Experiment: polyatomic+polyatomic on qm9

FOLD 1/5
GNN Train Loss: 0.1272
RMSE: 1.3036, MAE: 0.9929, R2: 0.9081

FOLD 2/5
GNN Train Loss: 0.1336
RMSE: 1.2810, MAE: 0.9819, R2: 0.9128

FOLD 3/5
GNN Train Loss: 0.1297
RMSE: 1.4949, MAE: 1.2115, R2: 0.8838

FOLD 4/5
GNN Train Loss: 0.1260
RMSE: 1.8907, MAE: 1.6264, R2: 0.8157

FOLD 5/5
GNN Train Loss: 0.1265
RMSE: 1.4262, MAE: 1.1400, R2: 0.8914

====== K-FOLD SUMMARY ======
RMSE: 1.4814  [1.3191, 1.6941]
MAE: 1.1924  [1.0201, 1.4145]
R2: 0.8822  [0.8478, 0.9067]

★ Using fold 2 model for test inference
RMSE: 1.2848, MAE: 0.9849, R2: 0.9128

====== HELD-OUT TEST METRICS ======
{'rmse': 1.2848398114709658, 'mae': 0.9848641157150269, 'r2': 0.912821888923645}
Test RMSE: 1.2799  (95 % CI: 1.1559–1.3954)

Test MAE : 0.9838  (95 % CI: 0.8926–1.0708)

Test R²  : 0.9126  (95 % CI: 0.8916–0.9307)