Deepxde

Areas of improvement

- Bug fix: `Model.predict()` works for DeepONet

Areas of improvement

- TensorFlow 2.x and PyTorch support loss_weights
- Improve L-BFGS for TensorFLow 2.x and PyTorch
- `Geometry.random_boundary_points()` doesn't sample corner points

New APIs

- Add `dde.optimizers.set_LBFGS_options()`

Areas of improvement

- Backend TensorFlow 2.x supports L-BFGS via TFP
- Backend PyTorch supports L-BFGS
- Backend PyTorch uses GPU by default, if available
- Improve BC/IC performance for backend PyTorch

New APIs

- Add `dde.Variable` for inverse problems

- Backend PyTorch supports PDE inverse problems

Areas of improvement

- Refactor some parts of `Model`
- Bug fix for backend TensorFlow 2.x and PyTorch

API changes

- Rename `dde.maps` to `dde.nn` to be more explicit

- Backend PyTorch supports PDE forward problems

API changes

- Refactor uncertainty via MC dropout as the callback `DropoutUncertainty`; remove "uncertainty" argument from `Model.train()`.

DeepXDE supports PyTorch backend for function approximation.

Areas of improvement

- Backend TensorFlow supports auxiliary variables

New APIs

- Add `dde.config.set_default_float()` and `dde.config.default_float()`