BeeGass commited on
Commit
70dc2c2
·
verified ·
1 Parent(s): 414d2a4

Update loading script to v5.0.0 with all 10 groups and dynamic filtering

Browse files
Files changed (1) hide show
  1. permutation-groups.py +96 -207
permutation-groups.py CHANGED
@@ -1,41 +1,12 @@
1
  import datasets
2
  import json
3
  import os
4
- import pyarrow as pa
5
 
6
  _DESCRIPTION = "Permutation composition datasets with dynamic filtering by group degree, order, and sequence length."
7
  _HOMEPAGE = "https://huggingface.co/datasets/BeeGass/permutation-groups"
8
  _LICENSE = "MIT"
9
 
10
- # TEMPORARY: Define the actual file structure explicitly
11
- # TODO: Revert to wildcard patterns once datasets library supports them properly
12
- _DATA_FILES = {
13
- "symmetric_superset": {
14
- "train": ["data-00000-of-00003.arrow", "data-00001-of-00003.arrow", "data-00002-of-00003.arrow"],
15
- "test": ["data-00000-of-00001.arrow"]
16
- },
17
- "alternating_superset": {
18
- "train": ["data-00000-of-00003.arrow", "data-00001-of-00003.arrow", "data-00002-of-00003.arrow"],
19
- "test": ["data-00000-of-00001.arrow"]
20
- },
21
- "cyclic_superset": {
22
- "train": ["data-00000-of-00001.arrow"],
23
- "test": ["data-00000-of-00001.arrow"]
24
- },
25
- "dihedral_superset": {
26
- "train": ["data-00000-of-00001.arrow"],
27
- "test": ["data-00000-of-00001.arrow"]
28
- },
29
- "psl25_data": {
30
- "train": ["data-00000-of-00001.arrow"],
31
- "test": ["data-00000-of-00001.arrow"]
32
- },
33
- "f20_data": {
34
- "train": ["data-00000-of-00001.arrow"],
35
- "test": ["data-00000-of-00001.arrow"]
36
- }
37
- }
38
-
39
  class PermutationGroupsConfig(datasets.BuilderConfig):
40
  def __init__(
41
  self,
@@ -52,7 +23,8 @@ class PermutationGroupsConfig(datasets.BuilderConfig):
52
  Configuration for loading permutation groups.
53
 
54
  Args:
55
- group_type: Type of group (symmetric, alternating, cyclic, dihedral, psl25, f20)
 
56
  min_degree: Minimum group degree to include
57
  max_degree: Maximum group degree to include
58
  min_order: Minimum group order to include
@@ -79,10 +51,14 @@ class PermutationGroupsConfig(datasets.BuilderConfig):
79
  class PermutationGroups(datasets.GeneratorBasedBuilder):
80
  """Permutation groups dataset with dynamic filtering."""
81
 
82
- VERSION = datasets.Version("4.0.0")
83
 
84
- # Define available group types
85
- GROUP_TYPES = ["symmetric", "alternating", "cyclic", "dihedral", "psl25", "f20"]
 
 
 
 
86
 
87
  BUILDER_CONFIGS = []
88
 
@@ -105,66 +81,6 @@ class PermutationGroups(datasets.GeneratorBasedBuilder):
105
  )
106
  )
107
 
