Datasets:

Languages:
English
ArXiv:
License:
aspartate commited on
Commit
953dbaf
·
1 Parent(s): c9117af

exclude PNGs from download

Browse files
Files changed (1) hide show
  1. Patho-Bench.py +48 -18
Patho-Bench.py CHANGED
@@ -30,12 +30,19 @@ class PathoBenchConfig(datasets.BuilderConfig):
30
 
31
 
32
  class PathoBenchDataset(datasets.GeneratorBasedBuilder):
 
 
 
 
33
  BUILDER_CONFIGS = [
34
  PathoBenchConfig(name="custom_config", version="1.0.0", description="PathoBench config")
35
  ]
36
  BUILDER_CONFIG_CLASS = PathoBenchConfig
37
 
38
-
 
 
 
39
  def _info(self):
40
  return datasets.DatasetInfo(
41
  description="PathoBench: collection of canonical computational pathology tasks",
@@ -56,36 +63,60 @@ class PathoBenchDataset(datasets.GeneratorBasedBuilder):
56
  # Ensure the base local directory exists
57
  os.makedirs(local_dir, exist_ok=True)
58
 
59
- # download available_splits.yaml if not yet downloaded
60
  snapshot_download(
61
  repo_id=repo_id,
62
- allow_patterns=["available_splits.yaml"],
63
  repo_type="dataset",
64
  local_dir=local_dir,
65
  force_download=force_download,
66
  )
67
 
68
- # open yaml and get a list of datasets implemented
69
  with open(os.path.join(local_dir, "available_splits.yaml"), 'r') as file:
70
  available_splits = yaml.safe_load(file)
71
 
72
- # ensure dataset_to_download is in implemented_datasets
73
  if dataset_to_download != "*":
74
- assert dataset_to_download in available_splits, f"{dataset_to_download} was not found. Available splits: ({available_splits})"
 
 
 
75
  if task_in_dataset != "*":
76
- assert task_in_dataset in available_splits[dataset_to_download], f"{task_in_dataset} was not found in {dataset_to_download}. Available tasks: ({available_splits[dataset_to_download]})"
 
 
 
77
 
78
- # Determine parent folder based on dataset naming
79
- os.makedirs(local_dir, exist_ok=True)
80
-
81
- # Determine the download pattern
82
  if dataset_to_download == "*":
83
- allow_patterns = [f"*/*"]
 
 
 
 
 
 
84
  else:
85
- task_path = "*" if task_in_dataset == '*' else f"{task_in_dataset}/*"
86
- allow_patterns = [f"{dataset_to_download}/{task_path}"]
87
-
88
- # Download the required datasets
 
 
 
 
 
 
 
 
 
 
 
 
89
  snapshot_download(
90
  repo_id=repo_id,
91
  allow_patterns=allow_patterns,
@@ -94,7 +125,7 @@ class PathoBenchDataset(datasets.GeneratorBasedBuilder):
94
  force_download=force_download,
95
  )
96
 
97
- # Locate all .tsv files
98
  search_pattern = os.path.join(local_dir, '**', '*.tsv')
99
  all_tsv_splits = glob.glob(search_pattern, recursive=True)
100
 
@@ -105,7 +136,6 @@ class PathoBenchDataset(datasets.GeneratorBasedBuilder):
105
  )
106
  ]
107
 
108
-
109
  def _generate_examples(self, filepath):
110
  idx = 0
111
  for file in filepath:
 
30
 
31
 
32
  class PathoBenchDataset(datasets.GeneratorBasedBuilder):
33
+ """
34
+ Downloads only the .tsv and .yaml files needed to construct the dataset.
35
+ Excludes .png images so they don't break the builder.
36
+ """
37
  BUILDER_CONFIGS = [
38
  PathoBenchConfig(name="custom_config", version="1.0.0", description="PathoBench config")
39
  ]
40
  BUILDER_CONFIG_CLASS = PathoBenchConfig
41
 
42
+ @property
43
+ def repo_id(self):
44
+ return "MahmoodLab/patho-bench"
45
+
46
  def _info(self):
47
  return datasets.DatasetInfo(
48
  description="PathoBench: collection of canonical computational pathology tasks",
 
63
  # Ensure the base local directory exists
64
  os.makedirs(local_dir, exist_ok=True)
65
 
66
+ # 1) Download the top-level available_splits.yaml
67
  snapshot_download(
68
  repo_id=repo_id,
69
+ allow_patterns=["available_splits.yaml"], # only this file
70
  repo_type="dataset",
71
  local_dir=local_dir,
72
  force_download=force_download,
73
  )
74
 
75
+ # Read available splits
76
  with open(os.path.join(local_dir, "available_splits.yaml"), 'r') as file:
77
  available_splits = yaml.safe_load(file)
78
 
79
+ # Basic validation
80
  if dataset_to_download != "*":
81
+ assert dataset_to_download in available_splits, (
82
+ f"{dataset_to_download} was not found. "
83
+ f"Available splits: {list(available_splits.keys())}"
84
+ )
85
  if task_in_dataset != "*":
86
+ assert task_in_dataset in available_splits[dataset_to_download], (
87
+ f"{task_in_dataset} was not found in {dataset_to_download}. "
88
+ f"Available tasks: {available_splits[dataset_to_download]}"
89
+ )
90
 
91
+ # 2) Decide what to allow based on dataset/task
92
+ #
93
+ # We only want .tsv and the relevant .yaml files (like about.yaml, config.yaml).
94
+ # That way, we skip .png images which can cause issues or be large in LFS.
95
  if dataset_to_download == "*":
96
+ # Download every dataset subfolder's .tsv and about.yaml/config.yaml
97
+ allow_patterns = [
98
+ "**/*.tsv", # All tsv splits
99
+ "**/about.yaml", # The about files
100
+ "**/config.yaml", # The config files
101
+ "available_splits.yaml" # Already downloaded, but no harm
102
+ ]
103
  else:
104
+ if task_in_dataset == "*":
105
+ allow_patterns = [
106
+ f"{dataset_to_download}/**/*.tsv",
107
+ f"{dataset_to_download}/**/about.yaml",
108
+ f"{dataset_to_download}/**/config.yaml",
109
+ "available_splits.yaml"
110
+ ]
111
+ else:
112
+ allow_patterns = [
113
+ f"{dataset_to_download}/{task_in_dataset}/*.tsv",
114
+ f"{dataset_to_download}/{task_in_dataset}/config.yaml",
115
+ f"{dataset_to_download}/about.yaml",
116
+ "available_splits.yaml"
117
+ ]
118
+
119
+ # 3) Download the requested patterns
120
  snapshot_download(
121
  repo_id=repo_id,
122
  allow_patterns=allow_patterns,
 
125
  force_download=force_download,
126
  )
127
 
128
+ # 4) Locate all .tsv files to pass to _generate_examples
129
  search_pattern = os.path.join(local_dir, '**', '*.tsv')
130
  all_tsv_splits = glob.glob(search_pattern, recursive=True)
131
 
 
136
  )
137
  ]
138
 
 
139
  def _generate_examples(self, filepath):
140
  idx = 0
141
  for file in filepath: