What does sample_adj do?

[boxiangliu]

Dear Matthias,

Thank you for making pytorch-geometric. It has been a extremely useful tool. I am trying to use GraphSage in unsupervised training mode. To do that, I am following your graph_sage_unsup.py. As I am trying to understand the details of the NeighborSampler, I am lost on what self.adj_t.sample_adj(n_id, size, replace=False) (in sampler.py; line 152) is doing.

Could you help me understand? Thank you in advance!

Boxiang Liu, Ph.D.

[rusty1s]

sample_adj will sample size neighbors for each node in n_id. It's output should correspond to a sparse tensor of shape [n_id.size(0), xxx], where xxx is determined by the number of unique neighbors sampled.

[eedalong]

@boxiangliu you can find code about sample_adj in pytroch_sparse repo