Model architecture and training rewritten to use [pytorch-lightning](https://pytorchlightning.ai/).
Specifically this includes:
- `pytorch.Dataset` (`convml_tt.data.TripletDataset`) for handling loading of individual triplet datasets (and one for single-tile datasets). This has the option to load all data into a single numpy memory-mapped array to reduce number of disc reads
- `pytorch_lightning.LightningDataModule` (`convml_tt.data.TripletDataModule`) for handling transform, batching and splitting data for train/test
- includes transforms previously provided by fastai now using the
[kornia](https://kornia.github.io/) library
- correct normalisation when using pretrained network is used
- `pytorch_lightning.LightningModule` (`convml_tt.system.TripletTrainerModel`) which handles the training logic: building model, setting learning rate, batch size. This is now flexible in the number of embedding dimensions used
- Command-line interface for doing training (`convml_tt.trainer`)
- functionality for downloading and unpacking example datasets
- refactoring of utility functions and inference plots to ensure they work with
the new architecture
- ARC3 HPC and JASMIN HPC systems submission scripts (SGE and SLURM) for running training on single GPU
[Full Changelog](https://github.com/convml/convml_tt/compare/v0.6.0...v0.7.0)
**Implemented enhancements:**
- Simplifying software dependencies [\14](https://github.com/convml/convml_tt/issues/14)
- Fixes and notes for training on JASMIN [\20](https://github.com/convml/convml_tt/pull/20) ([leifdenby](https://github.com/leifdenby))
- Switch from fastai to pytorch-lightning [\17](https://github.com/convml/convml_tt/pull/17) ([leifdenby](https://github.com/leifdenby))
[v0.6.0](https://github.com/convml/convml_tt/tree/v0.6.0) (2021-02-24)
Last version using fastai v1. Contains model as used in L Denby 2020.
[Full Changelog](https://github.com/convml/convml_tt/compare/v0.5.0...v0.6.0)
**Merged pull requests:**
- Setup CI for windows, macos and linux [\18](https://github.com/convml/convml_tt/pull/18) ([leifdenby](https://github.com/leifdenby))
- Simplifying installation of convml\_tt [\15](https://github.com/convml/convml_tt/pull/15) ([leifdenby](https://github.com/leifdenby))
- Tile generation and analysis pipeline [\12](https://github.com/convml/convml_tt/pull/12) ([leifdenby](https://github.com/leifdenby))
- add test for creating embedding from tile list [\11](https://github.com/convml/convml_tt/pull/11) ([leifdenby](https://github.com/leifdenby))