag(nltk.word_tokenize(x))
|
177 |
+
elif engine == "flair":
|
178 |
+
self.tagger = SequenceTagger.load("flair/pos-english-fast").predict
|
179 |
+
|
180 |
+
def __call__(self, text: str):
|
181 |
+
"""
|
182 |
+
Args:
|
183 |
+
text (str): Text to remove words with specific POS tags from.
|
184 |
+
"""
|
185 |
+
if self.engine == "nltk":
|
186 |
+
word_tags = self.tagger(text)
|
187 |
+
text = " ".join([word for word, tag in word_tags if tag not in self.tags])
|
188 |
+
elif self.engine == "flair":
|
189 |
+
sentence = Sentence(text)
|
190 |
+
self.tagger(sentence)
|
191 |
+
text = " ".join([token.text for token in sentence.tokens if token.tag in self.tags])
|
192 |
+
|
193 |
+
if self.keep_compound_nouns:
|
194 |
+
compound_nouns = []
|
195 |
+
|
196 |
+
if self.engine == "nltk":
|
197 |
+
for i in range(len(word_tags) - 1):
|
198 |
+
if word_tags[i][1] == "NN" and word_tags[i + 1][1] == "NN":
|
199 |
+
# if they are the same word, skip
|
200 |
+
if word_tags[i][0] == word_tags[i + 1][0]:
|
201 |
+
continue
|
202 |
+
|
203 |
+
compound_noun = word_tags[i][0] + "_" + word_tags[i + 1][0]
|
204 |
+
compound_nouns.append(compound_noun)
|
205 |
+
elif self.engine == "flair":
|
206 |
+
for i in range(len(sentence.tokens) - 1):
|
207 |
+
if sentence.tokens[i].tag == "NN" and sentence.tokens[i + 1].tag == "NN":
|
208 |
+
# if they are the same word, skip
|
209 |
+
if sentence.tokens[i].text == sentence.tokens[i + 1].text:
|
210 |
+
continue
|
211 |
+
|
212 |
+
compound_noun = sentence.tokens[i].text + "_" + sentence.tokens[i + 1].text
|
213 |
+
compound_nouns.append(compound_noun)
|
214 |
+
|
215 |
+
text = " ".join([text, " ".join(compound_nouns)])
|
216 |
+
|
217 |
+
return text
|
218 |
+
|
219 |
+
def __repr__(self) -> str:
|
220 |
+
return f"{self.__class__.__name__}(tags={self.tags}, engine={self.engine})"
|
221 |
+
|
222 |
+
|
223 |
+
class FrequencyMinWordCount(BaseTextTransform):
|
224 |
+
"""Keep only words that occur more than a minimum number of times in the input text.
|
225 |
+
|
226 |
+
If the threshold is too strong and no words pass the threshold, the threshold is reduced to
|
227 |
+
the most frequent word.
|
228 |
+
|
229 |
+
Args:
|
230 |
+
min_count (int): Minimum number of occurrences of a word to keep.
|
231 |
+
"""
|
232 |
+
|
233 |
+
def __init__(self, min_count) -> None:
|
234 |
+
super().__init__()
|
235 |
+
self.min_count = min_count
|
236 |
+
|
237 |
+
def __call__(self, text: str):
|
238 |
+
"""
|
239 |
+
Args:
|
240 |
+
text (str): Text to remove infrequent words from.
|
241 |
+
"""
|
242 |
+
if self.min_count <= 1:
|
243 |
+
return text
|
244 |
+
|
245 |
+
words = text.split()
|
246 |
+
word_counts = {word: words.count(word) for word in words}
|
247 |
+
|
248 |
+
# if nothing passes the threshold, reduce the threshold to the most frequent word
|
249 |
+
max_word_count = max(word_counts.values() or [0])
|
250 |
+
min_count = max_word_count if self.min_count > max_word_count else self.min_count
|
251 |
+
|
252 |
+
text = " ".join([word for word in words if word_counts[word] >= min_count])
|
253 |
+
|
254 |
+
return text
|
255 |
+
|
256 |
+
def __repr__(self) -> str:
|
257 |
+
return f"{self.__class__.__name__}(min_count={self.min_count})"
|
258 |
+
|
259 |
+
|
260 |
+
class FrequencyTopK(BaseTextTransform):
|
261 |
+
"""Keep only the top k most frequent words in the input text.
