experiments/status-aware/logs/ablations/summarized_structdense.log

============================================================
  StructGNN (act=64d hash, pos=0d) [NO GNN] [adapted]: kuhperdata-summ-exp
============================================================
Loading pre-computed data...
/workspace/ta-statute-law-retrieval/src/evaluate_paragnn.py:140: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  bm25_train_scores = torch.load(f"{output_dir}/bm25_train_scores.pt")
/workspace/ta-statute-law-retrieval/src/evaluate_paragnn.py:141: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  bm25_val_scores = torch.load(f"{output_dir}/bm25_val_scores.pt")
/workspace/ta-statute-law-retrieval/src/evaluate_paragnn.py:142: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  bm25_test_scores = torch.load(f"{output_dir}/bm25_test_scores.pt")
Loading paragraph store...
/workspace/ta-statute-law-retrieval/src/paragnn/graph_builder.py:40: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self.rr_const_emb = torch.load(self.emb_dir / "EMBD_CONST.pt")  # (13, 1024)
Computing structure features (act_encoder=hash, act_feat=64d, pos=0d)...
  2127 corpus docs, feature dim=64
Creating training dataset...
Creating ParaGNN training dataset...
  Created 2534 training examples from 1035 queries
Building val graph...
/workspace/ta-statute-law-retrieval/src/paragnn/graph_builder.py:81: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self._emb_cache[qid] = torch.load(path, map_location="cpu")
/workspace/ta-statute-law-retrieval/src/paragnn/graph_builder.py:88: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  self._emb_cache[key] = torch.load(path, map_location="cpu")
Building test graph...
Training StructGNN for epochs 1-100...
  Early stopping on VAL set (148 queries)
Pre-warming corpus embedding cache (2127 docs)...
  Cache ready.

Epoch 1:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 1:  10%|█         | 1/10 [00:02<00:20,  2.25s/it]
Epoch 1:  30%|███       | 3/10 [00:02<00:06,  1.17it/s]
Epoch 1:  50%|█████     | 5/10 [00:04<00:03,  1.42it/s]
Epoch 1:  70%|███████   | 7/10 [00:05<00:01,  1.61it/s]
Epoch 1:  80%|████████  | 8/10 [00:05<00:01,  1.98it/s]
Epoch 1:  90%|█████████ | 9/10 [00:05<00:00,  2.01it/s]
                                                       
  Epoch 1: loss=4.3555 val_MRR=0.0448 val_R@10=0.1052 val_Hit=16.2% alpha=1.0 (learned=0.523)
  → New best val Recall@10=0.1052, saved model

Epoch 2:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 2:  10%|█         | 1/10 [00:01<00:12,  1.34s/it]
Epoch 2:  30%|███       | 3/10 [00:02<00:04,  1.50it/s]
Epoch 2:  40%|████      | 4/10 [00:02<00:03,  1.98it/s]
Epoch 2:  50%|█████     | 5/10 [00:03<00:03,  1.61it/s]
Epoch 2:  70%|███████   | 7/10 [00:04<00:01,  1.88it/s]
Epoch 2:  90%|█████████ | 9/10 [00:04<00:00,  2.06it/s]
                                                       
  Epoch 2: loss=3.1736 val_MRR=0.0674 val_R@10=0.1375 val_Hit=20.3% alpha=1.0 (learned=0.789)
  → New best val Recall@10=0.1375, saved model

Epoch 3:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 3:  10%|█         | 1/10 [00:01<00:12,  1.42s/it]
Epoch 3:  30%|███       | 3/10 [00:02<00:04,  1.42it/s]
Epoch 3:  50%|█████     | 5/10 [00:03<00:02,  1.73it/s]
Epoch 3:  70%|███████   | 7/10 [00:04<00:01,  1.58it/s]
Epoch 3:  90%|█████████ | 9/10 [00:05<00:00,  1.77it/s]
                                                       
  Epoch 3: loss=2.0722 val_MRR=0.1692 val_R@10=0.2376 val_Hit=41.9% alpha=1.0 (learned=0.934)
  → New best val Recall@10=0.2376, saved model

