Start a python package - Githubissues

Avsecz commented 6 years ago

Avsecz commented 6 years ago

Worklog

started a python package modelzoo
- setup the CLI
setup simple documentation with a desired end-user API: docs
started tests in tests with py.test. For now it only runs the two examples using a CLI tests/test_examples.py
setup travisCI for automatic unit-tests on python 3.5 and 2.7
- had some issues with installing the genomelake repository as it's not public. Solved by adding a deploy-key to my github fork of it
- Build for python 3.5 works: https://travis-ci.com/kipoi/model-zoo/jobs/88646792
  - if fails on python 2.7, as the rbp example is written for python 3.5. Need to wrap the preprocessor/model call into a virtualenv with python 3.5 or have a flag to skip the test.
Defined Dataset and Model abstract classes
- Dataset (or Extractor) is now the same as pytorch.utils.data.Dataset
- i.e. it defines: __len__ and __getitem__
- Model defines predict_on_batch and can be defined using custom code or deserialized from keras, sklearn
renamed preprocessor -> extractor
Structured modules into model, data, pipeline

[x] python 2.7 unit-test should go through
- [/ ] solve setting up conda environments at https://github.com/kipoi/model-zoo/blob/pypackage/modelzoo/model_handling.py#L158
  - will create a separate issue for this

Avsecz commented 6 years ago

python package
CLI
- $ modelzoo predict ...
- $ modelzoo test ...
- $ modelzoo preproc ...
python SDK
- modelzoo.load_extractor
- modelzoo.load_model
unit-tests with py.test running on travis-CI
Defined Dataset and Model abstract classes
- Dataset (or Extractor) is now the same as pytorch.utils.data.Dataset i.e. it defines: __len__ and __getitem__
- Model defines predict_on_batch and can be defined using custom code or deserialized from keras, sklearn
Preprocessor & model schema
- See the updated .yaml files in examples/
started documentation in docs/