95 lines
2.8 KiB
Python
95 lines
2.8 KiB
Python
from __future__ import annotations
|
|
|
|
import importlib.util
|
|
import json
|
|
import subprocess
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
|
|
SCRIPT = Path(__file__).parents[2] / "scripts" / "run_belgium_building_training_loop.py"
|
|
SPEC = importlib.util.spec_from_file_location("training_loop", SCRIPT)
|
|
assert SPEC and SPEC.loader
|
|
MODULE = importlib.util.module_from_spec(SPEC)
|
|
SPEC.loader.exec_module(MODULE)
|
|
|
|
|
|
def test_training_command_is_cuda_deterministic_and_bound_to_frozen_inputs(tmp_path: Path) -> None:
|
|
command = MODULE.training_command(
|
|
"yolo",
|
|
model=tmp_path / "base.pt",
|
|
data=tmp_path / "dataset.yaml",
|
|
project=tmp_path / "runs",
|
|
name="iteration-001",
|
|
epochs=160,
|
|
seed=42,
|
|
batch=2,
|
|
workers=4,
|
|
)
|
|
assert command[:2] == ["yolo", "train"]
|
|
assert "device=0" in command
|
|
assert "deterministic=True" in command
|
|
assert "seed=42" in command
|
|
assert "epochs=160" in command
|
|
assert "max_det=1000" in command
|
|
assert "imgsz=640" in command
|
|
assert "optimizer=auto" in command
|
|
assert "mosaic=1.0" in command
|
|
|
|
|
|
def test_training_command_supports_conservative_aerial_finetuning(tmp_path: Path) -> None:
|
|
command = MODULE.training_command(
|
|
"yolo",
|
|
model=tmp_path / "base.pt",
|
|
data=tmp_path / "dataset.yaml",
|
|
project=tmp_path / "runs",
|
|
name="aerial",
|
|
epochs=50,
|
|
seed=42,
|
|
batch=2,
|
|
workers=0,
|
|
optimizer="AdamW",
|
|
lr0=0.0001,
|
|
mosaic=0.0,
|
|
scale=0.2,
|
|
translate=0.05,
|
|
)
|
|
assert "optimizer=AdamW" in command
|
|
assert "lr0=0.0001" in command
|
|
assert "mosaic=0.0" in command
|
|
assert "scale=0.2" in command
|
|
assert "translate=0.05" in command
|
|
assert f"data={tmp_path / 'dataset.yaml'}" in command
|
|
|
|
|
|
def test_loop_refuses_failed_dataset_audit(tmp_path: Path) -> None:
|
|
audit = tmp_path / "audit.json"
|
|
audit.write_text(json.dumps({"status": "needs_attention", "low_variance_positive_tile_count": 4}))
|
|
result = subprocess.run(
|
|
[
|
|
sys.executable,
|
|
str(SCRIPT),
|
|
"--initial-model",
|
|
str(tmp_path / "base.pt"),
|
|
"--train-yaml",
|
|
str(tmp_path / "dataset.yaml"),
|
|
"--dataset-audit",
|
|
str(audit),
|
|
"--calibration-summary",
|
|
str(tmp_path / "cal.json"),
|
|
"--test-summary",
|
|
str(tmp_path / "test.json"),
|
|
"--background-summary",
|
|
str(tmp_path / "background.json"),
|
|
"--corpus-manifest",
|
|
str(tmp_path / "manifest.json"),
|
|
"--output-dir",
|
|
str(tmp_path / "output"),
|
|
],
|
|
capture_output=True,
|
|
text=True,
|
|
check=False,
|
|
)
|
|
assert result.returncode != 0
|
|
assert "Dataset audit is not ok" in result.stderr
|