Updated to use `torch 2.x` (+ latest versions of other dependencies). This is *not* a breaking change, DLKoopman will continue to work as before if the machine running it has `torch>=1.12.1`.
Added a new configuration parameter `torch_compile_backend` to specify the backend for the `torch.compile()` call if using `torch 2.x`.
Fixed the device for some tensors.
1.1.3
Updated "How to cite" section in README to reflect publication of the accompanying paper in L4DC 2023.
No changes to the source code.
1.1.0
Stable release, major changes are: - Configuration is now a class that can be instantiated to create new configs. - Added isort workflow. Run `isort .` in the repo root before pushing to pass the isort workflow. - Updates to README and documentation.
1.0.2
Stable release containing: - Modules for state predictor, trajectory predictor, and hyperparameter search. - Examples (tutorials) and tests for all the above. - Poetry for packaging, plus Docker support.
0.1
Version `0.1` – Initial release.
Planned upcoming changes: - Support for multiple trajectories - Approximating the Koopman matrix via a linear layer - More loss functions - Supporting inputs and control