graph2mat.bindings

Specific interfaces to other codes.

The core functionality of graph2mat is agnostic to the framework, and it is based on pure python and numpy.

For running a specific ML workflow, we need interfaces of the core functionality with ML frameworks, e.g. torch, e3nn. The implementation of the interfaces are mostly thin wrappers around the core functionality, as well as functions that only make sense on the specific framework. For example, the e3nn bindings contains functions that use irreps.

Whatever framework that we interface with, it should not be a required import for the core functionality. So basically, the criteria for creating a new submodule in bindings is that we can’t add the functionality to the core without requiring the framework as a dependency.

Modules

e3nn	Interface to e3nn, as well as functions that use irreps.
torch	Interface with pytorch.