GitTaskBench / test_scripts /Scrapy_03 /test_script.py
Nicole-Yi's picture
Upload 103 files
4a5e815 verified
Raw
History Blame Contribute Delete
14.3 kB
import argparse
import os
import xml.etree.ElementTree as ET
import json
import re
from datetime import datetime
def check_file_valid(file_path: str) -> bool:
if not os.path.isfile(file_path):
print(f"❌ File does not exist: {file_path}")
return False
if os.path.getsize(file_path) == 0:
print(f"❌ File is empty: {file_path}")
return False
return True
def clean_text(text):
"""Clean and normalize text for comparison"""
if text is None:
return ""
# Remove quotes at beginning and end (different agents might handle quotes differently)
text = text.strip()
text = re.sub(r'^[""]|[""]$', '', text)
# Normalize whitespace
text = re.sub(r'\s+', ' ', text)
return text.strip()
def extract_xml_content(file_path):
"""Extract content from XML file, filtering out possible comments and non-XML content"""
try:
with open(file_path, 'r', encoding='utf-8') as f:
content = f.read()
# Keep XML declaration if found, otherwise add one
xml_decl_match = re.search(r'<\?xml[^>]+\?>', content)
if xml_decl_match:
xml_decl = xml_decl_match.group(0)
content = content[xml_decl_match.end():]
else:
xml_decl = '<?xml version="1.0" encoding="UTF-8"?>'
# Remove XML comments
content = re.sub(r'<!--.*?-->', '', content, flags=re.DOTALL)
# Remove illegal XML characters
content = re.sub(r'[\x00-\x08\x0B\x0C\x0E-\x1F\x7F]', '', content)
# Find content between first < and last > (remove possible non-XML prefixes/suffixes)
start_idx = content.find('<')
end_idx = content.rfind('>')
if start_idx >= 0 and end_idx > start_idx:
content = content[start_idx:end_idx + 1]
# Ensure there's a root element
if not re.search(r'<\w+[^>]*>.*</\w+>', content, re.DOTALL):
content = f"<root>{content}</root>"
return xml_decl + content
except Exception as e:
print(f"❌ Failed to read XML file: {e}")
return None
def parse_xml(file_path):
"""Generic XML parser that identifies various possible structures"""
try:
xml_content = extract_xml_content(file_path)
if not xml_content:
return []
# Try to parse XML
root = ET.fromstring(xml_content)
except ET.ParseError as e:
print(f"❌ XML parse error, attempting repair: {e}")
try:
# Try to handle incomplete XML
xml_content = re.sub(r'<\?xml[^>]+\?>', '', xml_content)
xml_content = f"<root>{xml_content}</root>"
root = ET.fromstring(xml_content)
except ET.ParseError as e2:
print(f"❌ XML repair failed: {e2}")
# Final attempt with more aggressive repair
try:
pattern = r'<(.*?)>(.*?)</(.*?)>'
matches = re.findall(pattern, xml_content, re.DOTALL)
items_xml = ''.join([f'<item>{m[1]}</item>' for m in matches if len(m) >= 2])
root = ET.fromstring(f"<root>{items_xml}</root>")
except:
print("❌ All XML repair attempts failed")
return []
records = []
