Berom0227 commited on
Commit
7f071ee
·
verified ·
1 Parent(s): efd59c0

Upload scripts/sample_atomic_commites.py with huggingface_hub

Browse files
Files changed (1) hide show
  1. scripts/sample_atomic_commites.py +70 -54
scripts/sample_atomic_commites.py CHANGED
@@ -1,13 +1,16 @@
1
  #!/usr/bin/env python3
2
  """
3
  Sample atomic commits from CCS dataset for concern extraction.
4
- Implements atomic sampling with token limits and SHA deduplication.
5
  """
6
 
7
  import pandas as pd
8
 
9
  import tiktoken
10
- from typing import Dict, List, Set
 
 
 
11
 
12
  # Processing configuration
13
  CONVENTIONAL_COMMIT_TYPES: List[str] = ["feat", "fix", "refactor", "test", "docs", "build", "cicd"]
@@ -38,7 +41,7 @@ DIFF_OUTPUT_DIR: str = "data/types"
38
 
39
 
40
  def normalize_dataset(df: pd.DataFrame) -> pd.DataFrame:
41
- """Normalize CI labels to CICD for consistent categorization."""
42
  df[COLUMN_ANNOTATED_TYPE] = (
43
  df[COLUMN_ANNOTATED_TYPE]
44
  .str.lower()
@@ -49,8 +52,8 @@ def normalize_dataset(df: pd.DataFrame) -> pd.DataFrame:
49
  return df
50
 
51
 
52
- def apply_token_filtering(df: pd.DataFrame) -> pd.DataFrame:
53
- """Filter commits exceeding token limit to prevent context overflow."""
54
  encoding = tiktoken.get_encoding(ENCODING_MODEL)
55
 
56
  combined_text = (
@@ -70,8 +73,8 @@ def apply_token_filtering(df: pd.DataFrame) -> pd.DataFrame:
70
  return filtered_df
71
 
72
 
73
- def apply_sha_deduplication(df: pd.DataFrame, excluded_shas: Set[str]) -> pd.DataFrame:
74
- """Remove previously sampled commits to avoid training data contamination."""
75
  original_count = len(df)
76
 
77
  sha_mask = ~df[COLUMN_SHA].astype(str).isin(excluded_shas)
@@ -82,23 +85,25 @@ def apply_sha_deduplication(df: pd.DataFrame, excluded_shas: Set[str]) -> pd.Dat
82
  return filtered_df
83
 
84
 
85
- def load_shas(file_path: str) -> Set[str]:
86
- """Load existing SHAs from sampled dataset to exclude duplicates."""
87
  try:
88
  df = pd.read_csv(file_path)
89
  sha_set = set(df[COLUMN_SHA].astype(str))
 
90
  print(f"Loaded {len(sha_set)} SHAs for deduplication")
91
- return sha_set
 
92
  except FileNotFoundError:
93
  print(f"No existing samples found at {file_path}")
94
- return set()
95
  except Exception as e:
96
- print(f"Error loading existing SHAs: {e}")
97
- return set()
98
 
99
 
100
  def load_ccs_dataset(file_path: str) -> pd.DataFrame:
101
- """Load and validate CCS dataset structure."""
102
  try:
103
  df = pd.read_csv(file_path)
104
  if df.empty:
@@ -118,7 +123,7 @@ def load_ccs_dataset(file_path: str) -> pd.DataFrame:
118
  def save_to_csv(
119
  data: List[Dict[str, str]], output_path: str, columns: List[str]
120
  ) -> None:
121
- """Append new samples to existing dataset or create new file."""
122
  import os
123
 
124
  os.makedirs(os.path.dirname(output_path), exist_ok=True)
@@ -140,7 +145,7 @@ def save_to_csv(
140
  def group_commits_by_type(
141
  df: pd.DataFrame, valid_types: List[str]
142
  ) -> Dict[str, pd.DataFrame]:
143
- """Group commits by concern type for balanced sampling."""
144
  type_mask = df[COLUMN_ANNOTATED_TYPE].isin(valid_types)
145
  valid_df = df[type_mask].copy()
146
 
@@ -158,13 +163,13 @@ def group_commits_by_type(
158
  def sample_commits_for_type(
159
  df: pd.DataFrame, count: int, output_columns: List[str]
160
  ) -> List[Dict[str, str]]:
161
- """Sample fixed number of commits per type for balanced dataset."""
162
- sampled_df = df.sample(n=count, random_state=None)
163
  return sampled_df[output_columns].to_dict("records")
164
 
