============================================================ Teacher-Mark Binary Classifier Training ============================================================ Data dir: outlook/data_correction Device: cuda Epochs: 100 Batch size: 32 LR: 0.001 Latent dim: 64 MP steps: 3 Checkpoint: outlook/ckpt/correction.pt Log: /public/home/dxw/Codes/afem/outlook/ckpt/correction_train_log.json Loading dataset... Found 1998 files across 100 samples Train samples: 80 Val samples: 20 Train: 1598 graphs, 2203757 nodes, 65355 positive (3.0%) Val: 400 graphs, 513740 nodes, 15208 positive (3.0%) Loaded in 5.6s Training... Epoch 0 | train=1.2278 val=1.1452 | auc=0.7556 topk=0.165 phys=0.120 lr=1.00e-03 | 3.5s Epoch 10 | train=0.7870 val=0.8045 | auc=0.8856 topk=0.287 phys=0.120 lr=1.00e-03 | 1.9s Epoch 20 | train=0.6734 val=0.7438 | auc=0.9029 topk=0.318 phys=0.120 lr=1.00e-03 | 1.9s Epoch 30 | train=0.6258 val=0.7229 | auc=0.9079 topk=0.342 phys=0.120 lr=1.00e-03 | 1.9s Epoch 40 | train=0.5536 val=0.7068 | auc=0.9134 topk=0.351 phys=0.120 lr=5.00e-04 | 1.9s Epoch 50 | train=0.5223 val=0.7220 | auc=0.9140 topk=0.359 phys=0.120 lr=2.50e-04 | 1.9s Epoch 60 | train=0.5032 val=0.7522 | auc=0.9117 topk=0.363 phys=0.120 lr=1.25e-04 | 1.9s Epoch 70 | train=0.4880 val=0.7545 | auc=0.9127 topk=0.362 phys=0.120 lr=1.25e-04 | 1.9s Epoch 80 | train=0.4804 val=0.7735 | auc=0.9121 topk=0.368 phys=0.120 lr=6.25e-05 | 1.9s Epoch 90 | train=0.4763 val=0.7756 | auc=0.9121 topk=0.365 phys=0.120 lr=3.13e-05 | 1.9s Epoch 99 | train=0.4737 val=0.7832 | auc=0.9116 topk=0.364 phys=0.120 lr=1.56e-05 | 1.9s Training completed in 191.7s Checkpoint saved: outlook/ckpt/correction.pt Training log saved: /public/home/dxw/Codes/afem/outlook/ckpt/correction_train_log.json ============================================================ Best epoch: 38 val_loss: 0.6904 val_auc: 0.9152 val_topk: 0.347 phys_topk: 0.120 GNN beats physics: YES ============================================================ Done.