Got the RuntimeError. Would anyone help me?

[huangteng1220]

(cv2.7) gpuserver@ubuntu:~/ht/labs/temporal-ensembling$ python mnist_eval.py
Traceback (most recent call last):
File "mnist_eval.py", line 50, in
acc, acc_best, l, sl, usl, indices = train(model, seed, **cfg)
File "/home/gpuserver/ht/labs/temporal-ensembling/temporal_ensembling.py", line 86, in train
k, n_classes, seed, shuffle_train=False)
File "/home/gpuserver/ht/labs/temporal-ensembling/temporal_ensembling.py", line 25, in sample_train
indices[i * card: (i + 1) * card] = class_items[rd[:card]]
RuntimeError: expand(torch.LongTensor{[10, 1]}, size=[10]): the number of sizes provided (1) must be greater or equal to the number of dimensions in the tensor (2)

[sweetdream33]

I get the same error.

Traceback (most recent call last):
File "mnist_eval.py", line 50, in
acc, acc_best, l, sl, usl, indices = train(model, seed, **cfg)
File "/storage/ssd/jupyter-nb-workspace/temporal-ensembling-master/temporal_ensembling.py", line 86, in train
k, n_classes, seed, shuffle_train=False)
File "/storage/ssd/jupyter-nb-workspace/temporal-ensembling-master/temporal_ensembling.py", line 25, in sample_train
indices[i * card: (i + 1) * card] = class_items[rd[:card]]
RuntimeError: expand(torch.LongTensor{[10, 1]}, size=[10]): the number of sizes provided (1) must be greater or equal to the number of dimensions in the tensor (2)

[ferretj]

Hey, sorry about the delay.

The error is a dimension mismatch and is due to breaking changes in PyTorch 0.4.0 and later. I used PyTorch 0.3.0.post4 version with torchvision version 0.2.0 to run my experiments.

I recommend to use the same version as I did, and will update the README so that users will know which version to use.

If you want to stick with the latest PyTorch version, in my case a very simple addition in temporal_ensembling.py did the job (line 22) :
class_items = (train_dataset.train_labels == i).nonzero()[:, 0]

Hope this helps, I will close the issue for now.