165
 
166
  def extract_diffs(sampled_data: List[Dict[str, str]], output_dir: str) -> None:
167
- """Extract git diff files organized by type into separate directories."""
168
  import os
169
 
170
  type_counts = {}
@@ -196,13 +201,13 @@ def extract_diffs(sampled_data: List[Dict[str, str]], output_dir: str) -> None:
196
  record[COLUMN_GIT_DIFF],
197
  ]
198
 
199
- with open(filepath, "w") as f:
200
  f.write("\n".join(content_lines))
201
 
202
  print(f"Extracted {len(sampled_data)} diff files to {output_dir}")
203
 
204
  def remove_excluded_commits(df: pd.DataFrame, excluded_shas: Set[str]) -> pd.DataFrame:
205
- """Remove manually excluded commits based on quality issues."""
206
  before_count = len(df)
207
  print(f"Initial commit count: {before_count}")
208
 
@@ -217,69 +222,80 @@ def remove_excluded_commits(df: pd.DataFrame, excluded_shas: Set[str]) -> pd.Dat
217
 
218
  def main() -> None:
219
  """
220
- Main function implementing atomic sampling strategy:
221
- 1. Load dataset and backup SHAs
222
- 2. Apply CI->CICD normalization
223
- 3. Apply token-based filtering
224
- 4. Apply SHA deduplication
225
- 5. Group by type and randomly sample
226
- 6. Save results and extract diffs
 
227
  """
228
  print("Starting atomic sampling strategy for CCS dataset")
229
  print("=" * 50)
230
 
231
- # Step 1: Load dataset and backup SHAs
232
- print("Step 1: Loading dataset and backup SHAs")
233
- existing_shas = load_shas(SAMPLED_CSV_PATH)
234
- excluded_shas = load_shas(EXCLUDED_COMMITS_PATH)
235
  ccs_df = load_ccs_dataset(CCS_SOURCE_PATH)
236
 
237
  # Step 2: Remove excluded commits
238
  print("\nStep 2: Removing excluded commits")
239
  ccs_df = remove_excluded_commits(ccs_df, excluded_shas)
240
 
241
- # Step 3: Apply CI->CICD normalization
242
- print("\nStep 3: Applying CI->CICD normalization")
 
 
 
 
243
  ccs_df = normalize_dataset(ccs_df)
244
 
245
- # Step 4: Apply token-based filtering
246
- print("\nStep 4: Applying token-based filtering")
247
- ccs_df = apply_token_filtering(ccs_df)
248
 
249
- # Step 5: Apply SHA deduplication
250
- print("\nStep 5: Applying SHA deduplication")
251
- ccs_df = apply_sha_deduplication(ccs_df, existing_shas)
252
 
253
- # Step 6: Group by type and randomly sample
254
  print("\nStep 6: Grouping by type and random sampling")
255
  commits_by_type = group_commits_by_type(ccs_df, CONVENTIONAL_COMMIT_TYPES)
256
 
257
  all_sampled_data = []
258
- for commits_df in commits_by_type.values():
 
 
 
 
 
 
 
 
 
 
 
 
 
259
  sampled_data = sample_commits_for_type(
260
- commits_df, SAMPLES_PER_TYPE, OUTPUT_COLUMNS
261
  )
262
  all_sampled_data.extend(sampled_data)
 
263
 
264
  print(f"Random sampling: generated {len(all_sampled_data)} samples total")
265
 
266
  # Step 7: Save results and extract diffs
267
  print("\nStep 7: Saving results and extracting diffs")
268
- save_to_csv(all_sampled_data, SAMPLED_CSV_PATH, OUTPUT_COLUMNS)
269
- extract_diffs(all_sampled_data, DIFF_OUTPUT_DIR)
 
 
 
270
 
271
  # Final summary
272
  print("\n" + "=" * 50)
273
  print("Atomic sampling completed successfully!")
274
-
275
- type_counts = {}
276
- for record in all_sampled_data:
277
- commit_type = record.get(COLUMN_ANNOTATED_TYPE, "")
278
- type_counts[commit_type] = type_counts.get(commit_type, 0) + 1
279
-
280
- print("Final sample distribution:")
281
- for commit_type in sorted(type_counts.keys()):
282
- print(f" {commit_type}: {type_counts[commit_type]} samples")
283
 
