the problem about PPI dataset

[MrLinNing]

❓ Questions & Help

When I use this APIs train_dataset = PPI(path, split='train') .
Because of the internet reason, I can not download the ppi.zip from https://s3.us-east-2.amazonaws.com/dgl.ai/dataset/ppi.zip.
So, If I already have copy ppi.zip from other peoples to my local folds, how to directly use this ppi.zip rather than downloading it again from the https://s3.us-east-2.amazonaws.com/dgl.ai/dataset/ppi.zip ?

[rusty1s]

Unpack it into the root folder and name it raw. The dataset will not get downloaded again if it finds the files such as test_feats.npy inside root/raw.

[im-wll]

I use the sentence “ test_dataset = GeoData.PPI('/home/work/wll/GCN/demo/deep_gcns_torch-master/data/deepgcn/ppi/data-wll', split='test')”，but get an error, I don't know how to solve it?
the raw data has downloaded,but when it comes to processing, the error appears

[rusty1s]

Mh, works like a charm for me. Which networkx version are you using?

[im-wll]

thank you for your reply. I have installed torch 1.2.0 , torch-geometric and a series of package，but my cuda is 8.0, so the torch.cuda.available( ) is False. Now, I only call the GeoData.PPI class without using network, but the error still appear. I don't kmow Whether the environment is satisfied.

[rusty1s]

No, I mean the version of the networkx package. It recently changed its API.

[LiaoLW]

❓ Questions & Help

When I use this APIs train_dataset = PPI(path, split='train') .
Because of the internet reason, I can not download the ppi.zip from https://s3.us-east-2.amazonaws.com/dgl.ai/dataset/ppi.zip.
So, If I already have copy ppi.zip from other peoples to my local folds, how to directly use this ppi.zip rather than downloading it again from the https://s3.us-east-2.amazonaws.com/dgl.ai/dataset/ppi.zip ?

Excuse me, can you share this dataset with me? The one I downloaded from http://snap.stanford.edu/graphsage/ppi.zip is not the same as the one used in PyG.

[rusty1s]

This is the same dataset, but it is already processed to speed up the dataset creation part. We load the pre-processed version provided by DGL, so I do not have access to the pre-processing part.

[LiaoLW]

This is the same dataset, but it is already processed to speed up the dataset creation part. We load the pre-processed version provided by DGL, so I do not have access to the pre-processing part.

But there is not any 'train_graph.json' (and others required in the script) in this file, all I have is

[rusty1s]

The training split can be found in ppi-G.json:

"nodes": [{"test": false, "id": 0, "val": false}, {"test": false, "id": 1, "val": false}, ...