Repository navigation
meent 0.13.2
Pre-release
Pre-release
Highlights
PyTorch solver inputs now preserve their autograd graphs throughout setup. This fixes theta, phi, psi, period, and thickness parameters silently receiving no gradient during joint optimization, while retaining the existing behavior for ordinary scalar, sequence, NumPy, and tensor inputs.
This release also adds anisotropic and evanescent-input solver support, expands the bundled refractive-index material tables, aligns reflection/transmission conventions, fixes sign(0), and exports fields in RETICOLO's basis.
What's Changed
- Anisotropic and evanescent-input solver support, plus nk material tables by @Taesang-Yun in #103
- Put R and T on one convention, fix
sign(0), and export fields in RETICOLO's basis by @Taesang-Yun in #106 - Keep the autograd graph on every differentiable solver input by @Taesang-Yun in #107
New Contributors
- @Taesang-Yun made their first contribution in #103
Full Changelog: v0.12.0...v0.13.2