- adapted to the newest breaking changes from [`e3nn 0.19.0`](https://github.com/e3nn/e3nn-jax/releases/tag/0.19.0) - added `O3TensorProductSCN` using `e3nn.experimental.linear_shtp.shtp` and a linear layer (not fully integrated) - removed `pooling_irreps` from segnn (_graph_ task case) - changed the O3 layer structure - fixed the validation script (mistake in the loss wrapper function)
0.6
- Updated to jax 0.4.8 and e3nn-jax 0.17.4 - Simplified tensor product interface - Improved QM9 validation results
0.5
- Updated tensor product, ~5-10% faster in training ([e3nn-jax 0.14.0](https://github.com/e3nn/e3nn-jax/tree/0.14.0)) - Fixed a bug on __normalization parametrization__ - Module refactor - Updated results
0.3
- `SEGNN` as an `hk.Module` class to allow easier extensions - updated readme