Open PGijsbers opened 5 years ago
In particular coverage from unit tests need to expand. With e.g. mocking ensembling and search algorithms can also be included in quick tests. Missing coverage:
score
~cross_val_predict_score
It would also be desirable to create smaller test modules (e.g. separate tests for arff
input or str
-labels)
Initial effort was merged in #50. I think the gama/genetic_programming/compilers/scikitlearn.py
code is reported as not covered due to it being executed in a separate process only (in system tests).
More progress in #84. Coverage of subprocesses is definitely correctly reported. Despite reaching 90% coverage, how to properly test AutoML systems reliably and quickly is still not obvious to me.
How to adequately test AutoML systems remains an open issue, especially how to do this well with unit tests as opposed to simply running a (small) benchmark. Updating the title to reflect that more general code coverage is (currently) not a goal, but rather to test for more potential issues in the unit test suite.
Not all code is currently covered by unit and/or system tests. In some cases, this does not matter (e.g. not all ValueError scenarios need to be automatically checked, I think), but for other functionality still needs coverage (e.g. time-out behavior in evaluation.py).