# Get all possible item elements
# Common tag patterns: quote, item, entry, record, etc.
item_tags = ['quote', 'item', 'entry', 'record', 'row', 'data']
# Try to find direct children of root element
direct_children = list(root)
# If there are multiple direct children with similar structure, they might be our items
candidates = []
if len(direct_children) > 1:
# Check if these elements have similar structure
first_child_tags = set([child.tag for child in direct_children[0]] if len(direct_children) > 0 else [])
if all(set([child.tag for child in elem]) == first_child_tags for elem in direct_children[1:]):
candidates.extend(direct_children)
# If no suitable direct children found, try predefined item tags
if not candidates:
for tag in item_tags:
items = root.findall(f".//{tag}")
if items:
candidates.extend(items)
break
# If still not found, try to find any element containing author and text sub-elements
if not candidates:
for elem in root.findall(".//*"):
has_author = any(child.tag == 'author' or 'author' in child.tag.lower() for child in elem)
has_text = any(child.tag == 'text' or 'text' in child.tag.lower() for child in elem)
if has_author and has_text:
candidates.append(elem)
# Extract author and text information from candidate items
for item in candidates:
author_content = ""
text_content = ""
# Method 1: Directly look for specific tags
for child in item:
if child.tag.lower() == 'author' or 'author' in child.tag.lower():
author_content = clean_text(child.text)
elif child.tag.lower() == 'text' or 'text' in child.tag.lower() or child.tag.lower() == 'quote':
text_content = clean_text(child.text)
# Method 2: Look for elements containing specific keywords
if not author_content or not text_content:
for elem in item.findall(".//*"):
if (elem.tag.lower() == 'author' or 'author' in elem.tag.lower()) and not author_content:
author_content = clean_text(elem.text)
elif (
elem.tag.lower() == 'text' or 'text' in elem.tag.lower() or 'quote' in elem.tag.lower() or 'content' in elem.tag.lower()) and not text_content:
text_content = clean_text(elem.text)
# Method 3: Look in attributes
for attr_name, attr_value in item.attrib.items():
if ('author' in attr_name.lower()) and not author_content:
author_content = clean_text(attr_value)
elif (
'text' in attr_name.lower() or 'quote' in attr_name.lower() or 'content' in attr_name.lower()) and not text_content:
text_content = clean_text(attr_value)
# Method 4: If item only has plain text with specific pattern, try to extract
if (not author_content or not text_content) and item.text and item.text.strip():
# Try to match author-quote pattern like "Author Name: 'Quote text'"
author_quote_match = re.search(r'([^:]+):\s*[\'"](.+?)[\'"]', item.text.strip())
if author_quote_match:
if not author_content:
author_content = clean_text(author_quote_match.group(1))
if not text_content:
text_content = clean_text(author_quote_match.group(2))
# Only include records with content
if author_content and text_content:
records.append({
"author": author_content,
"text": text_content
})
# If no records found with above methods, try direct text extraction
if not records:
print("⚠️ No records found with standard parsing methods, attempting direct text extraction...")
try:
with open(file_path, 'r', encoding='utf-8') as f:
content = f.read()
# Try to match author and quote patterns
pattern = r'<(author|text)>(.*?)</(author|text)>'
matches = re.findall(pattern, content, re.IGNORECASE | re.DOTALL)
authors = [clean_text(m[1]) for m in matches if m[0].lower() == 'author' and m[2].lower() == 'author']
texts = [clean_text(m[1]) for m in matches if m[0].lower() == 'text' and m[2].lower() == 'text']
# If equal number of authors and texts found, assume they are pairs
if len(authors) == len(texts) and len(authors) > 0:
records = [{"author": a, "text": t} for a, t in zip(authors, texts)]
except Exception as e:
print(f"❌ Text extraction failed: {e}")
return records
def find_best_matches(preds, gts):
"""Find best matches between predictions and ground truth"""
matched_pairs = []
remaining_preds = list(range(len(preds)))
for i, gt_item in enumerate(gts):
best_score = -1
best_idx = -1
for j in remaining_preds:
pred_item = preds[j]
score = 0
# Score based on text similarity
for field in ["author", "text"]:
gt_val = gt_item.get(field, "").lower()
pred_val = pred_item.get(field, "").lower()
