Datasets:
Size:
1M<n<10M
ArXiv:
Tags:
Document_Understanding
Document_Packet_Splitting
Document_Comprehension
Document_Classification
Document_Recognition
Document_Segmentation
DOI:
License:
File size: 4,311 Bytes
165da3c | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 | # Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: CC-BY-NC-4.0
from typing import List, Dict
import uuid
from loguru import logger
from models import DocumentAsset, SplicedDocument, SourceDocument, GroundTruthPage
from services.shuffle_strategies.base_strategy import BaseStrategy
class PolyRand(BaseStrategy):
"""DocSplit-Poly-Rand: Multi category document pages randomization.
Similar to Poly-Seq but applies complete randomization across all pages,
representing maximum entropy scenarios. Stress-tests model robustness under
worst-case conditions where no structural assumptions hold.
"""
def generate(
self,
documents_by_type: Dict[str, List[DocumentAsset]],
doc_names_for_split: Dict[str, List[str]],
num_spliced_docs: int
) -> List[SplicedDocument]:
available_docs = self._get_available_docs(documents_by_type, doc_names_for_split)
doc_types = list(available_docs.keys())
if len(doc_types) < 2:
raise ValueError("Need at least 2 document types for multi-category strategy")
spliced_documents = []
for i in range(num_spliced_docs):
# Set target page count
target_pages = self.rng.randint(self.min_pages, self.max_pages)
# Collect entire documents from different types
selected_docs = []
used_types = set()
total_pages = 0
while total_pages < target_pages:
available_types = [t for t in doc_types if t not in used_types]
if not available_types:
break
doc_type = self.rng.choice(available_types)
used_types.add(doc_type)
available = [d for d in available_docs[doc_type] if d.page_count <= self.max_pages]
if not available:
continue
doc = self.rng.choice(available)
if total_pages + doc.page_count > self.max_pages:
continue
selected_docs.append(doc)
total_pages += doc.page_count
# Collect all pages from selected documents
all_pages = []
for doc in selected_docs:
for page_num in range(1, doc.page_count + 1):
all_pages.append({
'doc_type': doc.doc_type,
'doc_name': doc.doc_name,
'source_page': page_num
})
# Fully shuffle all pages
self.rng.shuffle(all_pages)
# Build source documents and ground truth
source_docs_dict = {}
ground_truth = []
for idx, page_info in enumerate(all_pages, start=1):
doc_key = (page_info['doc_type'], page_info['doc_name'])
if doc_key not in source_docs_dict:
source_docs_dict[doc_key] = []
source_docs_dict[doc_key].append(page_info['source_page'])
ground_truth.append(GroundTruthPage(
page_num=idx,
doc_type=page_info['doc_type'],
source_doc=page_info['doc_name'],
source_page=page_info['source_page']
))
source_documents = [
SourceDocument(doc_type=doc_type, doc_name=doc_name, pages=pages)
for (doc_type, doc_name), pages in source_docs_dict.items()
]
if len(all_pages) >= self.min_pages:
spliced_doc = SplicedDocument(
spliced_doc_id=str(uuid.uuid4()),
source_documents=source_documents,
ground_truth=ground_truth,
total_pages=len(all_pages)
)
spliced_documents.append(spliced_doc)
logger.info(f"Generated {len(spliced_documents)} DocSplit-Poly-Rand documents")
return spliced_documents
|