This repository comes with a pre-defined python package structure.
To use it for setting up a new repository, just click the Use this template button on the project page.
Once a repository is created from this template, it will setup itself with a github action, deriving the actual package name from the repository name.
Furthermore it comes with pre-defined docker images for deep learning (derived from either a base conda one or the NVIDIA NGC PyTorch one). The docker images are automatically built on each PR and can be automatically uploaded to dockerhub via github actions after each commit to main. For this to work, the user needs to set the secrets DOCKERHUB_USERNAME and DOCKERHUB_TOKEN to the associated username and token respectively.
Whenever a release is created on github, there is a github action running to automatically test if the package can be built and installed and tries to automatically publish it to PyPi and PyPi-Test. The PyPi release is only triggered, if the PyPi-Test release succeeds. For this to work the user has to set the secrets TEST_PYPI_API_TOKEN and PYPI_API_TOKEN. These tokens can either be obtained on a general per-user basis or per project after publishing manually for the first time.
Tests lying under the tests directory, are automatically run after each commit. The dependencies will thereby be cached on a weekly basis to speedup the development progress. In order to get proper code coverage details, users may want to alter the ci.yaml file in the workflows directory to include --source=YOUR_PACKAGE_NAME to the coverage run command in the Tests step. Furthermore users can specify the CODECOV_TOKEN secret to upload their coverage results to codecov.
This repository also contains an opinionated pre-commit config. For more information, on how pre-commit works and how to use it, visit their website. Pre-Commit is also run on each commit to the main branch and the results will be commited automatically.