Scandium-Dataset / scripts /audit_phase3_property.py
Scandiumlabs's picture
Upload folder using huggingface_hub
df32294 verified
Raw
History Blame Contribute Delete
8.62 kB
"""Phase 3: Property Audit — distributions, outliers, physical constraints."""
import json, time
from pathlib import Path
from collections import Counter
import numpy as np
OUT = Path("scripts/audit_reports")
OUT.mkdir(parents=True, exist_ok=True)
DATASET = "dataset/entries_final_v3.json"
AUDIT_DIR = Path.cwd() if Path.cwd().name == "Scandium-Dataset" else Path("/home/shamique/Scandium Labs SSB/Scandium-Dataset")
severity_counts = {"CRITICAL": 0, "HIGH": 0, "MEDIUM": 0, "LOW": 0, "PASS": 0}
findings = []
def finding(severity, phase, check, status, detail):
severity_counts[severity] += 1
findings.append({"severity": severity, "phase": phase, "check": check, "status": status, "detail": str(detail)[:200]})
s = "🔴" if severity == "CRITICAL" else "🟠" if severity == "HIGH" else "🟡" if severity == "MEDIUM" else "🔵" if severity == "LOW" else "✅"
print(f" {s} [{severity:8s}] {check}: {str(detail)[:120]}")
def analyze_prop(values, name, unit, phys_min=None, phys_max=None):
vals = np.array([v for v in values if v is not None])
n_missing = sum(1 for v in values if v is None)
print(f"\n {name} ({unit}):")
print(f" n={len(vals):,}, missing={n_missing:,}, mean={np.mean(vals):.4f}, "
f"median={np.median(vals):.4f}, std={np.std(vals):.4f}")
print(f" min={np.min(vals):.4f}, max={np.max(vals):.4f}")
# Percentiles
for p in [1, 5, 25, 50, 75, 95, 99]:
print(f" P{p:2d}={np.percentile(vals, p):.4f}", end="")
print()
# Skewness and kurtosis
from scipy import stats
print(f" skewness={stats.skew(vals):.4f}, kurtosis={stats.kurtosis(vals):.4f}")
return vals, n_missing
def main():
print("=" * 60)
print(" PHASE 3: PROPERTY AUDIT")
print("=" * 60)
with open(AUDIT_DIR / DATASET) as f:
entries = json.load(f)
N = len(entries)
print(f" Loaded {N:,} entries")
FE = [e.get("formation_energy_per_atom") for e in entries]
EaH = [e.get("energy_above_hull") for e in entries]
BG = [e.get("band_gap") for e in entries]
# --- Distribution analysis ---
print("\n--- Distribution Analysis ---")
fe_vals, fe_missing = analyze_prop(FE, "Formation Energy", "eV/atom")
eah_vals, eah_missing = analyze_prop(EaH, "Energy Above Hull", "eV/atom")
bg_vals, bg_missing = analyze_prop(BG, "Band Gap", "eV")
# --- Physical constraints ---
print("\n--- Physical Constraints ---")
# BG >= 0
neg_bg = np.sum(bg_vals < -0.01)
if neg_bg:
finding("CRITICAL", "physical", "negative_band_gap", f"{neg_bg:,} entries with BG < 0", "")
else:
finding("PASS", "physical", "band_gap_non_negative", f"{np.sum(bg_vals < 0)} negative", "")
# EaH >= 0
neg_eah = np.sum(eah_vals < -0.01)
if neg_eah:
finding("CRITICAL", "physical", "negative_hull_energy", f"{neg_eah:,} entries with EaH < 0", "")
else:
finding("PASS", "physical", "hull_non_negative", f"{neg_eah} negative", "")
# FE within physical range (-5 to +3 eV/atom)
extreme_fe = np.sum((fe_vals < -6) | (fe_vals > 4))
if extreme_fe:
finding("HIGH", "physical", "extreme_formation_energy",
f"{extreme_fe:,} entries with FE outside [-6, 4]", "")
else:
finding("PASS", "physical", "fe_reasonable_range", "all within [-6, 4]", "")
# --- Outliers ---
print("\n--- Outliers ---")
# IQR-based outlier detection
def iqr_outliers(vals, name, n_iqr=3):
q1, q3 = np.percentile(vals, [25, 75])
iqr = q3 - q1
low = q1 - n_iqr * iqr
high = q3 + n_iqr * iqr
outliers = vals[(vals < low) | (vals > high)]
return outliers, low, high
fe_out, fe_low, fe_high = iqr_outliers(fe_vals, "FE")
eah_out, eah_low, eah_high = iqr_outliers(eah_vals, "EaH")
bg_out, bg_low, bg_high = iqr_outliers(bg_vals, "BG")
finding("LOW", "outliers", "fe_iqr_outliers", f"{len(fe_out):,} (thresholds: {fe_low:.3f}, {fe_high:.3f})", "")
