How Does SparseConvNet Treat Active Sites Outside of the Stated Spatial Size of the Sparse Tensor?

[andrewcking]

I ran a small test with an input layer that had an active site outside of the stated spatial size of the sparse tensor. The input layer didn't complain and the network still provided an output for that value and did not omit it (which is good for my use case). Do I need to be ensuring that this does not occur or is this behavior occasionally okay to rely on? What implications does this have under the hood? I suppose the the question may be what is spatial_size used for internally?