PACT-Net / logs /freesolv-bench /sage /sage_ecfp_freesolv_20250729_150029.txt
rk-random's picture
Upload folder using huggingface_hub
9a67fbe verified
raw
history blame
849 Bytes
Experiment: sage+ecfp on freesolv
FOLD 1/5
GNN Train Loss: 0.0044
RMSE: 2.1845, MAE: 1.5875, R2: 0.6237
FOLD 2/5
GNN Train Loss: 0.0059
RMSE: 1.8647, MAE: 1.1633, R2: 0.7881
FOLD 3/5
GNN Train Loss: 0.0060
RMSE: 2.0170, MAE: 1.3163, R2: 0.7362
FOLD 4/5
GNN Train Loss: 0.0036
RMSE: 1.6653, MAE: 1.2153, R2: 0.7606
FOLD 5/5
GNN Train Loss: 0.0050
RMSE: 1.9074, MAE: 1.2864, R2: 0.7491
====== K-FOLD SUMMARY ======
RMSE: 1.9318 [1.7841, 2.0840]
MAE: 1.3189 [1.2043, 1.4671]
R2: 0.7308 [0.6739, 0.7716]
★ Using fold 4 model for test inference
RMSE: 2.5718, MAE: 1.7177, R2: 0.6006
====== HELD-OUT TEST METRICS ======
{'rmse': 2.5717555114503106, 'mae': 1.717659831047058, 'r2': 0.600629985332489}
Test RMSE: 2.5430 (95 % CI: 2.0052–3.2978)
Test MAE : 1.7124 (95 % CI: 1.4085–2.0726)
Test R² : 0.5959 (95 % CI: 0.4749–0.7242)