Kipoi

Features
- New API for contributing groups of models: https://github.com/kipoi/kipoi/pull/349
- Updated PyTorchModel to new api: https://github.com/kipoi/kipoi/pull/348
- Unify the python object loading API: https://github.com/kipoi/kipoi/pull/346

Bugfixes
- default using local path 347
- Fix occasional test failing due to parallel workers 345
- Fix `kipoi env create` with kipoiseq dataloaders https://github.com/kipoi/kipoi/pull/344
- fix parameter overriding https://github.com/kipoi/kipoi/pull/338

Bugfix

- parameters specified in model.yaml `default_args` now override the ones specified in the yaml docstring

Features
- speedup HDF5BatchWriter (327)
- Files may now be referred to by an url (https://github.com/kipoi/kipoi/pull/334)
- `kipoi_dataloader` decorator which allows to write new dataloaders by parsing the doc string as the dataloader.yaml file (https://github.com/kipoi/kipoi/pull/336)
- example: https://github.com/kipoi/kipoi/blob/master/example/models/kipoi_dataloader_decorator/dataloader.py
- Refer to the dataloader by a python import string in `default_dataloader` and allow overriding the default variables
- example: https://github.com/kipoi/kipoi/blob/master/example/models/kipoi_dataloader_decorator/model.yaml#L8

Bugfixes

- Retain author order in model groups, closes 318 (319)
- hdf5 writers dump method

Features and re-factoring

- variant effect prediction and interpretation modules are now part of `kipoi_veff` and `kipoi_interpret` plugins
- Instead of `kipoi postproc` you should use `kipoi <plugin>`. E.g. `kipoi veff score_variants`
- rename `kipoi.remote` to `kipoi.sources`
- rename `kipoi.components` to `kipoi.specs`
- remove `-i` flags in commands like `kipoi predict`
- `kipoi.postprocessing` doesn't exist anymore. The functionality can be found in `kipoi_veff` and `kipoi_interpret` packages

Fixes
- prevent values to be None in GenomicRanges

Bug fixes

- minor fix in gradient visualisation
- Freeze urllib3 dependency so that requests package works

Refactoring

- rename preds to grads in the gradient visualization
- rename `filter_ind` to `filter_idx`

Features
- kipoi env create: add option where to write the tmpfile
- Enable models to be installed `kipoi env ...` or pulled `kipoi pull` by referring to a wildcard
- Add the `predict_method` option to the SklearnModel class

Bugfix

- fix the score variants res https://github.com/kipoi/kipoi/pull/276
- fix: train_iter now leverages iterables https://github.com/kipoi/kipoi/pull/277