Epoch 4:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 4:  10%|█         | 1/10 [00:01<00:12,  1.38s/it]
Epoch 4:  30%|███       | 3/10 [00:02<00:04,  1.40it/s]
Epoch 4:  50%|█████     | 5/10 [00:03<00:02,  1.77it/s]
Epoch 4:  70%|███████   | 7/10 [00:04<00:01,  1.96it/s]
Epoch 4:  90%|█████████ | 9/10 [00:04<00:00,  2.35it/s]
                                                       
  Epoch 4: loss=1.6867 val_MRR=0.2341 val_R@10=0.3198 val_Hit=52.0% alpha=1.0 (learned=0.973)
  → New best val Recall@10=0.3198, saved model

Epoch 5:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 5:  10%|█         | 1/10 [00:01<00:11,  1.33s/it]
Epoch 5:  30%|███       | 3/10 [00:02<00:04,  1.53it/s]
Epoch 5:  50%|█████     | 5/10 [00:03<00:02,  1.82it/s]
Epoch 5:  70%|███████   | 7/10 [00:03<00:01,  1.97it/s]
Epoch 5:  90%|█████████ | 9/10 [00:04<00:00,  2.27it/s]
                                                       
  Epoch 5: loss=1.4813 val_MRR=0.1205 val_R@10=0.3137 val_Hit=45.9% alpha=1.0 (learned=0.984)

Epoch 6:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 6:  10%|█         | 1/10 [00:01<00:13,  1.53s/it]
Epoch 6:  30%|███       | 3/10 [00:02<00:04,  1.42it/s]
Epoch 6:  50%|█████     | 5/10 [00:03<00:02,  1.76it/s]
Epoch 6:  70%|███████   | 7/10 [00:04<00:01,  1.88it/s]
Epoch 6:  90%|█████████ | 9/10 [00:04<00:00,  2.26it/s]
                                                       
  Epoch 6: loss=1.3228 val_MRR=0.2229 val_R@10=0.4135 val_Hit=59.5% alpha=0.9 (learned=0.988)
  → New best val Recall@10=0.4135, saved model

Epoch 7:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 7:  10%|█         | 1/10 [00:01<00:13,  1.53s/it]
Epoch 7:  30%|███       | 3/10 [00:02<00:04,  1.45it/s]
Epoch 7:  50%|█████     | 5/10 [00:03<00:02,  1.81it/s]
Epoch 7:  70%|███████   | 7/10 [00:04<00:01,  1.97it/s]
Epoch 7:  90%|█████████ | 9/10 [00:04<00:00,  2.27it/s]
                                                       
  Epoch 7: loss=1.2840 val_MRR=0.2844 val_R@10=0.4734 val_Hit=64.9% alpha=0.9 (learned=0.988)
  → New best val Recall@10=0.4734, saved model

Epoch 8:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 8:  10%|█         | 1/10 [00:01<00:13,  1.50s/it]
Epoch 8:  30%|███       | 3/10 [00:02<00:04,  1.43it/s]
Epoch 8:  50%|█████     | 5/10 [00:03<00:02,  1.79it/s]
Epoch 8:  70%|███████   | 7/10 [00:04<00:01,  2.02it/s]
Epoch 8:  90%|█████████ | 9/10 [00:04<00:00,  2.28it/s]
                                                       
  Epoch 8: loss=1.2200 val_MRR=0.3081 val_R@10=0.4756 val_Hit=64.2% alpha=0.9 (learned=0.988)
  → New best val Recall@10=0.4756, saved model

Epoch 9:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 9:  10%|█         | 1/10 [00:01<00:17,  1.91s/it]
Epoch 9:  30%|███       | 3/10 [00:02<00:05,  1.21it/s]
Epoch 9:  50%|█████     | 5/10 [00:03<00:03,  1.62it/s]
Epoch 9:  70%|███████   | 7/10 [00:04<00:01,  1.85it/s]
Epoch 9:  90%|█████████ | 9/10 [00:05<00:00,  2.25it/s]
                                                       
  Epoch 9: loss=1.1852 val_MRR=0.3672 val_R@10=0.5268 val_Hit=68.2% alpha=0.9 (learned=0.988)
  → New best val Recall@10=0.5268, saved model

Epoch 10:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 10:  10%|█         | 1/10 [00:01<00:14,  1.63s/it]
Epoch 10:  30%|███       | 3/10 [00:02<00:05,  1.37it/s]
Epoch 10:  50%|█████     | 5/10 [00:03<00:02,  1.71it/s]
Epoch 10:  70%|███████   | 7/10 [00:04<00:01,  1.89it/s]
Epoch 10:  90%|█████████ | 9/10 [00:04<00:00,  2.24it/s]
                                                        