108
- # Keep backwards compatibility configs
109
- LEGACY_GROUPS = {
110
- "s3": ("symmetric", 3, 3), "s4": ("symmetric", 4, 4), "s5": ("symmetric", 5, 5),
111
- "s6": ("symmetric", 6, 6), "s7": ("symmetric", 7, 7), "s8": ("symmetric", 8, 8),
112
- "s9": ("symmetric", 9, 9), "s10": ("symmetric", 10, 10),
113
- "a3": ("alternating", 3, 3), "a4": ("alternating", 4, 4), "a5": ("alternating", 5, 5),
114
- "a6": ("alternating", 6, 6), "a7": ("alternating", 7, 7), "a8": ("alternating", 8, 8),
115
- "a9": ("alternating", 9, 9), "a10": ("alternating", 10, 10),
116
- "c3": ("cyclic", 3, 3), "c4": ("cyclic", 4, 4), "c5": ("cyclic", 5, 5),
117
- "c6": ("cyclic", 6, 6), "c7": ("cyclic", 7, 7), "c8": ("cyclic", 8, 8),
118
- "c9": ("cyclic", 9, 9), "c10": ("cyclic", 10, 10), "c12": ("cyclic", 12, 12),
119
- "c15": ("cyclic", 15, 15), "c20": ("cyclic", 20, 20), "c25": ("cyclic", 25, 25),
120
- "c30": ("cyclic", 30, 30),
121
- "z3": ("cyclic", 3, 3), "z4": ("cyclic", 4, 4), "z5": ("cyclic", 5, 5), "z6": ("cyclic", 6, 6),
122
- "d3": ("dihedral", 3, 3), "d4": ("dihedral", 4, 4), "d5": ("dihedral", 5, 5),
123
- "d6": ("dihedral", 6, 6), "d7": ("dihedral", 7, 7), "d8": ("dihedral", 8, 8),
124
- "d9": ("dihedral", 9, 9), "d10": ("dihedral", 10, 10), "d12": ("dihedral", 12, 12),
125
- "d15": ("dihedral", 15, 15), "d20": ("dihedral", 20, 20),
126
- }
127
-
128
- for name, (group_type, min_deg, max_deg) in LEGACY_GROUPS.items():
129
- # Simple name (e.g., "s5")
130
- BUILDER_CONFIGS.append(
131
- PermutationGroupsConfig(
132
- name=name,
133
- description=f"Legacy config for {name.upper()}",
134
- group_type=group_type,
135
- min_degree=min_deg,
136
- max_degree=max_deg,
137
- )
138
- )
139
- # Old style name (e.g., "s5_data")
140
- BUILDER_CONFIGS.append(
141
- PermutationGroupsConfig(
142
- name=f"{name}_data",
143
- description=f"Legacy config for {name.upper()}",
144
- group_type=group_type,
145
- min_degree=min_deg,
146
- max_degree=max_deg,
147
- )
148
- )
149
-
150
- # Add legacy configs for special groups
151
- BUILDER_CONFIGS.extend([
152
- PermutationGroupsConfig(
153
- name="psl25_data",
154
- description="Legacy config for PSL(2,5)",
155
- group_type="psl25",
156
- min_degree=6,
157
- max_degree=6,
158
- ),
159
- PermutationGroupsConfig(
160
- name="f20_data",
161
- description="Legacy config for F20",
162
- group_type="f20",
163
- min_degree=5,
164
- max_degree=5,
165
- ),
166
- ])
167
-
168
  DEFAULT_CONFIG_NAME = "symmetric"
169
 
170
  def _info(self):
@@ -185,137 +101,110 @@ class PermutationGroups(datasets.GeneratorBasedBuilder):
185
  def _split_generators(self, dl_manager):
186
  # Determine which datasets to load
187
  if self.config.group_type:
188
- if self.config.group_type in ["psl25", "f20"]:
189
- # PSL25 and F20 are stored as individual datasets
190
- datasets_to_load = [f"{self.config.group_type}_data"]
191
- else:
192
- # Load the superset for this group type
193
- datasets_to_load = [f"{self.config.group_type}_superset"]
194
  else:
195
  # Load all supersets
196
- datasets_to_load = ["symmetric_superset", "alternating_superset",
197
- "cyclic_superset", "dihedral_superset",
198
- "psl25_data", "f20_data"]
 
 
 
 
199
 
200
- # TEMPORARY: Build explicit file URLs instead of using wildcards
201
- # TODO: Revert to wildcard pattern once supported:
202
- # data_urls = {
203
- # "train": f"data/{dataset_name}/train/data-*-of-*.arrow",
204
- # "test": f"data/{dataset_name}/test/data-*-of-*.arrow",
205
- # }
206
  train_urls = []
207
  test_urls = []
208
 
209
  for dataset_name in datasets_to_load:
210
- if dataset_name in _DATA_FILES:
211
- # Build full URLs for each file
212
- for filename in _DATA_FILES[dataset_name]["train"]:
213
- train_urls.append(f"data/{dataset_name}/train/{filename}")
214
- for filename in _DATA_FILES[dataset_name]["test"]:
215
- test_urls.append(f"data/{dataset_name}/test/{filename}")
216
 
