Delira

This is a major release since it changes the license to a dual licensing strategy. This may cause some troubles if you want to use the latest version, so make sure to check the [README](https://github.com/delira-dev/delira/blob/v0.6.0/README.md) and the [LICENSE](https://github.com/delira-dev/delira/blob/v0.6.0/LICENSE).

Bugfixes

- removed many mutable default arguments [[Commit]](d870195cae1f8b9a1f35b51e1e9d9829e411efe3)
- Correctly unload TF in Tests [[Commit]](6b9d4c432119f1cbfd79e9cfa3aac94af71d7e09)
- Split TF-Tests in Graph and Eager Backend [[Commit]](7bd3c37a11906a0a296d935ee7ef26ed4e1f65bd)
- Backend-Checks for Tests (avoid duplicate testing) [[Commit]](70f01bfd8aec754e64031fe5a690f0af601d3caf)
- Move to new APEX API [[Commit]](183b97d8f20226bbe21599522ed8b5d45e7ab41d)
- Fix Graph-Based Logging [[Commit]](c754caa337632574ab94e00fe09e91f2916e2f11)
- Fix Utils Imports [[Commit]](b84c229cd37be12d60a26623602c21e4eb9e60fc)
- Move to new Slack API [[Commit]](93498accf96b390a24bb5c386874cab6f7cb6fbb)
- Fix Travis CI/CD [[Commit]](488bc882fbf642cec375fc203e8aab667b97112b)
- Fix Config overwrites [[Commit]](024a4028856661ac8328443ef3cf3d456a30991c)
- Minor type fixes inside the trainer [Commit]](e7294fd5580c10e2d2acc3f92145d1e814e2566d)
- Fix Typecheck for Transforms [[Commit]](47d2a14aacd94da99b611904bc5c36c1fd370367)
- Config Traversal [[Commit]](176acb75628198994697f74c4b6ca340de612716)
- Fix logging frequency and timeout [[Commit]](64ce57ba0839b775c4afb9d367ff29a8f0a17448)
- Fix Logging Steps [[Commit]](e33bdb93135007748f0f2e155bf5d080ea4b9769)

Breaking Changes

- Logging is now done outside the `closure`
- Metrics are now based on numpy and are calculated the same way for `train` and `val`
- Remove `trixi` dependency [[Commit]](480170929f6c5fc68f29eeb14e10bf0a2263acc1)
- Remove deprecated codeparts (`TorchvisionClassificationDataset` , `train_test_split` , `Nii3DLazyDataset` , `Nii3DCacheDataset` , `[delira.utils.imageops.py](http://delira.utils.imageops.py)` , `DefaultOptimWrapperTorch` , `DebugMode` Context Manager and the following requirements: `scikit-image`, `pillow`, `SimpleITK`, `psutil` [[Commit]](22f81ddcd9e65fbe8660f97aa47fd8034fd81722)

New Features

- Slack Notificator [[Commit]](069fef5eddfb1397b45d43e6e12cd40c2edea5c2)
- Function to seed all backends [[Commit]](f79bef8250294cf4fb5c18945150e2dad9223d7e)
- Add Callbacks after and before epoch, after and before training and after and before every step (regardless of train and val) [[Commit]](b491b62a9fc06af4b7be9b87f8c91f1167fd08d8)
- New Custom Data Pipeline (completely replaces the batchgenerators pipeline) for more customization and better integration of our sampling [[Commit]](8f2bd610f27c1e07122fd93c021af03db6609f19)
- Register Logger in Callbacks [[Commit]](df0a7e4bf05e7fb9d730ffe9ab1e93f91b6a5919)
- Onecycle Scheduler Callback [[Commit]](bd6f4caf6f08af2854641a30138f32d1c494be3a)
- Allow creation of config from system args [[Commit]](be0f9f0773b4cb19c6af97ccdb859e26d5e6e5dc)

Bugfixes
* Several improvements to reduce memory usage (and fixed an actual memory leak)
* Fix the Search for previous trainings [[commit]](b6396a859b2a6076181ceceaec32c40c9d7c8913)
* Convert Metric Values to `Numpy` [[commit]](0850487927b1161b9626a4b8ce1e542cbcf61d5c)
* Epochs correctly saved in checkpoints [[commit]](db58949cebcc8596e5eef1451ae4ca8cec6a4165)
* Refactoring to clean up global namespace [[commit]](75863902c3491c93ac21c5f8533c4b79f4ae22b3)

Improvements
* Add `numba` jitting for transforms [[commit]](299b7ac879942fe26a5935addb218e2d919d7e1d)
* Bump `TensorFlow` to 1.14 [[commit]](784cba6dce73ed2b31db8fdce7f93b1d19f6dbc2)
* Versioning via `versioneer` for more detailed versions [[commit]](9491fadf6b17308b9080550e6c44b612092cc5b8)

