doc_split / src /assets /services /pdf_loader.py
mmiakashs's picture
Release dataset generator
165da3c verified
# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: CC-BY-NC-4.0
import os
from pathlib import Path
from typing import List, Optional
import csv
from loguru import logger
import fitz
from models import Document
class PdfLoader:
"""Loads PDF files from raw data directory."""
def __init__(self, raw_data_path: str = ".data/rvl-cdip-mp/raw_data", max_file_size_mb: float = 150):
"""Initialize PDF loader.
Args:
raw_data_path: Path to raw data directory
max_file_size_mb: Maximum file size in MB for valid PDFs
"""
self.raw_data_path = Path(raw_data_path)
self.max_file_size_mb = max_file_size_mb
def validate_pdf(self, file_path: Path) -> tuple[bool, str, int]:
"""Validate if a PDF is readable and within size limits.
Args:
file_path: Path to PDF file
Returns:
Tuple of (is_valid, status_message, page_count)
"""
try:
# Check file size
size_bytes = file_path.stat().st_size
size_mb = size_bytes / (1024 * 1024)
if size_bytes == 0:
return False, "ZERO_SIZE", 0
if size_mb > self.max_file_size_mb:
return False, f"TOO_LARGE ({size_mb:.1f}MB)", 0
# Check PDF header
with open(file_path, "rb") as f:
header = f.read(4)
if header != b"%PDF":
return False, "INVALID_PDF_HEADER", 0
# Test readability with PyMuPDF
doc = fitz.open(str(file_path))
if doc.is_encrypted:
doc.close()
return False, "ENCRYPTED_PDF", 0
page_count = doc.page_count
if page_count == 0:
doc.close()
return False, "NO_PAGES", 0
# Try to access first page
page = doc[0]
page.get_text()
doc.close()
return True, "VALID", page_count
except Exception as e:
return False, f"ERROR: {str(e)[:50]}", 0
def get_all_documents(self, exclude_types: Optional[List[str]] = None) -> List[Document]:
"""Get all valid PDF documents from raw data directory.
Args:
exclude_types: List of document types to exclude (e.g., ['language'])
"""
if exclude_types is None:
exclude_types = []
documents = []
for doc_type_dir in self.raw_data_path.iterdir():
if not doc_type_dir.is_dir():
continue
doc_type = doc_type_dir.name
if doc_type in exclude_types:
logger.info(f"Skipping excluded type: {doc_type}")
continue
for pdf_file in doc_type_dir.glob("*.pdf"):
is_valid, status, page_count = self.validate_pdf(pdf_file)
if not is_valid:
logger.warning(f"Skipping invalid PDF {pdf_file}: {status}")
continue
doc = Document(
doc_type=doc_type,
doc_name=pdf_file.stem,
filename=pdf_file.name,
absolute_filepath=str(pdf_file.absolute()),
page_count=page_count
)
documents.append(doc)
logger.info(f"Loaded {len(documents)} valid documents")
return documents
def save_document_mapping(self, documents: List[Document], output_path: str):
"""Save document mapping to CSV."""
os.makedirs(os.path.dirname(output_path), exist_ok=True)
with open(output_path, 'w', newline='') as f:
writer = csv.DictWriter(f, fieldnames=['type', 'doc_name', 'filename', 'pages', 'validation_status'])
writer.writeheader()
for doc in documents:
writer.writerow({
'type': doc.doc_type,
'doc_name': doc.doc_name,
'filename': doc.filename,
'pages': doc.page_count,
'validation_status': 'VALID'
})
logger.info(f"Saved document mapping to {output_path}")