pytorch_sparse icon indicating copy to clipboard operation
pytorch_sparse copied to clipboard

New feature: weighted neighborhood sampling (sample by value)

Open deepakagrawal opened this issue 4 years ago • 3 comments

I have a SparseTensor with value, how can extend sample method or some other function where the neighborhood sampling probability is based on the value of the SparseTensor.

deepakagrawal avatar Jul 11 '21 21:07 deepakagrawal

This is sadly not supported yet. I think this needs its own condition here, where we want to perform weighted neighborhood sampling rather than uniform sampling.

rusty1s avatar Jul 12 '21 06:07 rusty1s

This is sadly not supported yet. I think this needs its own condition here, where we want to perform weighted neighborhood sampling rather than uniform sampling.

How can I create a feature request for this?

deepakagrawal avatar Jul 12 '21 07:07 deepakagrawal

Feel free to send a PR with the applied changes. Other-wise, I will find some time to implement it by myself.

rusty1s avatar Sep 16 '21 06:09 rusty1s