What is the difference of batch size and the number of seed nodes?

[Gibyeng]

What is the relationship of batch size and the number of seed nodes in NeighborSampler?
To my understanding, the batch size is equal to the number of seed nodes, which means the first nodes in the batch.
However I found the two variables "seed.size(0)" (in .neighbor_kernel.cpp) and "batch_size" (in .py file) have different values.

[rusty1s]

seed refers to the input nodes you want to process in one epoch, and batch_size is determines how to divide these input nodes into different chunks.

[Gibyeng]

@rusty1s Could you explain it with an example, say if seed.size(0) is 207 and batch_size is 512. Does it mean we need two epoches to collect 512 samples in total to form a batch?

[rusty1s]

The last batch doesn't need to have full capacity. If you have seed.size() == 207 and batch_size = 512, then all 207 seed nodes will be handled within a single batch.

[Gibyeng]

@rusty1s Thanks a lot, I got it :)