  Epoch 10: loss=1.1463 val_MRR=0.4437 val_R@10=0.5935 val_Hit=75.0% alpha=0.9 (learned=0.988)
  → New best val Recall@10=0.5935, saved model

Epoch 11:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 11:  10%|█         | 1/10 [00:02<00:18,  2.00s/it]
Epoch 11:  30%|███       | 3/10 [00:02<00:05,  1.21it/s]
Epoch 11:  50%|█████     | 5/10 [00:03<00:03,  1.61it/s]
Epoch 11:  70%|███████   | 7/10 [00:04<00:01,  1.82it/s]
Epoch 11:  90%|█████████ | 9/10 [00:05<00:00,  2.19it/s]
                                                        
  Epoch 11: loss=1.1204 val_MRR=0.4886 val_R@10=0.6573 val_Hit=79.7% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.6573, saved model

Epoch 12:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 12:  10%|█         | 1/10 [00:01<00:12,  1.40s/it]
Epoch 12:  20%|██        | 2/10 [00:01<00:05,  1.36it/s]
Epoch 12:  30%|███       | 3/10 [00:02<00:05,  1.37it/s]
Epoch 12:  40%|████      | 4/10 [00:02<00:02,  2.02it/s]
Epoch 12:  50%|█████     | 5/10 [00:03<00:02,  1.72it/s]
Epoch 12:  60%|██████    | 6/10 [00:03<00:01,  2.30it/s]
Epoch 12:  70%|███████   | 7/10 [00:04<00:01,  1.82it/s]
Epoch 12:  90%|█████████ | 9/10 [00:04<00:00,  2.28it/s]
                                                        
  Epoch 12: loss=1.1160 val_MRR=0.5248 val_R@10=0.7167 val_Hit=83.1% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.7167, saved model

Epoch 13:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 13:  10%|█         | 1/10 [00:01<00:13,  1.50s/it]
Epoch 13:  30%|███       | 3/10 [00:02<00:05,  1.30it/s]
Epoch 13:  50%|█████     | 5/10 [00:03<00:02,  1.67it/s]
Epoch 13:  70%|███████   | 7/10 [00:04<00:01,  1.89it/s]
Epoch 13:  90%|█████████ | 9/10 [00:04<00:00,  2.20it/s]
                                                        
  Epoch 13: loss=1.0150 val_MRR=0.5578 val_R@10=0.7797 val_Hit=87.2% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.7797, saved model

Epoch 14:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 14:  10%|█         | 1/10 [00:01<00:12,  1.43s/it]
Epoch 14:  30%|███       | 3/10 [00:02<00:05,  1.33it/s]
Epoch 14:  50%|█████     | 5/10 [00:03<00:02,  1.67it/s]
Epoch 14:  70%|███████   | 7/10 [00:04<00:01,  1.90it/s]
Epoch 14:  90%|█████████ | 9/10 [00:04<00:00,  2.28it/s]
                                                        
  Epoch 14: loss=0.9499 val_MRR=0.5690 val_R@10=0.8062 val_Hit=89.2% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8062, saved model

Epoch 15:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 15:  10%|█         | 1/10 [00:01<00:12,  1.44s/it]
Epoch 15:  30%|███       | 3/10 [00:02<00:04,  1.51it/s]
Epoch 15:  50%|█████     | 5/10 [00:03<00:02,  1.70it/s]
Epoch 15:  60%|██████    | 6/10 [00:03<00:01,  2.12it/s]
Epoch 15:  70%|███████   | 7/10 [00:04<00:01,  1.77it/s]
Epoch 15:  80%|████████  | 8/10 [00:04<00:00,  2.25it/s]
Epoch 15:  90%|█████████ | 9/10 [00:04<00:00,  2.35it/s]
Epoch 15: 100%|██████████| 10/10 [00:05<00:00,  2.58it/s]
                                                         
  Epoch 15: loss=0.9519 val_MRR=0.5805 val_R@10=0.8458 val_Hit=91.9% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8458, saved model