# Exact match
if gt_val == pred_val:
score += 2
# Partial match (one contains the other)
elif gt_val in pred_val or pred_val in gt_val:
score += 1
if score > best_score:
best_score = score
best_idx = j
if best_idx >= 0:
matched_pairs.append((i, best_idx))
remaining_preds.remove(best_idx)
else:
# If no match found, use default index matching
default_idx = i if i < len(preds) else 0
matched_pairs.append((i, default_idx))
return matched_pairs
def evaluate_scraping_xml(pred_file, gt_file, threshold=0.95, result_file=None):
if not check_file_valid(pred_file) or not check_file_valid(gt_file):
return {}, False
preds = parse_xml(pred_file)
gts = parse_xml(gt_file) # Changed from parse_csv to parse_xml
if not preds:
print("❌ Could not extract valid records from prediction XML")
if result_file:
result = {
"Process": True,
"Result": False,
"TimePoint": datetime.now().isoformat(),
"comments": "Prediction XML parsing failed, could not extract valid records"
}
with open(result_file, "a", encoding="utf-8") as f:
f.write(json.dumps(result, ensure_ascii=False, default=str) + "\n")
return {}, False
if not gts:
print("❌ Could not extract valid records from ground truth XML")
if result_file:
result = {
"Process": True,
"Result": False,
"TimePoint": datetime.now().isoformat(),
"comments": "Ground truth XML parsing failed, could not extract valid records"
}
with open(result_file, "a", encoding="utf-8") as f:
f.write(json.dumps(result, ensure_ascii=False, default=str) + "\n")
return {}, False
print(f"✓ Extracted {len(preds)} records from prediction XML")
print(f"✓ Extracted {len(gts)} records from ground truth XML")
for i, record in enumerate(preds[:3]): # Only show first 3 as examples
print(f" Prediction Record {i + 1}: Author='{record['author']}', Text='{record['text'][:30]}...'")
if len(preds) > 3:
print(f" ... plus {len(preds) - 3} more prediction records")
if len(preds) != len(gts):
print(
f"⚠️ Prediction and ground truth lengths differ (predicted {len(preds)}, truth {len(gts)}), finding best matches.")
num_samples = min(len(preds), len(gts))
if num_samples == 0:
print("❌ Error: No data to compare.")
return {}, False
fields = ["author", "text"]
correct_counts = {field: 0 for field in fields}
# Find best matches between predictions and ground truth
matched_pairs = find_best_matches(preds, gts[:num_samples])
for gt_idx, pred_idx in matched_pairs:
for field in fields:
gt_val = clean_text(gts[gt_idx].get(field, "")).lower()
pred_val = clean_text(preds[pred_idx].get(field, "")).lower()
# Exact match or near-exact match (allowing minor differences)
if gt_val == pred_val:
correct_counts[field] += 1
# If ground truth is fully contained in prediction, also count as correct
elif gt_val and gt_val in pred_val:
correct_counts[field] += 1
print(f"⚠️ Field '{field}' partial match: GT='{gt_val}', Pred='{pred_val}'")
accuracies = {field: correct_counts[field] / num_samples for field in fields}
for field, acc in accuracies.items():
print(f"Field '{field}' accuracy: {acc:.4f}")
success = all(acc >= threshold for acc in accuracies.values())
print(
f"✅ Validation passed: All fields accuracy >= {threshold * 100}%" if success
else f"❌ Validation failed: Some fields accuracy < {threshold * 100}%")
if result_file:
# Ensure directory exists
os.makedirs(os.path.dirname(result_file), exist_ok=True)
result = {
"Process": True,
"Results": success,
"TimePoint": datetime.now().isoformat(),
"comments": f"Field-level accuracy: {accuracies}, {'meets' if success else 'does not meet'} {threshold * 100}% threshold"
}
with open(result_file, "a", encoding="utf-8") as f:
f.write(json.dumps(result, ensure_ascii=False, default=str) + "\n")
return accuracies, success
def main():
parser = argparse.ArgumentParser(
description="Evaluate accuracy of XML extraction results against Ground Truth XML.")
parser.add_argument('--output', type=str, required=True, help="Prediction XML file path")
parser.add_argument('--groundtruth', type=str, required=True, help="Ground truth XML file path")
parser.add_argument('--threshold', type=float, default=0.95, help="Field accuracy threshold")
parser.add_argument('--result', type=str, required=False, help="Output JSONL file path for results")
args = parser.parse_args()
evaluate_scraping_xml(args.output, args.groundtruth, args.threshold, args.result)
if __name__ == "__main__":
main()