284
 
285
  if __name__ == "__main__":
 
1
  #!/usr/bin/env python3
2
  """
3
  Sample atomic commits from CCS dataset for concern extraction.
4
+ Implements balanced sampling pipeline with filtering, normalization, and deduplication.
5
  """
6
 
7
  import pandas as pd
8
 
9
  import tiktoken
10
+ from typing import Dict, List, Set, Tuple
11
+
12
+ # Random seed for reproducibility
13
+ RANDOM_SEED: int = 42
14
 
15
  # Processing configuration
16
  CONVENTIONAL_COMMIT_TYPES: List[str] = ["feat", "fix", "refactor", "test", "docs", "build", "cicd"]
 
41
 
42
 
43
  def normalize_dataset(df: pd.DataFrame) -> pd.DataFrame:
44
+ """Normalize CI commit type labels to CICD for consistent categorization."""
45
  df[COLUMN_ANNOTATED_TYPE] = (
46
  df[COLUMN_ANNOTATED_TYPE]
47
  .str.lower()
 
52
  return df
53
 
54
 
55
+ def remove_long_token_commits(df: pd.DataFrame) -> pd.DataFrame:
56
+ """Filter out commits exceeding TARGET_TOKEN_LIMIT to prevent model context overflow."""
57
  encoding = tiktoken.get_encoding(ENCODING_MODEL)
58
 
59
  combined_text = (
 
73
  return filtered_df
74
 
75
 
76
+ def remove_existing_commits(df: pd.DataFrame, excluded_shas: Set[str]) -> pd.DataFrame:
77
+ """Remove commits with SHAs that already exist in the sampled dataset."""
78
  original_count = len(df)
79
 
80
  sha_mask = ~df[COLUMN_SHA].astype(str).isin(excluded_shas)
 
85
  return filtered_df
86
 
87
 
88
+ def load_shas_and_type_counts(file_path: str) -> Tuple[Set[str], Dict[str, int]]:
89
+ """Load commit SHAs and type counts from CSV file for deduplication and intelligent sampling."""
90
  try:
91
  df = pd.read_csv(file_path)
92
  sha_set = set(df[COLUMN_SHA].astype(str))
93
+ type_counts = df[COLUMN_ANNOTATED_TYPE].value_counts().to_dict()
94
  print(f"Loaded {len(sha_set)} SHAs for deduplication")
95
+ print(f"Existing type counts: {type_counts}")
96
+ return sha_set, type_counts
97
  except FileNotFoundError:
98
  print(f"No existing samples found at {file_path}")
99
+ return set(), {}
100
  except Exception as e:
101
+ print(f"Error loading existing data: {e}")
102
+ return set(), {}
103
 
104
 
105
  def load_ccs_dataset(file_path: str) -> pd.DataFrame:
106
+ """Load CCS dataset CSV and validate required columns exist."""
107
  try:
108
  df = pd.read_csv(file_path)
109
  if df.empty:
 
123
  def save_to_csv(
124
  data: List[Dict[str, str]], output_path: str, columns: List[str]
125
  ) -> None:
126
+ """Save sampled commit data to CSV file, appending if file exists."""
127
  import os
128
 
129
  os.makedirs(os.path.dirname(output_path), exist_ok=True)
 
145
  def group_commits_by_type(
146
  df: pd.DataFrame, valid_types: List[str]
147
  ) -> Dict[str, pd.DataFrame]:
148
+ """Filter commits by valid types and group into separate DataFrames by type."""
149
  type_mask = df[COLUMN_ANNOTATED_TYPE].isin(valid_types)
150
  valid_df = df[type_mask].copy()
151
 
 
163
  def sample_commits_for_type(
164
  df: pd.DataFrame, count: int, output_columns: List[str]
165
  ) -> List[Dict[str, str]]:
166
+ """Randomly sample specified count of commits from DataFrame."""
167
+ sampled_df = df.sample(n=count, random_state=RANDOM_SEED)
168
  return sampled_df[output_columns].to_dict("records")
169
 
170
 
171
  def extract_diffs(sampled_data: List[Dict[str, str]], output_dir: str) -> None:
172
+ """Create individual diff files organized by commit type in subdirectories."""
173
  import os
174
 
175
  type_counts = {}
 
201
  record[COLUMN_GIT_DIFF],
202
  ]
203
 
204
+ with open(filepath, "w", encoding="utf-8") as f:
205
  f.write("\n".join(content_lines))
206
 