|
262 |
+
|
263 |
+
In case of a tie, all words with the same count as the last word are kept.
|
264 |
+
|
265 |
+
Args:
|
266 |
+
top_k (int): Number of top words to keep.
|
267 |
+
"""
|
268 |
+
|
269 |
+
def __init__(self, top_k: int) -> None:
|
270 |
+
super().__init__()
|
271 |
+
self.top_k = top_k
|
272 |
+
|
273 |
+
def __call__(self, text: str):
|
274 |
+
"""
|
275 |
+
Args:
|
276 |
+
text (str): Text to remove infrequent words from.
|
277 |
+
"""
|
278 |
+
if self.top_k < 1:
|
279 |
+
return text
|
280 |
+
|
281 |
+
words = text.split()
|
282 |
+
word_counts = {word: words.count(word) for word in words}
|
283 |
+
top_words = sorted(word_counts, key=word_counts.get, reverse=True)
|
284 |
+
|
285 |
+
# in case of a tie, keep all words with the same count
|
286 |
+
top_words = top_words[: self.top_k]
|
287 |
+
top_words = [word for word in top_words if word_counts[word] == word_counts[top_words[-1]]]
|
288 |
+
|
289 |
+
text = " ".join([word for word in words if word in top_words])
|
290 |
+
|
291 |
+
return text
|
292 |
+
|
293 |
+
def __repr__(self) -> str:
|
294 |
+
return f"{self.__class__.__name__}(top_k={self.top_k})"
|
295 |
+
|
296 |
+
|
297 |
+
class ReplaceSeparators(BaseTextTransform):
|
298 |
+
"""Replace underscores and dashes with spaces."""
|
299 |
+
|
300 |
+
def __call__(self, text: str):
|
301 |
+
"""
|
302 |
+
Args:
|
303 |
+
text (str): Text to replace separators in.
|
304 |
+
"""
|
305 |
+
text = re.sub(r"[_\-]", " ", text)
|
306 |
+
|
307 |
+
return text
|
308 |
+
|
309 |
+
def __repr__(self) -> str:
|
310 |
+
return f"{self.__class__.__name__}()"
|
311 |
+
|
312 |
+
|
313 |
+
class RemoveDuplicates(BaseTextTransform):
|
314 |
+
"""Remove duplicate words from the input text."""
|
315 |
+
|
316 |
+
def __call__(self, text: str):
|
317 |
+
"""
|
318 |
+
Args:
|
319 |
+
text (str): Text to remove duplicate words from.
|
320 |
+
"""
|
321 |
+
text = " ".join(list(set(text.split())))
|
322 |
+
|
323 |
+
return text
|
324 |
+
|
325 |
+
|
326 |
+
class TextCompose:
|
327 |
+
"""Compose several transforms together.
|
328 |
+
|
329 |
+
It differs from the torchvision.transforms.Compose class in that it applies the transforms to
|
330 |
+
a string instead of a PIL Image or Tensor. In addition, it automatically join the list of
|
331 |
+
input strings into a single string and splits the output string into a list of words.
|
332 |
+
|
333 |
+
Args:
|
334 |
+
transforms (list): List of transforms to compose.
