dataset-builder / data3 /extract_functions_v2.py
SunDou's picture
Upload data3/extract_functions_v2.py with huggingface_hub
be0bdd8 verified
#!/usr/bin/env python3
"""
Extract individual functions from enhanced_dataset.csv and create a new dataset.
Each function becomes a separate row in the new dataset.
Version 2: Better handling of malformed CSV/JSON
"""
import csv
import json
import re
from collections import defaultdict
import sys
def clean_json_string(json_str):
"""
Clean up malformed JSON strings that may have been corrupted by CSV formatting.
"""
# Remove extra spaces in key names that might have been inserted
# This is a bit risky but we'll try to handle common cases
# Replace common malformed patterns
json_str = re.sub(r'"\s*function_nam\s*e\s*"', '"function_name"', json_str)
json_str = re.sub(r'"\s*function_start_line\s*"', '"function_start_line"', json_str)
json_str = re.sub(r'"\s*function_end_line\s*"', '"function_end_line"', json_str)
json_str = re.sub(r'"\s*relevance_score\s*"', '"relevance_score"', json_str)
json_str = re.sub(r'"\s*relevance_reason\s*"', '"relevance_reason"', json_str)
json_str = re.sub(r'"\s*doc_start_line\s*"', '"doc_start_line"', json_str)
json_str = re.sub(r'"\s*doc_end_line\s*"', '"doc_end_line"', json_str)
# Remove markdown bold markers that might have been inserted
json_str = json_str.replace('**', '')
# Fix spacing issues in keys
json_str = re.sub(r'"\s*([a-z_]+)\s*([a-z_]+)\s*([a-z_]*)\s*":',
lambda m: '"' + m.group(1) + m.group(2) + (m.group(3) if m.group(3) else '') + '":',
json_str)
return json_str
def extract_function_content(text, start_line, end_line):
"""
Extract function content from text based on line number range.
Args:
text: The full code text
start_line: Starting line number (1-indexed)
end_line: Ending line number (1-indexed)
Returns:
Extracted function content as string
"""
lines = text.split('\n')
# Convert to 0-indexed (since start_line is 1-indexed, we subtract 1)
start_idx = max(0, start_line - 1)
end_idx = min(len(lines), end_line) # end_line is inclusive, so we don't subtract 1
function_lines = lines[start_idx:end_idx]
return '\n'.join(function_lines)
def process_dataset(input_file, output_file):
"""
Process enhanced_dataset.csv and extract functions.
Args:
input_file: Path to enhanced_dataset.csv
output_file: Path to output CSV file
"""
print(f"Reading from: {input_file}")
print(f"Writing to: {output_file}")
# Statistics
total_rows = 0
total_functions = 0
score_distribution = defaultdict(int)
skipped_rows = 0
parse_errors = 0
empty_function_info = 0
with open(input_file, 'r', encoding='utf-8') as infile, \
open(output_file, 'w', encoding='utf-8', newline='') as outfile:
reader = csv.DictReader(infile)
# Define output columns
fieldnames = [
'original_index', # Original row number
'function_index', # Index within the file
'repo_name',
'path',
'language',
'license',
'keyword',
'text_hash',
'config',
'split',
'repo_path',
'ds_source',
'function_name',
'function_start_line',
'function_end_line',
'doc_start_line',
'doc_end_line',
'relevance_score',
'relevance_reason',
'function_content'
]
writer = csv.DictWriter(outfile, fieldnames=fieldnames)
writer.writeheader()
# Store all function rows for later sorting
all_function_rows = []
print("\nProcessing rows...")
for row in reader:
total_rows += 1
if total_rows % 1000 == 0:
print(f"Processed {total_rows} rows, extracted {total_functions} functions, errors: {parse_errors}...", end='\r')
# Parse function_info JSON
function_info_str = row.get('function_info', '[]')
if not function_info_str or function_info_str.strip() == '':
empty_function_info += 1
skipped_rows += 1
continue
# Clean the JSON string
function_info_str = clean_json_string(function_info_str)
# Handle potential CSV escaping issues
try:
# First try direct JSON parsing
function_info_list = json.loads(function_info_str)
except (json.JSONDecodeError, ValueError) as e:
# If that fails, try with ast.literal_eval as backup
try:
import ast
function_info_list = ast.literal_eval(function_info_str)
except:
# If still fails, skip this row
parse_errors += 1
if parse_errors <= 5: # Only print first 5 errors
print(f"\nWarning: Failed to parse function_info in row {total_rows}")
if parse_errors == 5:
print("(Suppressing further parse error messages...)")