Epoch 16:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 16:  10%|█         | 1/10 [00:01<00:12,  1.41s/it]
Epoch 16:  30%|███       | 3/10 [00:02<00:05,  1.22it/s]
Epoch 16:  50%|█████     | 5/10 [00:04<00:03,  1.33it/s]
Epoch 16:  70%|███████   | 7/10 [00:05<00:01,  1.52it/s]
Epoch 16:  90%|█████████ | 9/10 [00:05<00:00,  1.92it/s]
                                                        
  Epoch 16: loss=0.9150 val_MRR=0.5963 val_R@10=0.8619 val_Hit=93.2% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8619, saved model

Epoch 17:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 17:  10%|█         | 1/10 [00:01<00:11,  1.32s/it]
Epoch 17:  30%|███       | 3/10 [00:02<00:04,  1.55it/s]
Epoch 17:  50%|█████     | 5/10 [00:03<00:02,  1.77it/s]
Epoch 17:  70%|███████   | 7/10 [00:03<00:01,  2.01it/s]
Epoch 17:  80%|████████  | 8/10 [00:04<00:00,  2.44it/s]
Epoch 17:  90%|█████████ | 9/10 [00:04<00:00,  2.29it/s]
                                                        
  Epoch 17: loss=0.9200 val_MRR=0.5980 val_R@10=0.8725 val_Hit=93.9% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8725, saved model

Epoch 18:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 18:  10%|█         | 1/10 [00:01<00:12,  1.44s/it]
Epoch 18:  30%|███       | 3/10 [00:02<00:04,  1.42it/s]
Epoch 18:  50%|█████     | 5/10 [00:03<00:02,  1.76it/s]
Epoch 18:  60%|██████    | 6/10 [00:03<00:01,  2.14it/s]
Epoch 18:  70%|███████   | 7/10 [00:04<00:01,  1.59it/s]
Epoch 18:  90%|█████████ | 9/10 [00:05<00:00,  1.96it/s]
                                                        
  Epoch 18: loss=0.8942 val_MRR=0.6137 val_R@10=0.8908 val_Hit=95.3% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8908, saved model

Epoch 19:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 19:  10%|█         | 1/10 [00:01<00:12,  1.39s/it]
Epoch 19:  30%|███       | 3/10 [00:02<00:04,  1.47it/s]
Epoch 19:  50%|█████     | 5/10 [00:03<00:02,  1.77it/s]
Epoch 19:  60%|██████    | 6/10 [00:03<00:01,  2.18it/s]
Epoch 19:  70%|███████   | 7/10 [00:04<00:01,  1.70it/s]
Epoch 19:  90%|█████████ | 9/10 [00:04<00:00,  2.13it/s]
                                                        
  Epoch 19: loss=0.8800 val_MRR=0.6295 val_R@10=0.8917 val_Hit=95.3% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8917, saved model

Epoch 20:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 20:  10%|█         | 1/10 [00:01<00:12,  1.44s/it]
Epoch 20:  30%|███       | 3/10 [00:02<00:04,  1.41it/s]
Epoch 20:  50%|█████     | 5/10 [00:03<00:02,  1.73it/s]
Epoch 20:  70%|███████   | 7/10 [00:04<00:01,  1.78it/s]
Epoch 20:  90%|█████████ | 9/10 [00:04<00:00,  2.16it/s]
                                                        
  Epoch 20: loss=0.8816 val_MRR=0.6259 val_R@10=0.8976 val_Hit=95.9% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8976, saved model

Epoch 21:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 21:  10%|█         | 1/10 [00:01<00:12,  1.38s/it]
Epoch 21:  30%|███       | 3/10 [00:02<00:04,  1.49it/s]
Epoch 21:  50%|█████     | 5/10 [00:03<00:02,  1.81it/s]
Epoch 21:  70%|███████   | 7/10 [00:03<00:01,  1.98it/s]
Epoch 21:  90%|█████████ | 9/10 [00:05<00:00,  1.94it/s]
                                                        
  Epoch 21: loss=0.8005 val_MRR=0.6460 val_R@10=0.8953 val_Hit=95.9% alpha=0.9 (learned=0.992)

Epoch 22:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 22:  10%|█         | 1/10 [00:01<00:16,  1.87s/it]
Epoch 22:  30%|███       | 3/10 [00:02<00:05,  1.22it/s]
Epoch 22:  50%|█████     | 5/10 [00:03<00:03,  1.60it/s]
Epoch 22:  70%|███████   | 7/10 [00:04<00:01,  1.83it/s]
Epoch 22:  90%|█████████ | 9/10 [00:05<00:00,  2.17it/s]
                                                        
