Closed HunterLep closed 1 year ago
Hi! Thanks for the comment.
In line 183, batch = np.random.permutation(self.class_dict2[c])[:num]
; it takes first num
samples from the c-th class. I feel it should not throw such an error and I am interested in what command you used.
Hi!
Thanks for your reply.
I used the command in your readme
python train_gcond_transduct.py --dataset cora --nlayers=2 --lr_feat=1e-4 --gpu_id=0 --lr_adj=1e-4 --r=0.5
Here's the full error message,just for your information;
Traceback (most recent call last):
File "train_gcond_transduct.py", line 57, in <module>
agent.train()
File "F:\GCond-main\gcond_agent_transduct.py", line 198, in train
c, adj, transductive=True, args=args)
File "F:\GCond-main\utils.py", line 185, in retrieve_class_sampler
out = self.samplers[c].sample(batch)
File "D:\Anaconda\envs\gcond\lib\site-packages\torch_geometric\data\sampler.py", line 151, in sample
adj_t, n_id = self.adj_t.sample_adj(n_id, size, replace=False)
File "D:\Anaconda\envs\gcond\lib\site-packages\torch_sparse\sample.py", line 31, in sample_adj
rowptr, col, subset, num_neighbors, replace)
RuntimeError: expected scalar type Long but found Int
I made a change in line 183 with "batch" just to see if it would make any difference and it did.
I just tried running the command and it worked fine to me. I guess it is due to the version of PyG (mine is for linux). But I'm glad that you figured out a solution.
That makes sense.I'm using the windows version of PyG.Anyway thanks for your time.
I've been trying reproducing the performance of your paper and this runtime error keeps popping up.After a brief look,I think Line 183 in utils.py should turn "batch" into a LongTensor type.I managed to fix this problem and reproduced the performance.I wonder if you could check if I'm right so that I can move on to other issues.