HongchengGao's picture
Align task taxonomy and IDs with the updated paper; main evaluation 306
3de3b50 verified
Raw History Blame Contribute Delete
2.19 kB
#!/usr/bin/env python3
import csv
import hashlib
import math
from eval_utils import (
desktop_root,
print_result,
read_numeric_csv,
run_submission,
)
JOB_HASHES = {
"job_baseline.py": "49a962bcda502b901c1a8d9fc5e018bd05e9a3218438e20f0cec9f3655c8a736",
"job_yaw.py": "b0f1c852d0ce7b514a649582c23b6a9c2bddd3af8e13970f768aa1c57c1842b5",
}
EXPECTED = [2190.397160, 436.442701, 2304.300414, 660.711994, 5.200119, 51.385736]
def check() -> bool:
root = desktop_root()
for name, expected_hash in JOB_HASHES.items():
actual_hash = hashlib.sha256((root / name).read_bytes()).hexdigest()
if actual_hash != expected_hash:
return False
run_submission(
root,
"run_pipeline.py",
["baseline.csv", "yaw.csv"],
timeout=300,
)
run_submission(root, "compare.py", ["comparison_report.txt"])
with (root / "comparison_report.txt").open("r", encoding="utf-8", newline="") as handle:
rows = list(csv.reader(handle))
if len(rows) != 4 or [row[0] for row in rows] != [
"baseline",
"yaw",
"Gain_Total_Percent",
"Gain_Downstream_Percent",
]:
return False
values = [
float(rows[0][1]),
float(rows[0][2]),
float(rows[1][1]),
float(rows[1][2]),
float(rows[2][1]),
float(rows[3][1]),
]
baseline_values = read_numeric_csv(root / "baseline.csv", 1, 2)[0]
yaw_values = read_numeric_csv(root / "yaw.csv", 1, 2)[0]
if not all(
math.isclose(actual, source, rel_tol=1e-6, abs_tol=1e-6)
for actual, source in zip(values[:4], baseline_values + yaw_values)
):
return False
if not all(math.isclose(a, e, rel_tol=1e-4, abs_tol=1e-3) for a, e in zip(values, EXPECTED)):
return False
total_gain = (values[2] - values[0]) / values[0] * 100.0
downstream_gain = (values[3] - values[1]) / values[1] * 100.0
return abs(values[4] - total_gain) <= 1e-3 and abs(values[5] - downstream_gain) <= 1e-3
if __name__ == "__main__":
try:
result = check()
except Exception:
result = False
raise SystemExit(print_result(result))