207
  print(f"Extracted {len(sampled_data)} diff files to {output_dir}")
208
 
209
  def remove_excluded_commits(df: pd.DataFrame, excluded_shas: Set[str]) -> pd.DataFrame:
210
+ """Remove commits with SHAs listed in the excluded commits file."""
211
  before_count = len(df)
212
  print(f"Initial commit count: {before_count}")
213
 
 
222
 
223
  def main() -> None:
224
  """
225
+ Execute atomic sampling pipeline for CCS dataset:
226
+ 1. Load dataset, existing SHAs and type counts for deduplication and sampling
227
+ 2. Remove excluded commits by SHA
228
+ 3. Remove existing commits to prevent duplicates
229
+ 4. Normalize CI commit types to CICD
230
+ 5. Filter commits exceeding token limits
231
+ 6. Sample needed amounts per type to reach target
232
+ 7. Save results and extract individual diff files (new samples only)
233
  """
234
  print("Starting atomic sampling strategy for CCS dataset")
235
  print("=" * 50)
236
 
237
+ # Step 1: Load dataset, backup SHAs and existing type counts
238
+ print("Step 1: Loading dataset, backup SHAs and existing type counts")
239
+ existing_shas, existing_type_counts = load_shas_and_type_counts(SAMPLED_CSV_PATH)
240
+ excluded_shas, _ = load_shas_and_type_counts(EXCLUDED_COMMITS_PATH)
241
  ccs_df = load_ccs_dataset(CCS_SOURCE_PATH)
242
 
243
  # Step 2: Remove excluded commits
244
  print("\nStep 2: Removing excluded commits")
245
  ccs_df = remove_excluded_commits(ccs_df, excluded_shas)
246
 
247
+ # Step 3: Remove existing commits
248
+ print("\nStep 3: Removing existing commits")
249
+ ccs_df = remove_existing_commits(ccs_df, existing_shas)
250
+
251
+ # Step 4: Apply CI->CICD normalization
252
+ print("\nStep 4: Applying CI->CICD normalization")
253
  ccs_df = normalize_dataset(ccs_df)
254
 
 
 
 
255
 
256
+ # Step 5: Apply token-based filtering
257
+ print("\nStep 5: Applying token-based filtering")
258
+ ccs_df = remove_long_token_commits(ccs_df)
259
 
260
+ # Step 6: Group by type and sample
261
  print("\nStep 6: Grouping by type and random sampling")
262
  commits_by_type = group_commits_by_type(ccs_df, CONVENTIONAL_COMMIT_TYPES)
263
 
264
  all_sampled_data = []
265
+
266
+ for commit_type, commits_df in commits_by_type.items():
267
+ existing_type_count = existing_type_counts.get(commit_type, 0)
268
+ needed_count = max(0, SAMPLES_PER_TYPE - existing_type_count) # Skip if target reached
269
+ available_type_count = len(commits_df)
270
+ actual_sample_count = min(needed_count, available_type_count)
271
+
272
+ if needed_count == 0:
273
+ print(f" {commit_type}: target reached, skipping")
274
+ continue
275
+ if actual_sample_count <= 0:
276
+ print(f" {commit_type}: no commits available")
277
+ continue
278
+
279
  sampled_data = sample_commits_for_type(
280
+ commits_df, actual_sample_count, OUTPUT_COLUMNS
281
  )
282
  all_sampled_data.extend(sampled_data)
283
+ print(f" {commit_type}: sampled {actual_sample_count} commits")
284
 
285
  print(f"Random sampling: generated {len(all_sampled_data)} samples total")
286
 
287
  # Step 7: Save results and extract diffs
288
  print("\nStep 7: Saving results and extracting diffs")
289
+ if all_sampled_data:
290
+ save_to_csv(all_sampled_data, SAMPLED_CSV_PATH, OUTPUT_COLUMNS)
291
+ extract_diffs(all_sampled_data, DIFF_OUTPUT_DIR)
292
+ else:
293
+ print("No new samples to save - all types have reached target counts")
294
 
295
  # Final summary
296
  print("\n" + "=" * 50)
297
  print("Atomic sampling completed successfully!")
298
+ print(f"New samples added: {len(all_sampled_data)}")
 
 
 
 
 
 
 
 
299
 
300
 
301
  if __name__ == "__main__":