finding("LOW", "outliers", "eah_iqr_outliers", f"{len(eah_out):,} (thresholds: {eah_low:.3f}, {eah_high:.3f})", "")
finding("LOW", "outliers", "bg_iqr_outliers", f"{len(bg_out):,} (thresholds: {bg_low:.3f}, {bg_high:.3f})", "")
# Extreme values (hard thresholds)
extreme_fe_hard = fe_vals[(fe_vals < -5) | (fe_vals > 5)]
max_fe = float(np.max(fe_vals))
min_fe = float(np.min(fe_vals))
severity = "CRITICAL" if max_fe > 10 or min_fe < -10 else "HIGH"
finding(severity, "outliers", "extreme_fe_hard",
f"{len(extreme_fe_hard):,} with |FE| > 5 (max={max_fe:.2f}, min={min_fe:.2f})",
f"values: {np.sort(extreme_fe_hard)[:10].tolist()}")
extreme_eah_hard = eah_vals[eah_vals > 5]
max_eah = float(np.max(eah_vals)) if len(eah_vals) > 0 else 0
if len(extreme_eah_hard) > 0:
sev = "CRITICAL" if max_eah > 10 else "HIGH"
finding(sev, "outliers", "extreme_eah_hard",
f"{len(extreme_eah_hard):,} with EaH > 5 (max={max_eah:.2f})", "")
else:
finding("PASS", "outliers", "extreme_eah_hard", "0", "")
# --- Correlations ---
print("\n--- Property Correlations ---")
from scipy import stats
# Build aligned arrays for correlation
import pandas as pd
df = pd.DataFrame({"fe": FE, "eah": EaH, "bg": BG}).dropna()
fe_aligned = df["fe"].values[:50000]
eah_aligned = df["eah"].values[:50000]
bg_aligned = df["bg"].values[:50000]
r1, _ = stats.pearsonr(fe_aligned, eah_aligned)
print(f" FE vs EaH: r={r1:.4f} (n={len(fe_aligned):,})")
r2, _ = stats.pearsonr(fe_aligned, bg_aligned)
print(f" FE vs BG: r={r2:.4f} (n={len(fe_aligned):,})")
r3, _ = stats.pearsonr(eah_aligned, bg_aligned)
print(f" EaH vs BG: r={r3:.4f} (n={len(eah_aligned):,})")
finding("PASS", "correlations", "fe_eah_correlation", f"r={r1:.4f}", "")
finding("PASS", "correlations", "fe_bg_correlation", f"r={r2:.4f}", "")
# --- Missing values per source ---
print("\n--- Missing Values Per Source ---")
for src in ["mp", "oqmd", "jarvis"]:
subset = [e for e in entries if e.get("source") == src]
for prop in ["formation_energy_per_atom", "energy_above_hull", "band_gap"]:
missing = sum(1 for e in subset if e.get(prop) is None)
if missing > 0:
pct = 100 * missing / len(subset)
sev = "CRITICAL" if pct > 50 else "HIGH" if pct > 10 else "MEDIUM"
finding(sev, "missing", f"{src}_{prop}_missing",
f"{missing:,}/{len(subset):,} ({pct:.1f}%)", "")
print(f"\n{'=' * 60}")
print(f" PHASE 3 SUMMARY")
print(f" CRITICAL: {severity_counts['CRITICAL']}")
print(f" HIGH: {severity_counts['HIGH']}")
print(f" MEDIUM: {severity_counts['MEDIUM']}")
print(f" LOW: {severity_counts['LOW']}")
print(f" PASS: {severity_counts['PASS']}")
print(f"{'=' * 60}")
report = {
"phase": "Phase 3: Property Audit",
"timestamp": time.strftime("%Y-%m-%d %H:%M:%S"),
"total_entries": N,
"properties": {
"formation_energy": {
"n": int(len(fe_vals)), "missing": fe_missing,
"mean": float(np.mean(fe_vals)), "median": float(np.median(fe_vals)),
"std": float(np.std(fe_vals)), "min": float(np.min(fe_vals)), "max": float(np.max(fe_vals)),
"skewness": float(stats.skew(fe_vals)), "kurtosis": float(stats.kurtosis(fe_vals)),
"outliers_iqr": int(len(fe_out)),
},
"energy_above_hull": {
"n": int(len(eah_vals)), "missing": eah_missing,
"mean": float(np.mean(eah_vals)), "median": float(np.median(eah_vals)),
"std": float(np.std(eah_vals)), "min": float(np.min(eah_vals)), "max": float(np.max(eah_vals)),
"outliers_iqr": int(len(eah_out)),
},
"band_gap": {
"n": int(len(bg_vals)), "missing": bg_missing,
"mean": float(np.mean(bg_vals)), "median": float(np.median(bg_vals)),
"std": float(np.std(bg_vals)), "min": float(np.min(bg_vals)), "max": float(np.max(bg_vals)),
"outliers_iqr": int(len(bg_out)),
},
},
"findings": findings,
"summary": dict(severity_counts),
}
with open(OUT / "phase3_property_audit.json", "w") as f:
json.dump(report, f, indent=2)
print(f"\n Report: {OUT / 'phase3_property_audit.json'}")
if __name__ == "__main__":
main()