============================================================ 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 1236 files across 100 samples Train samples: 80 Val samples: 20 Train: 991 graphs, 5274265 nodes, 157771 positive (3.0%) Val: 245 graphs, 1296751 nodes, 38781 positive (3.0%) Loaded in 7.1s Training... Epoch 0 | train=1.2012 val=1.1091 | auc=0.7760 topk=0.230 phys=0.153 lr=1.00e-03 | 4.7s Epoch 10 | train=0.5825 val=0.6353 | auc=0.9287 topk=0.372 phys=0.153 lr=1.00e-03 | 3.2s Epoch 20 | train=0.5019 val=0.5901 | auc=0.9378 topk=0.416 phys=0.153 lr=1.00e-03 | 3.2s Epoch 30 | train=0.5124 val=0.5745 | auc=0.9417 topk=0.423 phys=0.153 lr=1.00e-03 | 3.1s Epoch 40 | train=0.4355 val=0.5334 | auc=0.9498 topk=0.466 phys=0.153 lr=1.00e-03 | 3.2s Epoch 50 | train=0.4143 val=0.5467 | auc=0.9520 topk=0.490 phys=0.153 lr=1.00e-03 | 3.2s Epoch 60 | train=0.3744 val=0.5686 | auc=0.9506 topk=0.486 phys=0.153 lr=5.00e-04 | 3.2s Epoch 70 | train=0.3544 val=0.5963 | auc=0.9510 topk=0.491 phys=0.153 lr=2.50e-04 | 3.1s Epoch 80 | train=0.3443 val=0.6211 | auc=0.9505 topk=0.490 phys=0.153 lr=1.25e-04 | 3.3s Epoch 90 | train=0.3419 val=0.6288 | auc=0.9502 topk=0.491 phys=0.153 lr=6.25e-05 | 3.2s Epoch 99 | train=0.3382 val=0.6329 | auc=0.9501 topk=0.490 phys=0.153 lr=3.13e-05 | 3.2s Training completed in 321.8s Checkpoint saved: outlook/ckpt/correction.pt Training log saved: /public/home/dxw/Codes/afem/outlook/ckpt/correction_train_log.json ============================================================ Best epoch: 40 val_loss: 0.5334 val_auc: 0.9498 val_topk: 0.466 phys_topk: 0.153 GNN beats physics: YES ============================================================ Done.