File size: 5,377 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 |
# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: CC-BY-NC-4.0
"""
Create Benchmarks from Assets
Generates document splitting benchmarks from structured assets.
Usage:
python main.py --strategy multi_category_concat
python main.py --strategy single_category_concat --num-docs-train 500
"""
import argparse
from pathlib import Path
from loguru import logger
from services.asset_loader import AssetLoader
from services.split_manager import SplitManager
from services.benchmark_generator import BenchmarkGenerator
from services.benchmark_writer import BenchmarkWriter
from services.shuffle_strategies import get_strategy, STRATEGIES
def main():
parser = argparse.ArgumentParser(description='Create benchmarks from assets')
# Paths
parser.add_argument('--assets-path', default='data/assets',
help='Path to assets from create_assets')
parser.add_argument('--output-path', default='data/benchmarks',
help='Output path for benchmarks')
parser.add_argument('--split-mapping', default='data/metadata/split_mapping.json',
help='Path to split mapping JSON (created if not exists)')
# Strategy
parser.add_argument('--strategy', choices=list(STRATEGIES.keys()) + ['all'], default='all',
help='Shuffle strategy to use (default: all strategies)')
# Split configuration
parser.add_argument('--num-docs-train', type=int, default=800,
help='Number of spliced documents for training')
parser.add_argument('--num-docs-test', type=int, default=500,
help='Number of spliced documents for testing')
parser.add_argument('--num-docs-val', type=int, default=200,
help='Number of spliced documents for validation')
# Strategy parameters
parser.add_argument('--size', choices=['small', 'large'], default='small',
help='Benchmark size: small (5-20 pages) or large (20-500 pages)')
parser.add_argument('--random-seed', type=int, default=42,
help='Random seed for reproducibility')
args = parser.parse_args()
# Set page ranges based on size
if args.size == 'small':
min_pages, max_pages = 5, 20
else: # large
min_pages, max_pages = 20, 500
# Determine which strategies to run
if args.strategy == 'all':
strategies_to_run = list(STRATEGIES.keys())
logger.info(f"Creating benchmarks for all strategies: {strategies_to_run}")
else:
strategies_to_run = [args.strategy]
logger.info(f"Creating benchmark with strategy: {args.strategy}")
logger.info(f"Size: {args.size} ({min_pages}-{max_pages} pages)")
# Load assets
loader = AssetLoader(assets_path=args.assets_path)
documents_by_type = loader.load_all_documents()
if not documents_by_type:
logger.error("No documents loaded. Check assets path.")
return
# Create or load split
split_manager = SplitManager(random_seed=args.random_seed)
if args.split_mapping and Path(args.split_mapping).exists():
logger.info(f"Loading existing split from {args.split_mapping}")
splits = split_manager.load_split(args.split_mapping)
else:
logger.info("Creating new split")
splits = split_manager.create_split(documents_by_type)
# Save split mapping to metadata folder
split_path = Path(args.split_mapping)
split_path.parent.mkdir(parents=True, exist_ok=True)
split_manager.save_split(splits, str(split_path))
# Run for each strategy
for strategy_name in strategies_to_run:
logger.info(f"\n{'='*60}")
logger.info(f"Processing strategy: {strategy_name}")
logger.info(f"{'='*60}\n")
# Initialize strategy
strategy = get_strategy(
strategy_name,
min_pages=min_pages,
max_pages=max_pages,
random_seed=args.random_seed
)
# Initialize generator and writer
generator = BenchmarkGenerator(strategy=strategy)
writer = BenchmarkWriter(
output_base_path=str(Path(args.output_path) / strategy_name / args.size),
assets_path=args.assets_path
)
# Generate benchmarks for each split
split_configs = [
('train', args.num_docs_train),
('test', args.num_docs_test),
('validation', args.num_docs_val)
]
for split_name, num_docs in split_configs:
if num_docs <= 0:
logger.info(f"Skipping {split_name} (num_docs=0)")
continue
logger.info(f"Generating {split_name} benchmark...")
benchmark_set = generator.generate_for_split(
documents_by_type=documents_by_type,
doc_names_for_split=splits[split_name],
num_spliced_docs=num_docs,
split_name=split_name,
benchmark_name=strategy_name
)
writer.save_benchmark_set(benchmark_set, split_name)
logger.info(f"Completed {split_name}: {benchmark_set.statistics}")
logger.info("\n" + "="*60)
logger.info("All benchmark creation complete!")
logger.info("="*60)
if __name__ == '__main__':
main()
|