  Epoch 22: loss=0.8034 val_MRR=0.6421 val_R@10=0.8993 val_Hit=96.6% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.8993, saved model

Epoch 23:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 23:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 23:  30%|███       | 3/10 [00:02<00:04,  1.56it/s]
Epoch 23:  50%|█████     | 5/10 [00:02<00:02,  1.90it/s]
Epoch 23:  60%|██████    | 6/10 [00:03<00:01,  2.41it/s]
Epoch 23:  70%|███████   | 7/10 [00:03<00:01,  1.97it/s]
Epoch 23:  90%|█████████ | 9/10 [00:04<00:00,  2.43it/s]
                                                        
  Epoch 23: loss=0.8043 val_MRR=0.6806 val_R@10=0.9100 val_Hit=97.3% alpha=0.9 (learned=0.992)
  → New best val Recall@10=0.9100, saved model

Epoch 24:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 24:  10%|█         | 1/10 [00:01<00:13,  1.46s/it]
Epoch 24:  30%|███       | 3/10 [00:02<00:04,  1.41it/s]
Epoch 24:  50%|█████     | 5/10 [00:03<00:02,  1.77it/s]
Epoch 24:  70%|███████   | 7/10 [00:04<00:01,  1.98it/s]
Epoch 24:  90%|█████████ | 9/10 [00:04<00:00,  2.28it/s]
                                                        
  Epoch 24: loss=0.7746 val_MRR=0.6359 val_R@10=0.8984 val_Hit=95.9% alpha=0.9 (learned=0.996)

Epoch 25:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 25:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 25:  30%|███       | 3/10 [00:02<00:04,  1.55it/s]
Epoch 25:  50%|█████     | 5/10 [00:03<00:02,  1.83it/s]
Epoch 25:  70%|███████   | 7/10 [00:03<00:01,  2.04it/s]
Epoch 25:  90%|█████████ | 9/10 [00:04<00:00,  2.35it/s]
                                                        
  Epoch 25: loss=0.7572 val_MRR=0.6743 val_R@10=0.9063 val_Hit=96.6% alpha=0.9 (learned=0.996)

Epoch 26:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 26:  10%|█         | 1/10 [00:01<00:12,  1.39s/it]
Epoch 26:  30%|███       | 3/10 [00:02<00:04,  1.47it/s]
Epoch 26:  50%|█████     | 5/10 [00:03<00:02,  1.82it/s]
Epoch 26:  70%|███████   | 7/10 [00:03<00:01,  2.00it/s]
Epoch 26:  90%|█████████ | 9/10 [00:04<00:00,  2.37it/s]
                                                        
  Epoch 26: loss=0.7222 val_MRR=0.6650 val_R@10=0.9063 val_Hit=96.6% alpha=0.9 (learned=0.996)

Epoch 27:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 27:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 27:  30%|███       | 3/10 [00:02<00:04,  1.50it/s]
Epoch 27:  50%|█████     | 5/10 [00:03<00:02,  1.84it/s]
Epoch 27:  70%|███████   | 7/10 [00:03<00:01,  2.04it/s]
Epoch 27:  90%|█████████ | 9/10 [00:04<00:00,  2.40it/s]
                                                        
  Epoch 27: loss=0.7729 val_MRR=0.6785 val_R@10=0.9063 val_Hit=96.6% alpha=0.9 (learned=0.996)

Epoch 28:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 28:  10%|█         | 1/10 [00:01<00:12,  1.36s/it]
Epoch 28:  30%|███       | 3/10 [00:02<00:04,  1.52it/s]
Epoch 28:  50%|█████     | 5/10 [00:03<00:02,  1.82it/s]
Epoch 28:  70%|███████   | 7/10 [00:04<00:01,  1.94it/s]
Epoch 28:  90%|█████████ | 9/10 [00:04<00:00,  2.29it/s]
                                                        
  Epoch 28: loss=0.7600 val_MRR=0.6702 val_R@10=0.9080 val_Hit=96.6% alpha=0.9 (learned=0.996)

