The repository is under construction ... you're inveted to try stuff but strange things may happen ;)
In this repository I gather my extensions to PyTorch. The packaging structure tries to reproduce PyTorch's structure in order to facilitate usage for people familiar with PyTorch. In the following a list of its main features (admitably short ... yet ;) ).
nn.SLayer: this is an input layer which can operate on multi sets of points of some cartesian product of the real numbers. Its primary intention is to train networks on the output of a topological data analysis pipeline, but can be used on arbitrary (real vector) multi set input.. See Deep Learning with Topological Signatures for further reading.
The folder tutorials contains minimalistic examples in form of jupyter notebooks to improve usability.