Kipoi

Main highlights of this release are:

- Several bugs are fixed
- Support for python 3.7 has been added.
- Now users can specify the dataloader as a python function available in the current environment using --dataloader cli argument. Unlike before this function does not have to be specified by yaml representation of the dataloader or defined in kipoiseq with kipoi decorator
- CI now is parallelized allowing the nightly ci of the model repo to complete withhin ~50 minutes.
- Each kipoi model now can specify hdf5 chunk size for hdf5writer in their yaml representation. An example is [here](https://github.com/kipoi/models/blob/add-test-predictions/Basenji/model.yaml#L83-L84)

- Bug fix - 585
If you would like to output metadata along with the predictions add --keep_metadata or -m to kipoi CLI. By default, outputting metadata is False now.

In this release we have addressed the following
- One can  output metadata (such as gene_id, transcript_id, exon regions etc.) along with the predictions from kipoi now. The relevant CLI flags are -m or --keep_metadata. By default it is true.
- Docker containers for all the models are released. These containers create the model specific conda environments, activate them and return a bash shell with the activated environment.  Each model group page now has a tab named docker which gives information about which docker container is compatible with which model and how to use the containers with user defined datasets. For example, look [here](http://kipoi.org/models/DeepSEA/variantEffects/).
- We are hosting 13 containers for existing 29 model groups [here](https://hub.docker.com/repository/docker/haimasree/kipoi-docker). The repository responsible for building the containers and testing them is [here](https://github.com/haimasree/kipoi-containers). You will also find information about how to build singularity containers from these docker containers in the [README](https://github.com/haimasree/kipoi-containers).
- Continuous integration tests are fixed across this repository, [kipoi models](https://github.com/kipoi/models) and [kipoi website](https://github.com/kipoi/website).

Fixes
--------
- fix BigWigWriter (431)
- deprecating install_model_requirements and install_dataloader_requirements (429)
- use non-deprecated converter keyworkd in attr.attrib (420)

Features
- added ZarrWriter / Reader (418)
- added AsyncBatchwriter (415)

Fixes
- https://github.com/kipoi/kipoi/commit/eb65532e0e45527daceb0ce916cb1f656180e7ec
- Fix the kipoi test-source command
- fix encoding in related mixin (412)
- revert back pytorch model (411)

Bugfix

- https://github.com/kipoi/kipoi/pull/409 kipoi info now doesn't download the model/example files and doesn't require the framework (say pytorch) to be installed.