Epoch 29:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 29:  10%|█         | 1/10 [00:01<00:13,  1.49s/it]
Epoch 29:  30%|███       | 3/10 [00:02<00:05,  1.40it/s]
Epoch 29:  50%|█████     | 5/10 [00:03<00:03,  1.61it/s]
Epoch 29:  70%|███████   | 7/10 [00:04<00:01,  1.88it/s]
Epoch 29:  80%|████████  | 8/10 [00:04<00:00,  2.31it/s]
Epoch 29:  90%|█████████ | 9/10 [00:04<00:00,  2.26it/s]
                                                        
  Epoch 29: loss=0.7438 val_MRR=0.7059 val_R@10=0.9131 val_Hit=97.3% alpha=0.9 (learned=0.996)
  → New best val Recall@10=0.9131, saved model

Epoch 30:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 30:  10%|█         | 1/10 [00:01<00:12,  1.40s/it]
Epoch 30:  30%|███       | 3/10 [00:02<00:04,  1.48it/s]
Epoch 30:  50%|█████     | 5/10 [00:03<00:02,  1.82it/s]
Epoch 30:  70%|███████   | 7/10 [00:03<00:01,  2.02it/s]
Epoch 30:  90%|█████████ | 9/10 [00:04<00:00,  2.40it/s]
                                                        
  Epoch 30: loss=0.7554 val_MRR=0.7016 val_R@10=0.9193 val_Hit=97.3% alpha=0.9 (learned=0.996)
  → New best val Recall@10=0.9193, saved model

Epoch 31:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 31:  10%|█         | 1/10 [00:01<00:12,  1.44s/it]
Epoch 31:  30%|███       | 3/10 [00:02<00:04,  1.50it/s]
Epoch 31:  50%|█████     | 5/10 [00:03<00:02,  1.84it/s]
Epoch 31:  70%|███████   | 7/10 [00:03<00:01,  2.01it/s]
Epoch 31:  90%|█████████ | 9/10 [00:04<00:00,  2.34it/s]
                                                        
  Epoch 31: loss=0.6694 val_MRR=0.6931 val_R@10=0.9238 val_Hit=98.0% alpha=0.9 (learned=0.996)
  → New best val Recall@10=0.9238, saved model

Epoch 32:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 32:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 32:  30%|███       | 3/10 [00:02<00:04,  1.50it/s]
Epoch 32:  40%|████      | 4/10 [00:02<00:02,  2.10it/s]
Epoch 32:  50%|█████     | 5/10 [00:03<00:02,  1.74it/s]
Epoch 32:  60%|██████    | 6/10 [00:03<00:01,  2.26it/s]
Epoch 32:  70%|███████   | 7/10 [00:03<00:01,  1.91it/s]
Epoch 32:  80%|████████  | 8/10 [00:04<00:00,  2.53it/s]
Epoch 32:  90%|█████████ | 9/10 [00:04<00:00,  2.33it/s]
                                                        
  Epoch 32: loss=0.6898 val_MRR=0.7030 val_R@10=0.9328 val_Hit=98.6% alpha=0.9 (learned=0.996)
  → New best val Recall@10=0.9328, saved model

Epoch 33:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 33:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 33:  30%|███       | 3/10 [00:02<00:04,  1.49it/s]
Epoch 33:  50%|█████     | 5/10 [00:03<00:02,  1.83it/s]
Epoch 33:  70%|███████   | 7/10 [00:03<00:01,  2.01it/s]
Epoch 33:  90%|█████████ | 9/10 [00:04<00:00,  2.39it/s]
                                                        
  Epoch 33: loss=0.6686 val_MRR=0.7100 val_R@10=0.9243 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 34:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 34:  10%|█         | 1/10 [00:01<00:11,  1.33s/it]
Epoch 34:  30%|███       | 3/10 [00:02<00:04,  1.54it/s]
Epoch 34:  50%|█████     | 5/10 [00:03<00:02,  1.83it/s]
Epoch 34:  70%|███████   | 7/10 [00:03<00:01,  2.05it/s]
Epoch 34:  90%|█████████ | 9/10 [00:04<00:00,  2.41it/s]
                                                        
