File size: 2,927 Bytes
8035461 | 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 | #!/usr/bin/env python3
"""Run the backtest with TRAINED models (loads saved model weights).
Usage:
python scripts/run_backtest_trained.py --data data/parquet/BTCUSDT_2021-05-19.parquet
python scripts/run_backtest_trained.py --data data/parquet/LUNAUSDT_2022-05-10.parquet
"""
import argparse
import logging
import sys
from pathlib import Path
ML_DIR = Path(__file__).resolve().parent.parent / "ml"
sys.path.insert(0, str(ML_DIR))
PROJECT_ROOT = Path(__file__).resolve().parent.parent
sys.path.insert(0, str(PROJECT_ROOT))
from flash_crash_watchdog.cascade import DetectionCascade
from flash_crash_watchdog.data.historical_loader import load_parquet
from flash_crash_watchdog.eval.backtest import run_backtest
from flash_crash_watchdog.models.stage2_isolation_forest import Stage2IsolationForest
from flash_crash_watchdog.models.stage3_tcn import Stage3TCN, TCNConfig
def main() -> int:
parser = argparse.ArgumentParser(description="Run backtest with trained models")
parser.add_argument("--data", required=True, help="Parquet file of crash data")
parser.add_argument("--config", default="configs/pipeline.yml")
parser.add_argument("--models", default="models/", help="Directory with trained models")
parser.add_argument("--output", default="results/backtest_trained.json")
parser.add_argument("--max-ticks", type=int, default=0,
help="Max ticks to process (0 = all, for speed use 500000)")
args = parser.parse_args()
logging.basicConfig(level=logging.INFO, format="%(asctime)s %(levelname)s %(message)s")
# Load cascade
cascade = DetectionCascade.from_config(args.config)
# Load trained models if they exist
models_dir = Path(args.models)
stage2_path = models_dir / "stage2_isolation_forest.joblib"
stage3_path = models_dir / "stage3_tcn.pt"
if stage2_path.exists():
logger.info("Loading trained Stage 2 from %s", stage2_path)
cascade.s2.load(stage2_path)
else:
logger.warning("No trained Stage 2 found at %s — using untrained fallback", stage2_path)
if stage3_path.exists():
logger.info("Loading trained Stage 3 from %s", stage3_path)
cascade.s3.load(stage3_path)
else:
logger.warning("No trained Stage 3 found at %s — using untrained fallback", stage3_path)
# Load data
df = load_parquet(args.data)
if args.max_ticks > 0 and len(df) > args.max_ticks:
logger.info("Sampling down to %d ticks (from %d) for speed", args.max_ticks, len(df))
indices = range(0, len(df), len(df) // args.max_ticks)
df = df.iloc[indices[:args.max_ticks]].copy()
# Run backtest
results = run_backtest(cascade, df)
results.print_summary()
output_path = Path(args.output)
output_path.parent.mkdir(parents=True, exist_ok=True)
results.save(output_path)
return 0
if __name__ == "__main__":
raise SystemExit(main())
|