New Features
* **Slack**: We now have a community chat on [slack](https://join.slack.com/t/deliradev/shared_invite/enQtNjI1MjA4MjQzMzQ2LTUzNTQ0MjQyNjJjNzgyODczY2Y1YjYxNjA3ZmQ0MGFhODhkYzQ4M2RjMGM1YWM3YWU5MDM0ZjdiNTQ4MmQ0ZDk) [[commit](e3307db4d450f9c37d45d9c838819021418e3b1a)]
* **Paper**: We wrote and submitted a Paper to [JOSS](https://joss.theoj.org/) [[commit](9af1915df077fdc52901b6a1696c7f6d0efd77a0)]
* **Dataset Integration Guide**: New tutorial to switch from old dataset API to new dataset API [[commit](42b55c388e522b75451e86055ae943acfdafcc7a)]
* **Predictor**: Introduce new `Predictor` class to predict from network without having to initialize Trainer with dummy values [[commit](69080a87fa2255d6e8da3564bd11b157e7e00458)]
* **PEP-8**: Complete Codebase is now PEP-8 compatible and CI/CD now checks for it [[commit](5db9a6533129f4aebb60dc8a0587a7616313567c)]
* **Debug Mode**: New `debug mode` disables all multiprocessing (currently multiprocessing is only used during data loading); Might slow down the program but increases debug capabilities [[commit](b5713ddbe85bb356a4c7f97d5587b3c69750d063)]

Breaking Changes
* **Return dicts**: Models must now return `dict`s during prediction and training [[commit](6f023a8365dd2007a3d22862b7b4fb95652d77b5)]
* **Kfold**: The old `kfold`- function has now merged with the new `kfold`, `stratified_kfold` and `stratified_kfold_predict` to a new `kfold` function supporting all these behaviors. [[commit](69080a87fa2255d6e8da3564bd11b157e7e00458)]
* **Rename tf Backend**: The `tf` backend has been renamed to `tensorflow` during installation. The new command to install `delira` with TensorFlow backend is now: `pip install delira[tensorflow]` [[commit](74e0d6dd19d805d6ab84d719cd49ec322c3f9f36)]

Non-Breaking Changes
* **Unify Trainer**: Merging Code from Backend Trainers (`TfNetworkTrainer` and `PyTorchNetworkTrainer`) into `BaseNetworkTrainer`; Will increase maintainability and ease of adding new backends [[commit](69080a87fa2255d6e8da3564bd11b157e7e00458)]
* **Unify Experiment**: Merging Code from Backend Experiments to `BaseExperiment` [[commit](69080a87fa2255d6e8da3564bd11b157e7e00458)]
* **Sampler StopIteration**: Refactor samplers to use a common function to check if `StopIteration` must be raise (where possible) [[commit](9d25a68516757cd39482939cd599bb267727a4b6)]

Bugfixes
* **TF Resnet18**: ResNet18 in TF now matches the official paper [[commit](9a133c0bdfe148f861f013b756be70b76d480c4a)]
* **Parameters**: Correct Permutations in `Parameters` class [[commit](60fa85d7dcc07928c03b1db3ffdec5ab6428d0b8)]
* **Infinite Recursion**: Infinite recursion in `LookUpConfig` due to wrong typechecks is now fixed [[commit](ecc1286507942ecc6625c06287cc6e07c8acf0ce)]
* **DataLoading `num_batches`**: The number of batches is now calculated without any influence of `multiprocessing` [[commit](9d25a68516757cd39482939cd599bb267727a4b6)]
* **DataLoading**: Instabilities due to inter-process communication [[commit](71368ce3ed1414fe6054be76b47ce9efdf0cdf9a)]

Added
* `WeightedPrevalenceSampler` (433cccb99fdfdf51e319faab30a2d851644d93bc)
* `_DatasetIterator` allowing iteration over a dataset with a simple loop without having to iterate over indices explicitly (70d62327de55447b6753c751d980ba66bbee153f)
* Support for TF1.13 (5e2817297e58fb165a84668903e7a45ec2bf3999)
* Support for Python 3.5 (5db1a1776c97e6ec8483a6a640e9bc95fd4980a2)

Breaking Changes
* New Dataset API (88 )
* Changed test framework from `pytest` to `unittest` (a70980944f361b6d7c290b6be077c844db79782f)

Bugfixes
* Random sampler can now be seeded via numpy for completely determinitic behavior (a7c10ee5df6fe8714755aba5be1ea367413f608a)

Bugfixes:
* loading behavior with `weights_only=True` for torch backend (fff41cf775a10677c5c0c76e6c8d595ba82671cb)
* Behavior of `BlankDataset` (353648effdbfdaf02d81662d726d14a89de57caf)
* Refactoring of backend choosing (01f25deb0c524458d3bf02a8eefc9d9838c63196)