|
335 |
+
"""
|
336 |
+
|
337 |
+
def __init__(self, transforms: list[BaseTextTransform]) -> None:
|
338 |
+
self.transforms = transforms
|
339 |
+
|
340 |
+
def __call__(self, text: Union[str, list[str]]) -> Any:
|
341 |
+
if isinstance(text, list):
|
342 |
+
text = " ".join(text)
|
343 |
+
|
344 |
+
for t in self.transforms:
|
345 |
+
text = t(text)
|
346 |
+
return text.split()
|
347 |
+
|
348 |
+
def __repr__(self) -> str:
|
349 |
+
format_string = self.__class__.__name__ + "("
|
350 |
+
for t in self.transforms:
|
351 |
+
format_string += "\n"
|
352 |
+
format_string += f" {t}"
|
353 |
+
format_string += "\n)"
|
354 |
+
return format_string
|
355 |
+
|
356 |
+
|
357 |
+
class ToLowercase(BaseTextTransform):
|
358 |
+
"""Convert text to lowercase."""
|
359 |
+
|
360 |
+
def __call__(self, text: str):
|
361 |
+
"""
|
362 |
+
Args:
|
363 |
+
text (str): Text to convert to lowercase.
|
364 |
+
"""
|
365 |
+
text = text.lower()
|
366 |
+
|
367 |
+
return text
|
368 |
+
|
369 |
+
|
370 |
+
class ToSingular(BaseTextTransform):
|
371 |
+
"""Convert plural words to singular form."""
|
372 |
+
|
373 |
+
def __init__(self) -> None:
|
374 |
+
super().__init__()
|
375 |
+
self.transform = inflect.engine().singular_noun
|
376 |
+
|
377 |
+
def __call__(self, text: str):
|
378 |
+
"""
|
379 |
+
Args:
|
380 |
+
text (str): Text to convert to singular form.
|
381 |
+
"""
|
382 |
+
words = text.split()
|
383 |
+
for i, word in enumerate(words):
|
384 |
+
if not word.endswith("s"):
|
385 |
+
continue
|
386 |
+
|
387 |
+
if word[-2:] in ["ss", "us", "is"]:
|
388 |
+
continue
|
389 |
+
|
390 |
+
if word[-3:] in ["ies", "oes"]:
|
391 |
+
continue
|
392 |
+
|
393 |
+
words[i] = self.transform(word) or word
|
394 |
+
|
395 |
+
text = " ".join(words)
|
396 |
+
|
397 |
+
return text
|
398 |
+
|
399 |
+
def __repr__(self) -> str:
|
400 |
+
return f"{self.__class__.__name__}()"
|
401 |
+
|
402 |
+
|
403 |
+
def default_vocabulary_transforms() -> TextCompose:
|
404 |
+
"""Preprocess input text with preprocessing transforms."""
|
405 |
+
words_to_drop = [
|
406 |
+
"image",
|
407 |
+
"photo",
|
408 |
+
"picture",
|
409 |
+
"thumbnail",
|
410 |
+
"logo",
|
411 |
+
"symbol",
|
412 |
+
"clipart",
|
413 |
+
"portrait",
|
414 |
+
"painting",
|
415 |
+
"illustration",
|
416 |
+
"icon",
|
417 |
+
"profile",
|
418 |
+
]
|
419 |
+
pos_tags = ["NN", "NNS", "NNP", "NNPS", "JJ", "JJR", "JJS", "VBG", "VBN"]
|
420 |
+
|
421 |
+
transforms = []
|
422 |
+
transforms.append(DropTokens())
|
423 |
+
transforms.append(DropURLs())
|
424 |
+
transforms.append(DropSpecialCharacters())
|
425 |
+
transforms.append(DropFileExtensions())
|
426 |
+
transforms.append(ReplaceSeparators())
|
427 |
+
transforms.append(DropShortWords(min_length=3))
|
428 |
+
transforms.append(DropNonAlpha())
|
429 |
+
transforms.append(ToLowercase())
|
430 |
+
transforms.append(ToSingular())
|
431 |
+
transforms.append(DropWords(words=words_to_drop))
|
432 |
+
transforms.append(FrequencyMinWordCount(min_count=2))
|
433 |
+
transforms.append(FilterPOS(tags=pos_tags, engine="flair", keep_compound_nouns=False))
|
434 |
+
transforms.append(RemoveDuplicates())
|
435 |
+
|
436 |
+
transforms = TextCompose(transforms)
|
437 |
+
|
438 |
+
return transforms
|