This package implements a few polynomial basis types, and convenience methods for evaluation and derivatives, fast batched evaluation, for building small and fast ML type models. Layers currently implemented include:
- Various orthogonal polynomials via 3-point recursion
- Trigonometric polynomials
- Complex and real spherical and solid harmonics
- A few quantum chemistry (atomic orbitals) basis sets
- Utilities to recombine them into (tensor) product or compressed basis sets
We also aim to provide full Lux.jl integration. A possible application of this might be to implement various flavours of equivariant neural networks and related models.