skipped_rows += 1
continue
# Validate that we got a list
if not isinstance(function_info_list, list):
skipped_rows += 1
continue
# Get the original text
text = row.get('text', '')
# Extract each function
for func_idx, func_info in enumerate(function_info_list):
# Validate func_info is a dictionary
if not isinstance(func_info, dict):
continue
# Extract function content
start_line = func_info.get('function_start_line', 0)
end_line = func_info.get('function_end_line', 0)
# Ensure they are integers
try:
start_line = int(start_line) if start_line else 0
end_line = int(end_line) if end_line else 0
except (ValueError, TypeError):
start_line = 0
end_line = 0
if start_line > 0 and end_line > 0:
function_content = extract_function_content(text, start_line, end_line)
else:
function_content = ""
# Get relevance score
relevance_score = func_info.get('relevance_score', 0)
# Ensure it's an integer
try:
relevance_score = int(relevance_score) if relevance_score else 0
except (ValueError, TypeError):
relevance_score = 0
# Track score distribution (in buckets of 10)
score_bucket = (relevance_score // 10) * 10
score_distribution[score_bucket] += 1
# Create new row
new_row = {
'original_index': row.get('Unnamed: 0', row.get('Unnamed: 0.1', total_rows - 1)),
'function_index': func_idx,
'repo_name': row.get('repo_name', ''),
'path': row.get('path', ''),
'language': row.get('language', ''),
'license': row.get('license', ''),
'keyword': row.get('keyword', ''),
'text_hash': row.get('text_hash', ''),
'config': row.get('config', ''),
'split': row.get('split', ''),
'repo_path': row.get('repo_path', ''),
'ds_source': row.get('ds_source', ''),
'function_name': func_info.get('function_name', ''),
'function_start_line': start_line,
'function_end_line': end_line,
'doc_start_line': func_info.get('doc_start_line', ''),
'doc_end_line': func_info.get('doc_end_line', ''),
'relevance_score': relevance_score,
'relevance_reason': func_info.get('relevance_reason', ''),
'function_content': function_content
}
all_function_rows.append(new_row)
total_functions += 1
print(f"\n\nTotal rows processed: {total_rows}")
print(f"Total functions extracted: {total_functions}")
print(f"Skipped rows:")
print(f" - Empty function_info: {empty_function_info}")
print(f" - Parse errors: {parse_errors}")
print(f" - Total skipped: {skipped_rows}")
# Sort by relevance_score (descending - highest first)
print("\nSorting by relevance score...")
all_function_rows.sort(key=lambda x: x['relevance_score'], reverse=True)
# Write sorted rows
print("Writing sorted data to output file...")
for row in all_function_rows:
writer.writerow(row)
print(f"\nSuccessfully written {total_functions} functions to {output_file}")
# Print score distribution
print("\n" + "="*70)
print("SCORE DISTRIBUTION")
print("="*70)
print(f"{'Score Range':<15} {'Count':<12} {'Percentage':<12} {'Visualization'}")
print("-"*70)
# Sort by score range (descending)
sorted_scores = sorted(score_distribution.items(), reverse=True)
# Filter out anomalous scores (very negative values)
normal_scores = [(k, v) for k, v in sorted_scores if k >= 0]
anomalous_scores = [(k, v) for k, v in sorted_scores if k < 0]
for score_bucket, count in normal_scores:
percentage = (count / total_functions * 100) if total_functions > 0 else 0
bar = '█' * min(50, int(percentage / 2)) # Scale bar to fit
print(f"{score_bucket:>3}-{score_bucket+9:<9} {count:<12} {percentage:>6.2f}% {bar}")
if anomalous_scores:
print("\nAnomalous scores (negative or out of range):")
for score_bucket, count in anomalous_scores:
percentage = (count / total_functions * 100) if total_functions > 0 else 0
print(f"{score_bucket:>15} {count:<12} {percentage:>6.2f}%")
print("-"*70)
print(f"{'Total':<15} {total_functions:<12} {'100.00%':<12}")
print("="*70)
# Additional statistics
if total_functions > 0:
# Filter out anomalous scores for statistics
valid_scores = [row['relevance_score'] for row in all_function_rows
if 0 <= row['relevance_score'] <= 100]
if valid_scores:
avg_score = sum(valid_scores) / len(valid_scores)
max_score = max(valid_scores)
min_score = min(valid_scores)
print(f"\nScore Statistics (valid scores 0-100 only):")
print(f" Average Score: {avg_score:.2f}")
print(f" Maximum Score: {max_score}")
print(f" Minimum Score: {min_score}")
print(f" Valid Functions: {len(valid_scores)} / {total_functions}")
if __name__ == "__main__":
input_file = "enhanced_dataset.csv"
output_file = "function_dataset_v2.csv"
# Allow command line arguments
if len(sys.argv) > 1:
input_file = sys.argv[1]
if len(sys.argv) > 2:
output_file = sys.argv[2]
try:
process_dataset(input_file, output_file)
print("\n✅ Processing complete!")
except FileNotFoundError:
print(f"❌ Error: File '{input_file}' not found.")
sys.exit(1)
except Exception as e:
print(f"❌ Error: {e}")
import traceback
traceback.print_exc()
sys.exit(1)