For a new machine learning course we're looking to design a series of coding assignments in which students get some starter code, and make improvements until the unit tests pass. Then they commit and push their code back to the remote where an autograding workflow runs more tests to see if they did adequate work.
What we'd like to do would be to give the students some tests that they can look into, to see what the general programming goal is; but to also have a secret unit test to try their code on data that the students have never seen. On this unseen test data they'd have to reach at least a certain accuracy score to get a passing grade.
The question is: can this be done in github classroom? It seems that the default setup is to give all the tests openly in the starter code repository. But we want to have some tests that the students can't see, so that we can test if they're only narrowly writing to the visible test or actually writing a properly generic solution.
If this isn't directly possible, is there a workaround strategy?