  Epoch 34: loss=0.6488 val_MRR=0.7016 val_R@10=0.9226 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 35:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 35:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 35:  20%|██        | 2/10 [00:01<00:05,  1.47it/s]
Epoch 35:  30%|███       | 3/10 [00:02<00:04,  1.57it/s]
Epoch 35:  40%|████      | 4/10 [00:02<00:02,  2.11it/s]
Epoch 35:  50%|█████     | 5/10 [00:02<00:02,  1.95it/s]
Epoch 35:  60%|██████    | 6/10 [00:03<00:01,  2.34it/s]
Epoch 35:  70%|███████   | 7/10 [00:03<00:01,  2.16it/s]
Epoch 35:  80%|████████  | 8/10 [00:04<00:00,  2.38it/s]
Epoch 35:  90%|█████████ | 9/10 [00:04<00:00,  2.80it/s]
Epoch 35: 100%|██████████| 10/10 [00:04<00:00,  2.88it/s]
                                                         
  Epoch 35: loss=0.6556 val_MRR=0.7078 val_R@10=0.9232 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 36:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 36:  10%|█         | 1/10 [00:01<00:12,  1.43s/it]
Epoch 36:  30%|███       | 3/10 [00:02<00:04,  1.57it/s]
Epoch 36:  40%|████      | 4/10 [00:02<00:02,  2.12it/s]
Epoch 36:  50%|█████     | 5/10 [00:02<00:02,  1.84it/s]
Epoch 36:  60%|██████    | 6/10 [00:03<00:01,  2.31it/s]
Epoch 36:  70%|███████   | 7/10 [00:03<00:01,  2.03it/s]
Epoch 36:  80%|████████  | 8/10 [00:03<00:00,  2.53it/s]
Epoch 36:  90%|█████████ | 9/10 [00:04<00:00,  2.50it/s]
Epoch 36: 100%|██████████| 10/10 [00:04<00:00,  3.00it/s]
                                                         
  Epoch 36: loss=0.6065 val_MRR=0.7038 val_R@10=0.9232 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 37:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 37:  10%|█         | 1/10 [00:01<00:12,  1.42s/it]
Epoch 37:  30%|███       | 3/10 [00:02<00:04,  1.53it/s]
Epoch 37:  50%|█████     | 5/10 [00:02<00:02,  1.93it/s]
Epoch 37:  70%|███████   | 7/10 [00:03<00:01,  2.16it/s]
Epoch 37:  80%|████████  | 8/10 [00:03<00:00,  2.63it/s]
Epoch 37:  90%|█████████ | 9/10 [00:04<00:00,  2.55it/s]
Epoch 37: 100%|██████████| 10/10 [00:04<00:00,  3.08it/s]
                                                         
  Epoch 37: loss=0.6210 val_MRR=0.7157 val_R@10=0.9266 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 38:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 38:  10%|█         | 1/10 [00:01<00:12,  1.39s/it]
Epoch 38:  30%|███       | 3/10 [00:02<00:04,  1.48it/s]
Epoch 38:  50%|█████     | 5/10 [00:03<00:02,  1.85it/s]
Epoch 38:  70%|███████   | 7/10 [00:03<00:01,  2.03it/s]
Epoch 38:  90%|█████████ | 9/10 [00:04<00:00,  2.42it/s]
                                                        
  Epoch 38: loss=0.6576 val_MRR=0.6914 val_R@10=0.9232 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 39:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 39:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 39:  30%|███       | 3/10 [00:02<00:04,  1.54it/s]
Epoch 39:  40%|████      | 4/10 [00:02<00:02,  2.14it/s]
Epoch 39:  50%|█████     | 5/10 [00:02<00:02,  1.85it/s]
Epoch 39:  60%|██████    | 6/10 [00:03<00:01,  2.32it/s]
Epoch 39:  70%|███████   | 7/10 [00:03<00:01,  1.95it/s]
Epoch 39:  80%|████████  | 8/10 [00:04<00:00,  2.50it/s]
Epoch 39:  90%|█████████ | 9/10 [00:04<00:00,  2.37it/s]
                                                        
  Epoch 39: loss=0.6259 val_MRR=0.7080 val_R@10=0.9266 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 40:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 40:  10%|█         | 1/10 [00:01<00:13,  1.45s/it]
Epoch 40:  30%|███       | 3/10 [00:02<00:04,  1.46it/s]
Epoch 40:  50%|█████     | 5/10 [00:03<00:02,  1.77it/s]
Epoch 40:  70%|███████   | 7/10 [00:04<00:01,  1.97it/s]
Epoch 40:  90%|█████████ | 9/10 [00:04<00:00,  2.34it/s]
                                                        