217
  # Download files
218
- if train_urls and test_urls:
219
- downloaded_files = dl_manager.download({
220
- "train": train_urls,
221
- "test": test_urls
222
- })
223
-
224
- return [
225
- datasets.SplitGenerator(
226
- name=datasets.Split.TRAIN,
227
- gen_kwargs={
228
- "files": downloaded_files["train"],
229
- },
230
- ),
231
- datasets.SplitGenerator(
232
- name=datasets.Split.TEST,
233
- gen_kwargs={
234
- "files": downloaded_files["test"],
235
- },
236
- ),
237
- ]
238
- else:
239
- # Return empty splits if no files found
240
- return [
241
- datasets.SplitGenerator(
242
- name=datasets.Split.TRAIN,
243
- gen_kwargs={
244
- "files": [],
245
- },
246
- ),
247
- datasets.SplitGenerator(
248
- name=datasets.Split.TEST,
249
- gen_kwargs={
250
- "files": [],
251
- },
252
- ),
253
- ]
254
 
255
  def _generate_examples(self, files):
256
  """Yield examples with filtering."""
257
  idx = 0
258
- total_examined = 0
259
- total_filtered_out = 0
260
 
261
  for file_path in files:
262
- # Load the dataset file using datasets library
263
- dataset = datasets.Dataset.from_file(file_path)
264
 
265
- # Convert to pandas for easier row-wise iteration
266
- df = dataset.to_pandas()
267
 
268
- # Iterate through rows and apply filters
269
- for _, row in df.iterrows():
270
- total_examined += 1
271
-
272
- # Apply filters
273
- # Filter by group type (if specified)
274
- if self.config.group_type and row.get("group_type") != self.config.group_type:
275
- total_filtered_out += 1
276
- continue
277
-
278
- # Filter by degree
279
- if self.config.min_degree and row.get("group_degree", 0) < self.config.min_degree:
280
- total_filtered_out += 1
281
- continue
282
- if self.config.max_degree and row.get("group_degree", float('inf')) > self.config.max_degree:
283
- total_filtered_out += 1
284
- continue
285
-
286
- # Filter by order
287
- if self.config.min_order and row.get("group_order", 0) < self.config.min_order:
288
- total_filtered_out += 1
289
- continue
290
- if self.config.max_order and row.get("group_order", float('inf')) > self.config.max_order:
291
- total_filtered_out += 1
292
- continue
293
-
294
- # Filter by sequence length
295
- seq_len = row.get("sequence_length", len(row["input_sequence"].split()))
296
- if seq_len < self.config.min_len or seq_len > self.config.max_len:
297
- total_filtered_out += 1
298
- continue
299
-
300
- # Yield the example
301
  yield idx, {
302
  "input_sequence": row["input_sequence"],
303
  "target": row["target"],
304
- "group_type": row.get("group_type", ""),
305
- "group_degree": int(row.get("group_degree", 0)),
306
- "group_order": int(row.get("group_order", 0)),
307
- "sequence_length": int(row.get("sequence_length", seq_len)),
308
  }
309
- idx += 1
310
-
311
- # Log a warning if all examples were filtered out
312
- if idx == 0 and total_examined > 0:
313
- import warnings
314
- warnings.warn(
315
- f"All {total_examined} examples were filtered out with the current configuration:\n"
316
- f" group_type={self.config.group_type}\n"
317
- f" degree_range=[{self.config.min_degree}, {self.config.max_degree}]\n"
318
- f" order_range=[{self.config.min_order}, {self.config.max_order}]\n"
319
- f" length_range=[{self.config.min_len}, {self.config.max_len}]\n"
320
- f"This might be expected if the requested configuration doesn't exist in the dataset."
321
- )
 
1
  import datasets
2
  import json
3
  import os
4
+ import pandas as pd
5
 
6
  _DESCRIPTION = "Permutation composition datasets with dynamic filtering by group degree, order, and sequence length."
7
  _HOMEPAGE = "https://huggingface.co/datasets/BeeGass/permutation-groups"
8
  _LICENSE = "MIT"
9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
10
  class PermutationGroupsConfig(datasets.BuilderConfig):
11
  def __init__(
12
  self,
 
23
  Configuration for loading permutation groups.
24
 
25
  Args:
26
+ group_type: Type of group (symmetric, alternating, cyclic, dihedral, klein,
27
+ quaternion, elementary_abelian, psl, frobenius, mathieu)
28
  min_degree: Minimum group degree to include
29
  max_degree: Maximum group degree to include
30
  min_order: Minimum group order to include
 
51
  class PermutationGroups(datasets.GeneratorBasedBuilder):
52
  """Permutation groups dataset with dynamic filtering."""
53
 
