CodeCov would be reduced by current solution for issue #153

[alex-senger]

In Issue #153 a few lines of code were commented out to keep CodeCov at 100%. These lines contain a try-catch block to catch a possible LearningError and PredictionError for errors that we already cover. It may be possible that we didn't cover every Error and these would still be thrown. We cannot think of a case and can't write tests for these as a consequence.

Either we remove these lines completely or we take into account that CodeCov is not at 100% anymore.

Link to the PR: #355

[zzril]

We could implement dummy models that just always raise the required exception.
E.g. a ValueErrorOnFitClassifier that will always raise a ValueError when calling fit (or predict) on it, then a ValueErrorOnPredictClassifier that only raises one when calling predict etc.

Then use one of those dummy models whenever we can't cover some line using the normal ones.

[Marsmaennchen221]

We could implement dummy models that just always raise the required exception. E.g. a ValueErrorOnFitClassifier that will always raise a ValueError when calling fit (or predict) on it, then a ValueErrorOnPredictClassifier that only raises one when calling predict etc.

Then use one of those dummy models whenever we can't cover some line using the normal ones.

This solution will be implemented in #355 so this issue is irrelevant