Bobchen1980's picture
Upload folder using huggingface_hub (part 3)
cfbe945 verified
Raw History Blame Contribute Delete
4.52 kB
#!/usr/bin/env python3
"""Check table coverage and intervention identities without rerunning propagation."""
import csv
import hashlib
import json
from pathlib import Path
import numpy as np
HERE = Path(__file__).resolve().parent
def digest(path):
h = hashlib.sha256()
with path.open("rb") as stream:
for block in iter(lambda: stream.read(1024 * 1024), b""):
h.update(block)
return h.hexdigest()
def main():
summary = json.loads((HERE / "results" / "summary.json").read_text())
rows = list(csv.DictReader((HERE / "results" / "channel_errors.csv").open()))
groups = list(csv.DictReader((HERE / "results" / "group_errors.csv").open()))
assert len(rows) == 3 * 3 * 2 * 8 * 12 == summary["channel_rows"]
assert len(groups) == 3 * 3 * 2 * 8 * 8 == summary["group_rows"]
assert summary["config_sha256"] == digest(HERE / "config.json")
assert summary["analysis_code_sha256"] == digest(HERE / "analyze.py")
assert all(digest(HERE.parent / filename) == value for filename, value in summary["input_sha256"].items())
assert summary["max_normalized_squared_error_identity_residual"] < 1e-12
assert summary["max_relative_remove_ritz_trajectory_identity"] < 1e-12
max_sum_residual = 0.0
for group in groups:
if group["group"] != "all":
continue
keys = ("material", "method", "modes_per_patch", "reference_grid", "window_kind", "start", "end")
subset = [r for r in rows if all(r[k] == group[k] for k in keys)]
assert len(subset) == 12
np.testing.assert_allclose(sum(float(r["reference_energy_share"]) for r in subset), 1, atol=1e-14)
for intervention in ("remove_ritz", "fine_frequency"):
contribution = sum(float(r[f"{intervention}_squared_error_change_contribution"]) for r in subset)
target = float(group[f"{intervention}_relative_error"])**2 - float(group["original_relative_error"])**2
max_sum_residual = max(max_sum_residual, abs(contribution - target))
assert abs(contribution - target) < 1e-12
max_window_partition_residual = 0.0
for full in rows:
if full["window_kind"] != "cumulative" or float(full["end"]) != 8:
continue
keys = ("material", "method", "modes_per_patch", "reference_grid", "source", "receiver")
parts = [r for r in rows if r["window_kind"] == "disjoint" and all(r[k] == full[k] for k in keys)]
assert len(parts) == 4
den = float(full["reference_energy"])
np.testing.assert_allclose(sum(float(r["reference_energy"]) for r in parts), den, rtol=2e-14)
assert sum(int(r["samples"]) for r in parts) == int(full["samples"])
for metric in ("original_relative_error", "remove_ritz_relative_error", "fine_frequency_relative_error"):
reconstructed = sum(float(r[metric]) ** 2 * float(r["reference_energy"]) for r in parts) / den
difference = abs(reconstructed - float(full[metric]) ** 2)
max_window_partition_residual = max(max_window_partition_residual, difference)
assert difference < 1e-12
previous = list(csv.DictReader((HERE.parent / "modal_diagnostics" / "results" / "reference_evaluation.csv").open()))
for old in previous:
new = next(r for r in groups if r["reference_grid"] == "512" and r["window_kind"] == "cumulative"
and float(r["end"]) == 8 and r["group"] == "all"
and r["material"] == old["material"] and r["method"] == old["method"]
and r["modes_per_patch"] == old["m"])
for current, earlier in (("original_relative_error", "total_error_vs_512"),
("remove_ritz_relative_error", "error_without_ritz_shift_vs_512"),
("fine_frequency_relative_error", "error_with_fine_frequencies_vs_512")):
np.testing.assert_allclose(float(new[current]), float(old[earlier]), rtol=2e-14, atol=2e-14)
result = dict(passed=True, channel_rows=len(rows), group_rows=len(groups),
input_files_checked=len(summary["input_sha256"]), max_additive_contribution_residual=max_sum_residual,
max_disjoint_window_partition_residual=max_window_partition_residual,
previous_aggregate_comparisons_checked=3 * len(previous))
(HERE / "results" / "verification.json").write_text(json.dumps(result, indent=2) + "\n")
print(json.dumps(result, indent=2))
if __name__ == "__main__":
main()