Datasets:

ArneBinder commited on
Commit
e917db0
1 Parent(s): 70446e7

replace subdirecty_mapping by split_paths and allow list of files for values

Browse files
Files changed (1) hide show
  1. brat.py +28 -14
brat.py CHANGED
@@ -2,7 +2,7 @@ import glob
2
  import logging
3
  from dataclasses import dataclass
4
  from os import listdir, path
5
- from typing import Dict, List, Optional
6
 
7
  import datasets
8
  from datasets import BuilderConfig, DatasetInfo, Features, Sequence, SplitGenerator, Value
@@ -18,8 +18,9 @@ class BratConfig(BuilderConfig):
18
  description: Optional[str] = None
19
  citation: Optional[str] = None
20
  homepage: Optional[str] = None
21
-
22
- subdirectory_mapping: Optional[Dict[str, str]] = None
 
23
  file_name_blacklist: Optional[List[str]] = None
24
  ann_file_extension: str = "ann"
25
  txt_file_extension: str = "txt"
@@ -270,14 +271,15 @@ class Brat(datasets.GeneratorBasedBuilder):
270
  )
271
  return res
272
 
273
- def _generate_examples(self, files=None, directory=None):
274
  """Read context (.txt) and annotation (.ann) files."""
275
  if files is None:
276
- assert (
277
- directory is not None
278
- ), "If files is None, directory has to be provided, but it is also None."
279
  _files = glob.glob(f"{directory}/*.{self.config.ann_file_extension}")
280
  files = sorted(path.splitext(fn)[0] for fn in _files)
 
 
281
 
282
  for filename in files:
283
  basename = path.basename(filename)
@@ -311,24 +313,36 @@ class Brat(datasets.GeneratorBasedBuilder):
311
 
312
  data_dir = dl_manager.download_and_extract(self.config.url)
313
 
314
- subdirectory_mapping = self.config.subdirectory_mapping
315
  # if no subdirectory mapping is provided, ...
316
- if subdirectory_mapping is None:
317
  # ... use available subdirectories as split names ...
318
  subdirs = [f for f in listdir(data_dir) if path.isdir(path.join(data_dir, f))]
319
  if len(subdirs) > 0:
320
- subdirectory_mapping = {subdir: subdir for subdir in subdirs}
321
  else:
322
  # ... otherwise, default to a single train split with the base directory
323
- subdirectory_mapping = {"": "train"}
 
 
 
 
 
 
 
 
 
 
 
 
324
 
325
  return [
326
  SplitGenerator(
327
  name=split,
328
  # These kwargs will be passed to _generate_examples
329
  gen_kwargs={
330
- "directory": path.join(data_dir, subdir),
 
331
  },
332
  )
333
- for subdir, split in subdirectory_mapping.items()
334
- ]
 
2
  import logging
3
  from dataclasses import dataclass
4
  from os import listdir, path
5
+ from typing import Dict, List, Optional, Union
6
 
7
  import datasets
8
  from datasets import BuilderConfig, DatasetInfo, Features, Sequence, SplitGenerator, Value
 
18
  description: Optional[str] = None
19
  citation: Optional[str] = None
20
  homepage: Optional[str] = None
21
+
22
+ # paths to directories or files per split (relative to url or data_dir)
23
+ split_paths: Optional[Dict[str, Union[str, List[str]]]] = None
24
  file_name_blacklist: Optional[List[str]] = None
25
  ann_file_extension: str = "ann"
26
  txt_file_extension: str = "txt"
 
271
  )
272
  return res
273
 
274
+ def _generate_examples(self, base_dir: str, files: Optional[List[str]] = None, directory: Optional[str] = None):
275
  """Read context (.txt) and annotation (.ann) files."""
276
  if files is None:
277
+ if directory is None:
278
+ raise ValueError("Either files or directory has to be provided.")
 
279
  _files = glob.glob(f"{directory}/*.{self.config.ann_file_extension}")
280
  files = sorted(path.splitext(fn)[0] for fn in _files)
281
+ elif directory is not None:
282
+ raise ValueError("Only one of files or directory can be provided.")
283
 
284
  for filename in files:
285
  basename = path.basename(filename)
 
313
 
314
  data_dir = dl_manager.download_and_extract(self.config.url)
315
 
 
316
  # if no subdirectory mapping is provided, ...
317
+ if self.config.split_paths is None:
318
  # ... use available subdirectories as split names ...
319
  subdirs = [f for f in listdir(data_dir) if path.isdir(path.join(data_dir, f))]
320
  if len(subdirs) > 0:
321
+ split_paths = {subdir: {"directory": subdir} for subdir in subdirs}
322
  else:
323
  # ... otherwise, default to a single train split with the base directory
324
+ split_paths = {"train": {"directory": ""}}
325
+ else:
326
+ split_paths = {}
327
+ for split, paths in self.config.split_paths.items():
328
+ if isinstance(paths, str):
329
+ split_paths[split] = {"directory": paths}
330
+ elif isinstance(paths, list):
331
+ split_paths[split] = {"files": paths}
332
+ else:
333
+ raise ValueError(
334
+ f"split_paths must be a dict containing either a single path to a directory "
335
+ f"or a list of file paths, but found: {paths}"
336
+ )
337
 
338
  return [
339
  SplitGenerator(
340
  name=split,
341
  # These kwargs will be passed to _generate_examples
342
  gen_kwargs={
343
+ "base_dir": data_dir,
344
+ **split_kwargs,
345
  },
346
  )
347
+ for split, split_kwargs in split_paths.items()
348
+ ]