54
+ VERSION = datasets.Version("5.0.0")
55
 
56
+ # Define all available group types
57
+ GROUP_TYPES = [
58
+ "symmetric", "alternating", "cyclic", "dihedral",
59
+ "klein", "quaternion", "elementary_abelian", "psl",
60
+ "frobenius", "mathieu"
61
+ ]
62
 
63
  BUILDER_CONFIGS = []
64
 
 
81
  )
82
  )
83
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
84
  DEFAULT_CONFIG_NAME = "symmetric"
85
 
86
  def _info(self):
 
101
  def _split_generators(self, dl_manager):
102
  # Determine which datasets to load
103
  if self.config.group_type:
104
+ # Load the superset for this group type
105
+ datasets_to_load = [f"{self.config.group_type}_superset"]
 
 
 
 
106
  else:
107
  # Load all supersets
108
+ datasets_to_load = [
109
+ "symmetric_superset", "alternating_superset",
110
+ "cyclic_superset", "dihedral_superset",
111
+ "klein_superset", "quaternion_superset",
112
+ "elementary_abelian_superset", "psl_superset",
113
+ "frobenius_superset", "mathieu_superset"
114
+ ]
115
 
116
+ # Build file URLs using wildcards
 
 
 
 
 
117
  train_urls = []
118
  test_urls = []
119
 
120
  for dataset_name in datasets_to_load:
121
+ train_urls.append(f"data/{dataset_name}/train/data-*.arrow")
122
+ test_urls.append(f"data/{dataset_name}/test/data-*.arrow")
 
 
 
 
123
 
124
  # Download files
125
+ downloaded_files = dl_manager.download({
126
+ "train": train_urls,
127
+ "test": test_urls
128
+ })
129
+
130
+ # Flatten the lists of files
131
+ train_files = []
132
+ test_files = []
133
+
134
+ for file_list in downloaded_files["train"]:
135
+ if isinstance(file_list, list):
136
+ train_files.extend(file_list)
137
+ else:
138
+ train_files.append(file_list)
139
+
140
+ for file_list in downloaded_files["test"]:
141
+ if isinstance(file_list, list):
142
+ test_files.extend(file_list)
143
+ else:
144
+ test_files.append(file_list)
145
+
146
+ return [
147
+ datasets.SplitGenerator(
148
+ name=datasets.Split.TRAIN,
149
+ gen_kwargs={
150
+ "files": train_files,
151
+ },
152
+ ),
153
+ datasets.SplitGenerator(
154
+ name=datasets.Split.TEST,
155
+ gen_kwargs={
156
+ "files": test_files,
157
+ },
158
+ ),
159
+ ]
 
160
 
161
  def _generate_examples(self, files):
162
  """Yield examples with filtering."""
163
  idx = 0
 
 
164
 
165
  for file_path in files:
166
+ # Load the Arrow file
167
+ table = datasets.table.read_table(file_path)
168
 
169
+ # Convert to pandas for easier filtering
170
+ df = table.to_pandas()
171
 
172
+ # Apply filters
173
+ mask = pd.Series([True] * len(df))
174
+
175
+ # Filter by group type (if specified in config)
176
+ if self.config.group_type:
177
+ mask &= (df["group_type"] == self.config.group_type)
178
+
179
+ # Filter by degree
180
+ if self.config.min_degree is not None:
181
+ mask &= (df["group_degree"] >= self.config.min_degree)
182
+ if self.config.max_degree is not None:
183
+ mask &= (df["group_degree"] <= self.config.max_degree)
184
+
185
+ # Filter by order
186
+ if self.config.min_order is not None:
187
+ mask &= (df["group_order"] >= self.config.min_order)
188
+ if self.config.max_order is not None:
189
+ mask &= (df["group_order"] <= self.config.max_order)
190
+
191
+ # Filter by sequence length
192
+ if self.config.min_len is not None:
193
+ mask &= (df["sequence_length"] >= self.config.min_len)
194
+ if self.config.max_len is not None:
195
+ mask &= (df["sequence_length"] <= self.config.max_len)
196
+
197
+ # Apply mask
198
+ filtered_df = df[mask]
199
+
200
+ # Yield filtered examples
201
+ for _, row in filtered_df.iterrows():
 
 
 
202
  yield idx, {
203
  "input_sequence": row["input_sequence"],
204
  "target": row["target"],
205
+ "group_type": row["group_type"],
206
+ "group_degree": int(row["group_degree"]),
207
+ "group_order": int(row["group_order"]),
208
+ "sequence_length": int(row["sequence_length"]),
209
  }
210
+ idx += 1