Carrot2

I think the long-term strategy for testing the C# APIs should be based on two things:

1. running the JUnit tests of the algorithms, just as we do now for Lingo3G,
2. running a quick correctness check: clustering a few data sets with Java and C# versions and comparing the results (I'm not sure though if the results can be identical, e.g. due to some tiny differences in numerical computations).

I will add JUnit tests of open-source algorithms, no problem with this. As for cross-check between Java and C#, it's a good idea and it was once implemented, but remember about numerical instability in Java (i.e. compiled native code can yield different results than pre-JITted code, for instance). The way we did it was run in interpreted mode, but this was very, very slow.

It's impossible to separate test classes from actual code they're mixed. A split of test/classes output folders is required (like it used to be, I don't remember why they haven been merged).