File size: 5,518 Bytes
2adb41a | 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 | """
TrackChain CLI Step Skip-Logic Evaluator (tc.v1 SOTA).
Used by ml/scripts/run.sh to evaluate manifest fingerprints and quality gates.
Usage:
python ml/scripts/should_skip.py --step train_yolo [--force]
Exit code 0: Skip the step (valid checkpoint & manifest found)
Exit code 1: Run the step (missing, modified config/dataset, or failed gate)
"""
import sys
import argparse
from pathlib import Path
# Add repo root to python path
repo_root = Path(__file__).resolve().parent.parent.parent
if str(repo_root) not in sys.path:
sys.path.insert(0, str(repo_root))
from ml.core.manifest import (
compute_file_sha256,
compute_config_hash,
compute_dataset_hash,
check_training_skip,
check_export_skip,
load_manifest,
)
def evaluate_step(step_name: str, force: bool = False) -> bool:
"""Returns True if step should be SKIPPED, False if step should RUN."""
if force:
print(f"[{step_name}] Force flag enabled. Running step.")
return False
if step_name == "train_yolo":
# Resolve dataset yaml
data_yaml = repo_root / "data" / "external" / "rail_defects_expanded" / "data.yaml"
if not data_yaml.exists():
data_yaml = repo_root / "data" / "external" / "rail_defects" / "data.yaml"
config_yaml = repo_root / "ml" / "configs" / "detector.yaml"
ckpt = repo_root / "artifacts" / "checkpoints" / "vision" / "yolov8n_rail_best.pt"
manifest_p = repo_root / "artifacts" / "checkpoints" / "vision" / "yolo_manifest.json"
cfg_hash = compute_config_hash(config_yaml, overrides={"imgsz": 960, "freeze": 10, "dropout": 0.1, "close_mosaic": 10})
data_hash = compute_dataset_hash(data_yaml)
skip, reason = check_training_skip(
manifest_path=manifest_p,
checkpoint_path=ckpt,
current_config_hash=cfg_hash,
current_dataset_hash=data_hash,
min_metrics={"mAP50": 0.25},
force=force,
)
if skip:
print(f"[{step_name}] SKIP: {reason}")
return True
else:
print(f"[{step_name}] RUN: {reason}")
return False
elif step_name == "export_yolo_onnx":
src_pt = repo_root / "artifacts" / "checkpoints" / "vision" / "yolov8n_rail_best.pt"
export_file = repo_root / "artifacts" / "exports" / "yolov8n_rail_best.onnx"
manifest_p = repo_root / "artifacts" / "exports" / "export_manifest.json"
skip, reason = check_export_skip(
manifest_path=manifest_p,
export_path=export_file,
source_model_path=src_pt,
export_format="onnx",
force=force,
)
if skip:
print(f"[{step_name}] SKIP: {reason}")
return True
else:
print(f"[{step_name}] RUN: {reason}")
return False
elif step_name == "export_yolo_int8":
src_pt = repo_root / "artifacts" / "checkpoints" / "vision" / "yolov8n_rail_best.pt"
export_file = repo_root / "artifacts" / "exports" / "yolov8n_rail_best_int8.onnx"
manifest_p = repo_root / "artifacts" / "exports" / "export_manifest.json"
skip, reason = check_export_skip(
manifest_path=manifest_p,
export_path=export_file,
source_model_path=src_pt,
export_format="int8",
force=force,
)
if skip:
print(f"[{step_name}] SKIP: {reason}")
return True
else:
print(f"[{step_name}] RUN: {reason}")
return False
elif step_name == "train_patchcore":
ckpt = repo_root / "artifacts" / "checkpoints" / "vision" / "patchcore_memory_bank.npz"
calib = repo_root / "artifacts" / "calibration" / "patchcore_calibration.json"
if ckpt.exists() and calib.exists():
print(f"[{step_name}] SKIP: PatchCore memory bank and calibration exist.")
return True
print(f"[{step_name}] RUN: Memory bank or calibration missing.")
return False
elif step_name == "train_bilstm":
ckpt = repo_root / "artifacts" / "checkpoints" / "geometry" / "bilstm_fault_typing_enhanced.pt"
calib = repo_root / "artifacts" / "calibration" / "bilstm_temp.json"
if ckpt.exists() and calib.exists():
print(f"[{step_name}] SKIP: Bi-LSTM checkpoint and calibration exist.")
return True
print(f"[{step_name}] RUN: Bi-LSTM weights or calibration missing.")
return False
elif step_name == "train_vae":
ckpt = repo_root / "artifacts" / "checkpoints" / "geometry" / "sequence_vae_enhanced.pt"
calib = repo_root / "artifacts" / "calibration" / "vae_calibration.json"
if ckpt.exists() and calib.exists():
print(f"[{step_name}] SKIP: Seq-VAE checkpoint and EVT calibration exist.")
return True
print(f"[{step_name}] RUN: Seq-VAE weights or calibration missing.")
return False
# Default: Run step
return False
if __name__ == "__main__":
parser = argparse.ArgumentParser(description="Evaluate step skip-logic using manifests.")
parser.add_argument("--step", required=True, help="Step name to evaluate")
parser.add_argument("--force", action="store_true", help="Force execution")
args = parser.parse_args()
should_skip = evaluate_step(step_name=args.step, force=args.force)
# Exit 0 if skipping, Exit 1 if running
sys.exit(0 if should_skip else 1)
|