Open jiaruHithub opened 4 years ago
i have same problem, i want to reproduce five different networks model for point cloud on ModelNet10 dataset. I got stuck there for 3 days.
You need to run main.py. If you only want to train a specific implementation/dataset, you can comment out all other models/datasets in nets
and datasets
. For a better logging experience, you can also pass logger
to the cross_validation_with_val_set
method.
You need to run main.py. If you only want to train a specific implementation/dataset, you can comment out all other models/datasets in
nets
anddatasets
. For a better logging experience, you can also passlogger
to thecross_validation_with_val_set
method.
Fine, thanks. Em...but I can't find this file,how can I call this file 'main.py'? Please Forgive me for being slow... :)
You just call it via python, see here.
Thank you for your patient answer! the file eported an error:
Traceback (most recent call last):
File "main.py", line 76, in <module>
logger=None,
File "\pytorch_geometric\benchmark\kernel\train_eval.py", line 19, in cross_validation_with_val_set
val_idx) in enumerate(zip(*k_fold(dataset, folds))):
File "\pytorch_geometric\benchmark\kernel\train_eval.py", line 93, in k_fold
train_mask[test_indices[i]] = 0
IndexError: tensors used as indices must be long, byte or bool tensors
How can I handle this error? Thank you!
Seems to be Windows/numpy related. You can fix it by casting indices to long
explicitly:
test_indices.append(torch.from_numpy(idx).to(torch.long))
test_indices.append(torch.from_numpy(idx).to(torch.long))
It's working!Thank you! But I have a question that instability of acc and experiment acc can't attend the paper claimed. Is the question relate the seed or Environment configuration?
❓ Questions & Help
Hello,I try to run the benchmarks for graph classification.I don't know how to use the benchmarks. For example, I want to reproduce the SAGPool in D&D dataset like #859. Forgive me for being slow:).How do I operate?