Pytorch DataLoader doesn't work with remote interpreter

user_12809299 0
Viewed 986

I have the following error. 

Expected: /home/ubuntu/.pycharm_helpers/pydev/pydevd_attach_to_process/attach_linux_amd64.so to exist.

And Here is the code:

import torch_geometric.transforms as T
category = 'Airplane'
path = osp.join(osp.dirname(osp.realpath(__file__)), '..', 'data', 'ShapeNet')
transform = T.Compose([
    T.RandomTranslate(0.01),
    T.RandomRotate(15, axis=0),
    T.RandomRotate(15, axis=1),
    T.RandomRotate(15, axis=2)
])
pre_transform = T.NormalizeScale()
train_dataset = ShapeNet(path, category, split='trainval', transform=transform,
                         pre_transform=pre_transform)
test_dataset = ShapeNet(path, category, split='test',
                        pre_transform=pre_transform)
train_loader = DataLoader(train_dataset, batch_size=12, shuffle=True, num_workers=6)
test_loader = DataLoader(test_dataset, batch_size=12, shuffle=False,
                         num_workers=6)

When I'm trying sample from dataset using the dataloader, the debugger crashes and returns this erros.

Have tried deleting remote helpers, but it didn't solve my problem. My local machine in running on windows 10 and the remote one is running on Ubuntu 18.04

pytorch deep-learning helper remote-debugging python
1 Answers

Answering to myself here,

I've tried to follow this : https://intellij-support.jetbrains.com/hc/en-us/community/posts/360006791159--pycharm-helpers-pydev-pydevd-attach-to-process-attach-linux-amd64-so-undefined-symbol-AttachDebuggerTracing

But it has proven inefficient.

On the contrary, deleting the parameter 'num_workers' from the Dataloader solved this issue. So this is an easy workaround.

However, I haven't wrap my head around the problem yet. It's probably that multiprocessing is buggy. But if anyone has any insight of what's going on. I'll gladly take it.

user_128092990
Related