An Implementation of Set Transformer that can work with variable sized sets. Out of the box, this function takes in sparse inputs in the format provided by Pytorch Geometric.
The code requires the following libraries:
- Pytorch
- Pytorch Geometric
Import the modules from modules.py and run.
If you have your data in the form x and ptr (or sizes), you can convert it to the form required by the module as
>>> sizes = torch.tensor([3, 32, 18, 13, 90, 29])
>>> batch = torch.arange(sizes.size(0)).repeat_interleave(sizes)