Error on Tensor Sizes during Model Training

[radam1]

Hello, I am trying to run exercise 2, but when I try to train my model, I run into an error at the end of the first epoch as it appears there is one data point that is a different size than the rest. When I print out the shape of x in the train_epoch() function, for every value until the last one I get:

x shape: torch.Size([32, 4, 224, 224])

But the final data point is:

x shape: torch.Size([17, 4, 224, 224])

Which creates the error:

ValueError Traceback (most recent call last)
Cell In[61], line 12
4 writer = SummaryWriter(f"runs/{exp_name}", comment="")
6 for epoch in tqdm(range(cfg.epochs), desc="Epochs"):
7 (
8 train_loss,
9 (train_error_mean, train_error_median, train_error_tensor),
10 (val_err_mean, val_err_median, val_error_tensor),
11 val_pred,
---> 12 ) = train_epoch(train_loader, val_loader, model, optimizer, criterion)
13 writer.add_scalar("Loss/train", train_loss, epoch)
14 writer.add_scalar("Error/train/mean", train_error_mean, epoch)

Cell In[59], line 35
31 loss.backward()
33 optimizer.step()
---> 35 train_loss, train_performance, _ = evaluate(train_loader, model, criterion)
36 _, val_performance, val_pred = evaluate(val_loader, model, criterion)
38 return train_loss, train_performance, val_performance, val_pred

Cell In[52], line 34
31 angle_error = torch.min(angle_error, 360 - angle_error)
32 metric.extend(angle_error.tolist())
---> 34 metric = torch.tensor(metric)
35 pred = torch.tensor(pred)
37 return sum(loss) / len(loss), (metric.mean(), metric.median(), metric), pred

ValueError: expected sequence of length 32 at dim 1 (got 17)

I am not sure if it has something to do with an error in the data or there is something wrong with my method of handling it, but I would appreciate any guidance about what I can do to fix it. Thanks!

[yilun-wu]

Can you let me know the config of this experiment? Which parameters have you changed in the config files under ex_2/config?

[radam1]

I haven’t changed any of the configurations other than the learning rate and fixing a couple of file path errors that happened the first time I tried to run the hydra code. (taking the ex2 out of _target_: ex_2.dataset.PolImgDataset for example).

…

On Sep 16, 2024, at 12:07 PM, Yilun Wu ***@***.***> wrote: Can you let me know the config of this experiment? Which parameters have you changed in the config files under ex_2/config? — Reply to this email directly, view it on GitHub <https://urldefense.com/v3/__https://github.com/tudelft/AE4353-Y24/discussions/20*discussioncomment-10657731__;Iw!!LIr3w8kk_Xxm!tb2t_nW8DJ8SRlkvCttqPzpUejzmzeBFPnz7rYOEl8YeMMp1svP761jEffEuw67LLAmPf8kVbYRLmde995IMgw$>, or unsubscribe <https://urldefense.com/v3/__https://github.com/notifications/unsubscribe-auth/AY4LOUZP55F4OQTYYQMGSWDZW2UUPAVCNFSM6AAAAABOI5LM42VHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTANRVG43TGMI__;!!LIr3w8kk_Xxm!tb2t_nW8DJ8SRlkvCttqPzpUejzmzeBFPnz7rYOEl8YeMMp1svP761jEffEuw67LLAmPf8kVbYRLmdfNQ-JArw$>. You are receiving this because you authored the thread.

[yilun-wu]

Yup. I could reproduce the problem. Fixing now...

[yilun-wu]

This has been fixed in 65388fc.