My final acc is only 76%

[mobulan]

Eval Epoch 95: 100% 91/91 [01:49<00:00,  1.21s/it, Batch [91/91], Acc: 0.559]
Train Epoch 96: 100% 469/469 [05:52<00:00,  1.33it/s, Batch [469/469], Loss: 0.3
Eval Epoch 96: 100% 91/91 [01:51<00:00,  1.22s/it, Batch [91/91], Acc: 0.618]
Train Epoch 97: 100% 469/469 [05:55<00:00,  1.32it/s, Batch [469/469], Loss: 0.2
Eval Epoch 97: 100% 91/91 [01:47<00:00,  1.18s/it, Batch [91/91], Acc: 0.529]
Train Epoch 98: 100% 469/469 [05:52<00:00,  1.33it/s, Batch [469/469], Loss: 0.3
Eval Epoch 98: 100% 91/91 [01:48<00:00,  1.19s/it, Batch [91/91], Acc: 0.588]
Train Epoch 99: 100% 469/469 [05:56<00:00,  1.32it/s, Batch [469/469], Loss: 0.4
Eval Epoch 99: 100% 91/91 [01:46<00:00,  1.17s/it, Batch [91/91], Acc: 0.618]
Train Epoch 100: 100% 469/469 [05:53<00:00,  1.33it/s, Batch [469/469], Loss: 0.
Eval Epoch 100: 100% 91/91 [01:47<00:00,  1.18s/it, Batch [91/91], Acc: 0.647]
Eval Epoch pruned: 100% 91/91 [01:39<00:00,  1.10s/it, Batch [91/91], Acc: 0.647
Projection: 100% 375/375 [02:51<00:00,  2.19it/s, Batch: 375/375]
Eval Epoch pruned_and_projected: 100% 91/91 [01:36<00:00,  1.06s/it, Batch [91/9
Eval Epoch pruned_and_projected: 100% 91/91 [01:33<00:00,  1.03s/it, Batch [91/9
Eval Epoch pruned_and_projected: 100% 91/91 [02:16<00:00,  1.50s/it, Batch [91/9
Fidelity: 100% 91/91 [02:47<00:00,  1.84s/it, Batch [91/91]]
```bash
In the overview table
| 85   | 0.764411 | 0.927753346 | 0.315042642 |
| ---- | -------- | ----------- | ----------- |
| 86   | 0.766483 | 0.929271055 | 0.312246734 |
| 87   | 0.768208 | 0.929870736 | 0.310522149 |
| 88   | 0.77028  | 0.928586235 | 0.310088791 |
| 89   | 0.76631  | 0.929519071 | 0.306573199 |
| 90   | 0.763721 | 0.929437633 | 0.307344964 |
| 91   | 0.764066 | 0.929333985 | 0.301229446 |
| 92   | 0.765447 | 0.929681947 | 0.299217476 |
| 93   | 0.765274 | 0.930585169 | 0.298738088 |
| 94   | 0.768381 | 0.929467247 | 0.298132296 |
| 95   | 0.765965 | 0.929104478 | 0.296845926 |
| 96   | 0.76631  | 0.929918858 | 0.295910276 |
| 97   | 0.764066 | 0.930070629 | 0.295259658 |
| 98   | 0.763548 | 0.92906746  | 0.296624792 |
| 99   | 0.764411 | 0.929082267 | 0.296345691 |
| 100  | 0.761823 | 0.929915156 | 0.294409203 |

[M-Nauta]

Dear Mobulan, On which dataset did you train and did you use the same parameters as specified in the paper? It might also be the case that you had a bad random seed or unfortunate weight initialization, which you could verify by training it again.

[mobulan]

yes, i will run it again today.

python main_tree.py --epochs 100 --log_dir /DATA/linjing/Mobulan/ProtoTree/runs/protoree_cub_11.27 --dataset CUB-200-2011 --lr 0.001 --lr_block 0.001 --lr_net 1e-5 --num_features 256 --depth 9 --net resnet50_inat --freeze_epochs 30 --milestones 60,70,80,90,100

i change the num_wokers to 4 by editing

    trainloader = torch.utils.data.DataLoader(trainset,
                                              batch_size=args.batch_size,
                                              shuffle=True,
                                              pin_memory=cuda,
                                              num_workers = 4
                                              )
    projectloader = torch.utils.data.DataLoader(projectset,
                                            #    batch_size=args.batch_size,
                                              batch_size=int(args.batch_size/4), #make batch size smaller to prevent out of memory errors during projection
                                              shuffle=False,
                                              pin_memory=cuda,
                                              num_workers = 4
                                              )
    testloader = torch.utils.data.DataLoader(testset,
                                             batch_size=args.batch_size,
                                             shuffle=False,
                                             pin_memory=cuda,
                                             num_workers = 4
                                             )