Closed rpetit3 closed 9 months ago
The release process for medaka involves QA testing steps to determine that the code can be used for both training and inference. The unit tests run by make test
do not assert these properties; the furthest they go with respect to tensorflow is to check the model data can be loaded.
We test also on a matrix of linux and macOS devices with x86 and aarch64.
By the time code has makes it to GitHub medaka is considered fully tested. A GitHub action performing a subset of the testing with mutated requirements is not sufficient to ensure the mutated requirements would meet all of our internal testing procedure.
This PR adds a Github Action to test Medaka for multiple versions of Python and Tensorflow. It was written to mimic the setup that is used on GitLab.
After setup, it runs the tests available from
make test
, which assumes they are sufficient to test whether Medaka is properly functioning with the different versions of Python and TensorflowCurrently it tests:
Here is an example execution of this action: https://github.com/rpetit3/medaka/actions/runs/6398866670/job/17369898556