Upload examples/statistics/compute_statistics.py with huggingface_hub

examples/statistics/compute_statistics.py

@@ -0,0 +1,32 @@
+"""Example: Compute per-family and per-source statistics."""
+import json
+import numpy as np
+from collections import Counter
+
+with open("dataset/entries_final_v3.json") as f:
+    entries = json.load(f)
+
+# Per-family statistics
+families = Counter(f for e in entries for f in e.get("families", ["unknown"]))
+print("Family Distribution:")
+for fam, count in families.most_common():
+    pct = 100 * count / len(entries)
+    print(f"  {fam:25s}: {count:>7,} ({pct:.1f}%)")
+
+# Per-source FE distribution
+print("\nFE Distribution by Source:")
+for src in ["mp", "oqmd", "jarvis"]:
+    subset = [e for e in entries if e.get("source") == src]
+    fe_vals = [e.get("formation_energy_per_atom", 0) for e in subset
+               if e.get("formation_energy_per_atom") is not None]
+    print(f"  {src:8s}: mean={np.mean(fe_vals):.3f}  "
+          f"median={np.median(fe_vals):.3f}  "
+          f"std={np.std(fe_vals):.3f}  "
+          f"N={len(fe_vals):,}")
+
+# Coverage analysis
+print("\nProperty Coverage:")
+for prop in ["formation_energy_per_atom", "energy_above_hull", "band_gap"]:
+    present = sum(1 for e in entries if e.get(prop) is not None)
+    print(f"  {prop:35s}: {present:>7,} / {len(entries):,} "
+          f"({100*present/len(entries):.1f}%)")