Paradime

Changes

This version aligns the ParaDime codebase with the state described in the revised paper preprint.
This includes mostly changes to the specifications and their parsing:

- introduced registration functions
- moved parsing functionality to main module
- minor fixes in the docs

Changes

- Restructured ParaDime routines:
- Losses are now added as a dict and referenced in training phases.
- Data can be registered earlier, but that is no longer required. Instead data is added in the `train` call.
- Added specification parser (`ParametricDR.from_spec`).
- Added new examples and improved docs.

Changes

- Removed PyTorch from the dependencies. It must now be installed separately.
- Updated docs to reflect installation changes.
- Improved loss tracking for compound losses.
- Scatterplot now accepts an axes object.
- Added various examples to the docs.

Changes

- Dependencies are now installed automatically.
- Fixed a bug with the color pallette that broke the `scatterplot` function.

Changes

- Changed syntax for `data_key` in relations, to allow computation of global relations for different attributes of the data set.
- Added parameters for model methods to be used by losses.
- Various bugfixes.
- Improved docs and switched to `nbsphinx` for examples.

Changes

- Many bugfixes
- Refactored utils
- Switched to logging module
- Added first example script
- Added most important info to docs