  Epoch 40: loss=0.5579 val_MRR=0.6967 val_R@10=0.9266 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 41:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 41:  10%|█         | 1/10 [00:01<00:12,  1.37s/it]
Epoch 41:  30%|███       | 3/10 [00:02<00:04,  1.50it/s]
Epoch 41:  50%|█████     | 5/10 [00:03<00:02,  1.90it/s]
Epoch 41:  70%|███████   | 7/10 [00:03<00:01,  2.05it/s]
Epoch 41:  90%|█████████ | 9/10 [00:04<00:00,  2.40it/s]
                                                        
  Epoch 41: loss=0.5765 val_MRR=0.7048 val_R@10=0.9255 val_Hit=98.0% alpha=0.9 (learned=0.996)

Epoch 42:   0%|          | 0/10 [00:00<?, ?it/s]
Epoch 42:  10%|█         | 1/10 [00:01<00:12,  1.34s/it]
Epoch 42:  30%|███       | 3/10 [00:02<00:04,  1.54it/s]
Epoch 42:  50%|█████     | 5/10 [00:02<00:02,  1.88it/s]
Epoch 42:  70%|███████   | 7/10 [00:03<00:01,  2.10it/s]
Epoch 42:  90%|█████████ | 9/10 [00:04<00:00,  2.49it/s]
                                                        
  Epoch 42: loss=0.6181 val_MRR=0.7021 val_R@10=0.9238 val_Hit=98.0% alpha=0.9 (learned=0.996)
  Early stopping at epoch 42 (no val improvement for 10 epochs)

============================================================
  Post-training final evaluation (on TEST)
============================================================
/workspace/ta-statute-law-retrieval/src/paragnn/trainer.py:280: FutureWarning: You are using `torch.load` with `weights_only=False` (the current default value), which uses the default pickle module implicitly. It is possible to construct malicious pickle data which will execute arbitrary code during unpickling (See https://github.com/pytorch/pytorch/blob/main/SECURITY.md#untrusted-models for more details). In a future release, the default value for `weights_only` will be flipped to `True`. This limits the functions that could be executed during unpickling. Arbitrary objects will no longer be allowed to be loaded via this mode unless they are explicitly allowlisted by the user via `torch.serialization.add_safe_globals`. We recommend you start setting `weights_only=True` for any use case where you don't have full control of the loaded file. Please open an issue on GitHub for any issues related to this experimental feature.
  model.load_state_dict(torch.load(f"{output_dir}/best_model.pt", map_location="cpu"))

  Alpha sweep on VAL (original):
  Alpha    R@10       MRR@10     Hit       
  --------------------------------------
  0.0      0.0154     0.0079     4.7%       <-
  0.1      0.0154     0.0080     4.7%      
  0.2      0.0171     0.0096     5.4%       <-
  0.3      0.0329     0.0131     7.4%       <-
  0.4      0.0396     0.0165     8.1%       <-
  0.5      0.0577     0.0209     10.1%      <-
  0.6      0.0892     0.0371     14.9%      <-
  0.7      0.1950     0.0821     31.8%      <-
  0.8      0.5982     0.3631     78.4%      <-
  0.9      0.9328     0.7030     98.6%      <-
  1.0      0.8380     0.5146     92.6%     

  Alpha sweep on VAL (debiased):
  Alpha    R@10       MRR@10     Hit       
  --------------------------------------
  0.0      0.0154     0.0079     4.7%       <-
  0.1      0.0154     0.0077     4.7%      
  0.2      0.0154     0.0078     4.7%      
  0.3      0.0154     0.0089     4.7%      
  0.4      0.0222     0.0094     5.4%       <-
  0.5      0.0329     0.0120     7.4%       <-
  0.6      0.0396     0.0153     8.1%       <-
  0.7      0.0554     0.0197     10.1%      <-
  0.8      0.0802     0.0378     12.2%      <-
  0.9      0.2180     0.1221     35.1%      <-
  1.0      0.4734     0.3315     62.2%      <-

  Grid Search (alpha=0.9, original, from val):
    Recall@10: 0.6664  MRR@10: 0.5514  Hit: 78.4%

Training complete. Test Recall@10: 0.6664
  Predictions saved: outputs/predictions/structgnn_nognn_statusaware_structdense_kuhperdata-summ-exp.jsonl (296 queries, top-100)

  Final best Recall@10: 0.6664