Closed xiaoliqiangxxqzz closed 4 years ago
yaohungt
commented
4 years ago I don't remember I have faced this issue. It seem like the problem happens in dataset.py.
What is the version of PyTorch you're using? Also, can you check if the loading of X/Y/META are correct in lines 51-53 in
https://github.com/yaohungt/Multimodal-Transformer/blob/master/src/dataset.py
xiaoliqiangxxqzz
commented
4 years ago 我不记得我遇到过这个问题。看来问题出在dataset.py中。
您使用的PyTorch是什么版本？另外，您可以检查X / Y / META的加载在第51-53行中是否正确
https://github.com/yaohungt/Multimodal-Transformer/blob/master/src/dataset.py
The version of PyTorch is 1.2.0. And when i use mosei_senti_data_noalign.pkl, error occurred:"_pickle.UnpicklingError: pickle data was truncated".So,i use mosi_data_noalign.pkl,but that's what happened.
yaohungt
commented
4 years ago Do you like to re-download the data from the dropbox? I never faced the issues you have, but I will examine it more tomorrow.
xiaoliqiangxxqzz
commented
4 years ago Do you like to re-download the data from the dropbox? I never faced the issues you have, but I will examine it more tomorrow.
Ok,I will re-download the data and try again
yaohungt
commented
4 years ago 我不记得我遇到过这个问题。看来问题出在dataset.py中。 您使用的PyTorch是什么版本？另外，您可以检查X / Y / META的加载在第51-53行中是否正确 https://github.com/yaohungt/Multimodal-Transformer/blob/master/src/dataset.py
The version of PyTorch is 1.2.0. And when i use mosei_senti_data_noalign.pkl, error occurred:"_pickle.UnpicklingError: pickle data was truncated".So,i use mosi_data_noalign.pkl,but that's what happened.
By the way, it surprises me that GitHub has an automatic translation function....
yaohungt
commented
4 years ago I guess maybe you do not set the datapath correctly? If this is the case, it will have your above issues.
xiaoliqiangxxqzz
commented
4 years ago 我不记得我遇到过这个问题。看来问题出在dataset.py中。 您使用的PyTorch是什么版本？另外，您可以检查X / Y / META的加载在第51-53行中是否正确 https://github.com/yaohungt/Multimodal-Transformer/blob/master/src/dataset.py
The version of PyTorch is 1.2.0. And when i use mosei_senti_data_noalign.pkl, error occurred:"_pickle.UnpicklingError: pickle data was truncated".So,i use mosi_data_noalign.pkl,but that's what happened.
By the way, it surprises me that GitHub has an automatic translation function....
hahaha,chrome has that function not Github
xiaoliqiangxxqzz
commented
4 years ago I guess maybe you do not set the datapath correctly? If this is the case, it will have your above issues.
yes, it works now.But it seems like there is some problems with mosei_senti_data_noalign.pkl.
Start loading the data....
Note: You are running in unaligned mode.
Traceback (most recent call last): File "main.py", line 151, in
test_loss = train.initiate(hyp_params, train_loader, valid_loader, test_loader)
File "/home/xlq/Multimodal-Transformer-master/src/train.py", line 63, in initiate
return train_model(settings, hyp_params, train_loader, valid_loader, test_loader)
File "/home/xlq/Multimodal-Transformer-master/src/train.py", line 236, in train_model
train(model, optimizer, criterion, ctc_a2l_module, ctc_v2l_module, ctc_a2l_optimizer, ctc_v2l_optimizer, ctc_criterion)
File "/home/xlq/Multimodal-Transformer-master/src/train.py", line 92, in train
for i_batch, (batch_X, batch_Y, batch_META) in enumerate(train_loader):
File "/home/xlq/anaconda3/envs/pyth/lib/python3.7/site-packages/torch/utils/data/dataloader.py", line 560, in next
batch = self.collate_fn([self.dataset[i] for i in indices])
File "/home/xlq/anaconda3/envs/pyth/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 68, in default_collate
return [default_collate(samples) for samples in transposed]
File "/home/xlq/anaconda3/envs/pyth/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 68, in
return [default_collate(samples) for samples in transposed]
File "/home/xlq/anaconda3/envs/pyth/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 68, in default_collate
return [default_collate(samples) for samples in transposed]
File "/home/xlq/anaconda3/envs/pyth/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 68, in
return [default_collate(samples) for samples in transposed]
File "/home/xlq/anaconda3/envs/pyth/lib/python3.7/site-packages/torch/utils/data/_utils/collate.py", line 55, in default_collate
return numpy_type_map[elem.dtype.name](list(map(py_type, batch)))
KeyError: 'bytes104'